WEBVTT
00:00:00.000 --> 00:00:05.790
We reached a position, we reached a place where there were 3000 experiments running in a particular sandbox at a particular time.
00:00:06.089 --> 00:00:17.710
We have built generative AI solutions that are cutting edge, but eventually we have seen no adoption and if we do not have any adoption, eventually it becomes, yet another experimentation.
00:00:18.234 --> 00:00:26.094
So I've seen people thinking of generative AI as a magic wand where it can fix things where people do not know what is broken.
00:00:26.394 --> 00:00:27.385
Generative ai.
00:00:28.344 --> 00:00:38.054
Is the first technology in my mind, in human history where it has been personally adopted first, then, being adopted in a corporate world, right?
00:00:52.097 --> 00:00:59.777
Almost every organization today is experimenting with generative ai, but very few are seeing meaningful impact at scale.
00:01:00.436 --> 00:01:07.522
The hardest part isn't the technology, it's the work behind the scenes, data foundations, operating models, and how decisions get made.
00:01:08.486 --> 00:01:18.867
In today's episode of Data Analytics Chat, I'm joined by Naan Paul, who's the managing Director for Chief Generative AI Center of Excellence at Accenture.
00:01:19.227 --> 00:01:24.266
And today we'll discuss what it really takes to adopt generative AI inside an organization.
00:01:24.867 --> 00:01:32.996
What separates experimentation from impact where companies get stuck, and what leaders should prioritize over the next six to 12 months.
00:01:33.537 --> 00:01:34.887
Naan great to have you here.
00:01:35.486 --> 00:01:36.596
Thank you so much, Ben.
00:01:36.867 --> 00:01:37.137
Thank you.
00:01:37.167 --> 00:01:39.447
Thank for hosting me and I'm really excited to be here.
00:01:39.807 --> 00:01:40.167
Thank you.
00:01:40.736 --> 00:01:46.557
So I guess, Anne, before we dive in, do you wanna give listeners a quick introduction to who you are and the work you are leading today?
00:01:47.156 --> 00:01:47.966
Absolutely.
00:01:48.337 --> 00:01:49.566
My name is Naan Paul.
00:01:49.566 --> 00:01:55.746
I'm based out of us, and for me, generative AI has been a very organic growth.
00:01:56.496 --> 00:01:58.236
Five years ago, my organization
00:01:58.537 --> 00:01:58.596
to
00:01:59.346 --> 00:02:01.087
opened up the Center of Excellence.
00:02:01.147 --> 00:02:06.216
It's a global entity, and as a part of this COE, as a part of the global organization.
00:02:06.816 --> 00:02:11.347
I supposed to understand help customers build
00:02:12.037 --> 00:02:12.456
when,
00:02:12.997 --> 00:02:13.717
generative ai.
00:02:14.317 --> 00:02:20.312
So for me, I have three major three ways that I contribute as a part of.
00:02:20.912 --> 00:02:23.106
One is being part of the consulting work.
00:02:23.706 --> 00:02:25.206
I work with my customers.
00:02:25.266 --> 00:02:27.247
I try to understand their North Star vision.
00:02:27.306 --> 00:02:28.611
I try to understand their gaps.
00:02:29.211 --> 00:02:36.981
I try to understand what it'll take for them, like you said, to move from an experimentation mindset to a productionized mindset.
00:02:37.581 --> 00:02:41.451
Two, I work with my products teams.
00:02:41.542 --> 00:02:54.502
I work with my partners to understand what it means to build solutions, jumpstart kits, frameworks, the entire scaffolding bit, because when we are talking about enterprise adoption, when we are talking about scalability.
00:02:55.192 --> 00:02:59.331
It is very important to do it right and do it in a repeatable way.
00:02:59.662 --> 00:03:02.481
We talk about building the right foundation.
00:03:02.542 --> 00:03:08.872
We talk about how generative AI needs to be designed in a way that we are not just chasing use cases and demos.
00:03:09.141 --> 00:03:10.581
We are chasing outcomes.
00:03:10.972 --> 00:03:13.942
So what it means to build those foundational capabilities.
00:03:14.331 --> 00:03:18.141
And three, because of how we are working very closely with our partners.
00:03:19.132 --> 00:03:23.301
It's about aligning our roadmap to the roadmap of our partners.
00:03:23.602 --> 00:03:27.891
What are, for example, Microsoft offering AWS offering, GCP offering.
00:03:28.431 --> 00:03:42.106
Now, what is Nvidia offering and how as Accenture, we are incorporating those understandings, incorporating those newness into our storyline, into our adoption, and how can we bring the latest thinking to our customers?
00:03:42.646 --> 00:03:43.787
So this is my day-to-day.
00:03:43.817 --> 00:03:46.967
This is how you know we have approached generative ai.
00:03:47.567 --> 00:03:55.192
From, like you said, change management to having ownership to ensuring that we are about the bigger picture and.
00:03:55.792 --> 00:03:56.092
Brilliant.
00:03:56.092 --> 00:04:02.657
Just so the listeners know, obviously, so are you specialized in one industry Accenture, or is it across the board?
00:04:03.257 --> 00:04:05.356
So within Accenture, we have multiple groups.
00:04:05.687 --> 00:04:09.747
However, my group as I was talking about, is part of Center of Advanced ai.
00:04:09.836 --> 00:04:12.896
So Center of Advanced AI is a very horizontal entity.
00:04:13.706 --> 00:04:21.836
We are supposed to be the tip of the sphere where we are supposed to understand, incorporate, and in vibe what is the latest and greatest technology offering.
00:04:22.242 --> 00:04:26.831
So my team and I, we specialize mostly across domains, across verticals.
00:04:27.132 --> 00:04:29.291
We work with our functional leads.
00:04:29.351 --> 00:04:34.572
We work with our domain experts, the subject matter experts, but we bring in the technology bits.
00:04:34.752 --> 00:04:41.531
We understand what is the rate latest offering, what it means to scale, what it means to secure.
00:04:41.591 --> 00:04:44.891
We talk about responsible AI security safety breaks.
00:04:45.252 --> 00:04:47.651
So we will bring in mostly the horizontal technology.
00:04:48.252 --> 00:04:58.942
We talk and adoption and others that we work with, like our business partners, our SMEs, our subject domain experts, they will bring in the functional knowledge.
00:04:58.971 --> 00:05:00.802
So together we will solve problems for.
00:05:01.401 --> 00:05:01.822
Brilliant.
00:05:01.851 --> 00:05:05.471
So you'll have a good blend of use cases, which would be great for the listeners today.
00:05:06.312 --> 00:05:16.932
And I guess something before we move on to the data topic, has there been a experience that you've gone through that's most shaped how you approach leading AI initiatives in organizations today?
00:05:17.531 --> 00:05:18.172
A hundred percent.
00:05:18.276 --> 00:05:20.346
Like it has been a growth cycle.
00:05:20.406 --> 00:05:21.786
It has been a lot of lessons learned for us.
00:05:22.387 --> 00:05:29.197
I remember five years ago when we started off this journey, our intention was how do we ensure that we are educating people?
00:05:30.007 --> 00:05:32.076
So we created our own sandboxes.
00:05:32.586 --> 00:05:41.997
We actually advertised, we actually, put in a lot of money where we asked our account teams, like we, we have multiple account teams.
00:05:41.997 --> 00:05:46.257
We asked, sorry we asked our account teams to onboard.
00:05:46.812 --> 00:05:48.432
Into these sandboxes.
00:05:48.701 --> 00:05:52.721
We provided, tenancy, we provided Azure services.
00:05:53.081 --> 00:05:56.721
We were early on bringing in, LLM models.
00:05:56.781 --> 00:06:02.791
We were exposing those LLM models and we are requesting all these account teams to go, and experiment.
00:06:03.692 --> 00:06:09.901
So the initial days, it was all about getting familiar, ensuring that people are not looking away.
00:06:10.651 --> 00:06:23.192
Giving them a safe space, giving them all the right tools and technologies, giving them the right support to ensure that people can bring in their thought process, they can bring in their ideas and just experiment, right?
00:06:23.461 --> 00:06:32.851
So we focused on ensuring that, just keep on experimenting, whatever thoughts you have, irrespective of whether these outcomes are going to add value or not.
00:06:33.041 --> 00:06:33.762
Just experiment.
00:06:34.617 --> 00:06:40.406
We reached a position, we reached a place where there were 3000 experiments running in a particular sandbox at a particular time.
00:06:41.007 --> 00:06:51.396
So what it did was we ensured that different account teams who are working with their respective clients, they were able to tell a story, they were able to showcase something.
00:06:51.817 --> 00:06:53.646
And again, showing is believing.
00:06:54.442 --> 00:06:58.567
And for the first one and a half years, we are doing a lot of this experimentation.
00:06:58.927 --> 00:07:02.166
We were not productionizing anything, but that was okay with us.
00:07:02.976 --> 00:07:06.666
Again, as time went by, we realized that we have to shift the gear.
00:07:06.997 --> 00:07:11.526
We have to understand what it means to just stop experimenting and start focusing on value.
00:07:11.627 --> 00:07:20.387
So for the next one, one and a half years we actually promoted, we talked about how we have to change our mindset from being, just experimenting to.
00:07:20.822 --> 00:07:31.531
Identifying, bubbling up the right use cases, focusing on what is important, and use those lessons learned from experimentation to start providing real value outcomes.
00:07:31.862 --> 00:07:35.791
And when we started building those real value outcomes, we saw the real challenges.
00:07:35.791 --> 00:07:37.291
We talked about data readiness.
00:07:37.291 --> 00:07:39.391
We talked about the operating model.
00:07:39.391 --> 00:07:42.182
You talked about the scalability, adoption.
00:07:42.182 --> 00:07:46.396
You talked about the real ROIs, and we've stumbled upon those challenges too.
00:07:47.161 --> 00:07:59.312
As we stumbled upon those challenges, we tried to understand whether it is a technology problem, whether it is an adoption problem, whether it is a responsible problem in terms of ownership, and we started mitigating a lot of those.
00:07:59.672 --> 00:08:00.872
So over a period of time.
00:08:00.872 --> 00:08:03.922
Now, if we look back, we did the right approach in my opinion.
00:08:03.971 --> 00:08:09.041
We didn't just go ahead and start saying, Hey, we are not going to start, or we are not going to enable anybody.
00:08:09.357 --> 00:08:12.206
To use generative AI unless you really provide a value.
00:08:12.266 --> 00:08:16.466
No, we said, don't worry about the first initial phase, just experiment.
00:08:16.586 --> 00:08:17.937
Just get yourself comfortable.
00:08:18.266 --> 00:08:20.487
Just ensure that you are technology aligned.
00:08:20.906 --> 00:08:23.716
And then over a period of time, we learned our lessons.
00:08:23.987 --> 00:08:29.896
We brought in business, we got the right buy-ins, and again, we organically grown into.
00:08:30.257 --> 00:08:32.206
What and how we can deliver values.
00:08:32.506 --> 00:08:36.197
So if you look back into those five years, it has been a very incremental journey for us.
00:08:36.197 --> 00:08:55.386
And now when we talk about business process automation, when we talk about augmentation, when we talk about, identifying those key functional capabilities within an organization and look into those from the lens of automation and augmentation, we understand what it takes, because of how the journey has been propagated for us.
00:08:55.787 --> 00:08:57.767
Long story, Ben, again, to your point.
00:08:58.667 --> 00:08:59.027
Yes.
00:08:59.027 --> 00:09:04.937
Now the important bits are ensuring that we are not just stuck into that POC world.
00:09:04.996 --> 00:09:07.096
We are not just looking into experimentation.
00:09:07.126 --> 00:09:09.937
We are not stuck, getting stuck into that mindset.
00:09:10.297 --> 00:09:16.927
But the journey had to start somewhere, and the journey for us started five years ago, and now it's all about value.
00:09:17.047 --> 00:09:22.606
It's all about looking into, as a company and saying, what is, what are those important things?
00:09:23.162 --> 00:09:27.581
Which are making us lose time, which is making us lose, money or growth opportunities.
00:09:27.581 --> 00:09:36.562
So now we are looking into generative AI opportunities from the lens of saving time, making money or growing as an organization.
00:09:36.772 --> 00:09:41.871
So again, to reach that particular milestone, it was a lot of hard work.
00:09:41.881 --> 00:09:48.287
It was a lot of technology adoption from my team and it was a lot of business alignment that we worked together with.
00:09:48.886 --> 00:09:56.017
Yeah, and I love the fact where you say, now it's about creating value because look, we're going for a transformational period now, isn't it?
00:09:56.017 --> 00:10:02.736
With businesses, and if you've got every business department coming to with you with a use case, you can't do it all, can you?
00:10:02.736 --> 00:10:08.706
You need to be choosing the one that's going to make the biggest ROI or the biggest impact for your business right now.
00:10:08.706 --> 00:10:09.667
You can't be doing everything.
00:10:10.267 --> 00:10:11.101
A hundred percent.
00:10:11.341 --> 00:10:12.812
And I think that's a mind shift.
00:10:12.871 --> 00:10:14.267
Shift change to your point.
00:10:14.866 --> 00:10:25.657
We couldn't have just talked about value from day one to your point, but now when we talk about how we are approaching generative ai, it's all about, who is owning it.
00:10:26.197 --> 00:10:27.817
It cannot own generative ai.
00:10:27.856 --> 00:10:37.037
If you give it the responsibility of owning generative ai, they'll just focus on architecture tools and what is the right technology stack.
00:10:37.831 --> 00:10:39.841
That is not the right measure of generative ai.
00:10:39.841 --> 00:10:43.532
Again, you mentioned it, and again, I think everybody understands it now.
00:10:44.131 --> 00:10:47.792
The real value of generative AI now is how is it transforming business?
00:10:47.851 --> 00:10:49.292
What is that underlying ROI?
00:10:49.892 --> 00:11:03.182
We are not just talking about automation because yes, automation is a good measure, but in my opinion, it's all about augmentation first, which means can we identify those important low risk, high impact use cases?
00:11:03.902 --> 00:11:20.341
Can we find out those business scenarios where the augmentation or automation can work hand in hand and allow businesses to own solution that can reduce, for example, the effort, the time, or even FTEs, right?
00:11:20.942 --> 00:11:32.131
And when we start looking into generative BI use cases from the lens of what can organizations do, how the decision making can be made faster, how the decision making can be de-risked.
00:11:32.731 --> 00:11:38.101
Those are the right measures to pick up generative use cases now, and that is how we are approaching it now.
00:11:38.131 --> 00:11:41.642
So again, the experimentation phase was important.
00:11:41.761 --> 00:11:55.677
Maybe it was important three years ago, four years ago, but now with all the technology changes with the way in which we are actually approaching generative ai, the way in which we should be approaching generative AI is business adoption and the ownership.
00:11:56.326 --> 00:12:01.216
The ownership at any point of time at this day and age is all about business.
00:12:01.317 --> 00:12:02.937
How are they looking into business?
00:12:02.937 --> 00:12:09.397
How are they looking into augmentation on automation and what business processes they feel comfortable, augmenting with generative ai.
00:12:09.846 --> 00:12:11.981
So I think those are the right questions we need to ask right now.
00:12:12.876 --> 00:12:13.386
Brilliant.
00:12:13.657 --> 00:12:17.307
So I guess most leaders feel pressure to do something with generative ai.
00:12:17.996 --> 00:12:20.621
In your experience, what separates organizations that.
00:12:21.221 --> 00:12:23.052
From those that create value.
00:12:23.652 --> 00:12:25.812
It's a multidimensional question to be honest.
00:12:25.812 --> 00:12:27.761
And you are, you're so right to point it out.
00:12:28.302 --> 00:12:32.172
So I have seen organizations where there is a lot of IT dependency.
00:12:32.772 --> 00:12:39.522
A lot of focus has gone into the CTO office where the intention is, do you have the right infrastructure?
00:12:39.971 --> 00:12:41.831
Do we have the right technology stack?
00:12:42.131 --> 00:12:42.881
What is that right?
00:12:42.881 --> 00:12:43.662
Technology stack.
00:12:44.501 --> 00:12:51.761
Can we set up the right technology foundation and can we onboard business and allow them to come and experiment?
00:12:52.152 --> 00:12:59.922
And that is a methodology where a lot of companies have gone into, and for the last two, two and a half years, they have been stuck into this POC loop, right?
00:13:00.522 --> 00:13:02.351
Technology is driving Gen ai.
00:13:02.351 --> 00:13:04.392
There is no adoption.
00:13:04.721 --> 00:13:06.552
Business is trying to prove a value.
00:13:06.581 --> 00:13:11.022
But even if there are certain values that are getting proved, there is no adoption because.
00:13:11.711 --> 00:13:13.392
It's not coming from the business.
00:13:13.991 --> 00:13:17.231
So what has happened right now is there's an organizational shift.
00:13:17.322 --> 00:13:18.822
We talked about change management.
00:13:19.302 --> 00:13:22.241
We talked about how it is an important management decision.
00:13:22.542 --> 00:13:30.032
So we are seeing and talking to a lot of our customers where we are educating them about, strategic way of adopting generated ai.
00:13:30.302 --> 00:13:33.542
And that strategic way of adopting generative AI is business first.
00:13:34.142 --> 00:13:37.142
So we have to bring business and technology in the same table.
00:13:37.591 --> 00:13:51.932
We have to talk to them, together, we have to talk to them about what value, like I talked about, as organizations, if you're looking into our organization and understanding why we are losing time, money, and growth opportunities, it has to come from business.
00:13:52.381 --> 00:13:57.182
And business can then tier down and tier up what kind of use cases they're looking at.
00:13:57.961 --> 00:14:02.322
And typically, again, technology can augment and help those decisions.
00:14:02.591 --> 00:14:04.272
They can help technologically.
00:14:05.201 --> 00:14:12.192
To tier down and say, oh, in my tier one I have these kind of use cases, which are high impact, but these are high risk.
00:14:12.461 --> 00:14:24.131
In the middle tier, we do have those use cases where we can have pretty significant opportunities of growth and we do not need, high volume, high risk speeder.
00:14:24.761 --> 00:14:35.042
We can go into a, tier three where the value of generative value use cases might not be significant, but we are also not looking into significant risks.
00:14:35.282 --> 00:14:47.486
So again, it's a measure of what is that fine line, what as an organization we are comfortable with a lot of time when we tear down and look into how we are going to.
00:14:48.432 --> 00:14:51.221
Position our use cases into these buckets.
00:14:51.491 --> 00:14:58.751
And what does it mean to onboard a new project and what does it mean to have ownership of the project in terms of outcome?
00:14:58.932 --> 00:15:11.432
But how does technology enable them through, like you said, data readiness, platform readiness technology adoption ensuring that the right guardrails, the right responsible ai, the safety nets are put in there.
00:15:11.761 --> 00:15:14.971
So again, technology will bring in those as an augmentation.
00:15:15.152 --> 00:15:23.611
But businesses driving the value by identifying the right use cases and technology can help them identify the right use cases in terms of risk profile.
00:15:24.152 --> 00:15:30.121
So once you do that, once you marry up the business and technology, then it becomes an easy, smoother, right?
00:15:30.121 --> 00:15:32.312
Then it becomes a game of adoption.
00:15:32.912 --> 00:15:36.152
So you have seen it more than again, in multiple ways.
00:15:36.182 --> 00:15:37.652
We have seen this multiple ways.
00:15:37.951 --> 00:15:43.081
A lot of time we have built brilliant, wonderful outcomes.
00:15:43.081 --> 00:15:54.701
We have built generative AI solutions that are cutting edge, but eventually we have seen no adoption and if we do not have any adoption, eventually it becomes, yet another experimentation.
00:15:55.226 --> 00:16:00.576
And it becomes yet another effort where, the organization is not getting a lot of value outta it.
00:16:01.027 --> 00:16:02.886
So again, like I said, it's a combination.
00:16:03.096 --> 00:16:15.277
But to my point, what I was trying to emphasize was it is a function of business adoption, business ownership, augmented by technology enablement.
00:16:15.876 --> 00:16:23.392
So I guess is there is there an early signal where this experiment is heading towards value rather than a proof of concept?
00:16:23.991 --> 00:16:24.741
Absolutely.
00:16:24.802 --> 00:16:27.621
And again, no, I come from a technology background.
00:16:27.621 --> 00:16:29.692
I can talk about how we have done it, right?
00:16:30.172 --> 00:16:36.392
So what we have done is we have created a framework and we call it a funnel framework, right?
00:16:36.392 --> 00:16:38.631
Again, the analogy is very simple.
00:16:39.231 --> 00:16:59.221
We have hundreds of ideas that actually come into our generative AI platform at any point of time, and out of hundreds of ideas, probably five of them, them actually go into production so we have created that setup where we have project onboarding and we have the use case onboarding process, which means.
00:16:59.912 --> 00:17:02.552
We work with our ciso, we work with our legal team.
00:17:02.552 --> 00:17:05.071
We work with our business to create a self-service form.
00:17:05.821 --> 00:17:14.402
We document and understand what use cases people want to bring in, and we allow our business to come and submit those forms.
00:17:15.182 --> 00:17:20.612
So as business owners, they can actually justify and discuss what is their value, what they want to solve for.
00:17:21.211 --> 00:17:27.852
What we do internally is we have actually converted those business request forms into technology workflows.
00:17:28.281 --> 00:17:38.821
We worked with our c cso, legal our engineering department, and we start providing weightage to these kind of ideas, and these weightage is actually convert to something meaningful.
00:17:38.852 --> 00:17:41.071
We can automatically bucket and tier those.
00:17:41.926 --> 00:17:48.237
So we can automatically know if a particular idea is high risk, low risk, high value, low value, and we create a rub rubric.
00:17:48.836 --> 00:17:56.517
So if a particular project is worth investing on, it actually gets onboarded into something we call it as a fail fast sandbox.
00:17:57.027 --> 00:18:00.926
And it is very important to your point, to have a concept of fail fast.
00:18:01.527 --> 00:18:08.067
So where we are differentiating ourselves from a technology point of view, we do not want to get stuck in that field loop forever.
00:18:08.666 --> 00:18:10.287
So we have technology enablement.
00:18:10.346 --> 00:18:14.487
Now we think of building generative AI solutions through a common foundation.
00:18:14.787 --> 00:18:23.067
When we onboard a particular use case into that platform, we will automatically see whether that use case is going somewhere.
00:18:23.436 --> 00:18:26.287
If we are providing value, success is good.
00:18:26.287 --> 00:18:28.487
If not we need to fail fast.
00:18:28.487 --> 00:18:30.267
We need to stop that buck right there.
00:18:30.866 --> 00:18:33.207
We do that through assessments.
00:18:33.207 --> 00:18:37.586
We have functional assessments where it can be human in the loop, it can be experts.
00:18:37.946 --> 00:18:42.416
We have technical assessments where we evaluate, scalability cost.
00:18:42.537 --> 00:18:49.977
We look into responsible AI biasness that comes out of the answers, the accuracy measures, the ground truth evaluations.
00:18:50.817 --> 00:18:57.646
We look into the finops we look into the observability, metricses, we look into the, all the logs and try to understand whether.
00:18:58.156 --> 00:19:03.616
That particular idea that we are trying to prove in that fail fast sandbox, is it worth investing on?
00:19:04.217 --> 00:19:13.426
So it has become like a standard for us where we are not going to keep an idea into a sandbox, which we call it as a fail fast sandbox for a long period of time.
00:19:13.666 --> 00:19:15.106
We have to prove a value out of it.
00:19:15.707 --> 00:19:19.156
And the value, like I said, is proven through technology assessment and human.
00:19:20.116 --> 00:19:35.057
If we can prove that value goes into scaled ideation, that is where we can start putting in the belts and whistles and we can convert that idea, which was a POC into something more meaningful like an MVP, and then we hand it over to the hands of business for a very small period of time.
00:19:35.146 --> 00:19:36.467
Again, shadow it and all.
00:19:37.186 --> 00:19:43.846
So we want to ensure that we are restricting business to have a good look and feel into what we have built, because again, they are the owners.
00:19:44.251 --> 00:19:52.602
In the scale ideation phase, we are understanding whether the outcome is, aligned to what the initial thought experiment was.
00:19:52.991 --> 00:20:01.092
And if it is, then again, we go through those entire assessment phase, and if everything looks good, then we start talking about packaging and putting into production through automation.
00:20:01.692 --> 00:20:04.301
So for us, it's a very stage gated process.
00:20:04.301 --> 00:20:11.201
There is a demand funnel, there is a fail fast sand sandbox where we try to establish value quickly.
00:20:11.741 --> 00:20:17.652
We do assessments, we do scaled ideation, we do reassessments, and then we decide whether it has to go into production.
00:20:17.981 --> 00:20:19.182
So it becomes a factory.
00:20:19.511 --> 00:20:40.271
And more importantly, we are continuously monitoring those factor, which means whoever is putting an idea into this funnel, whoever is talking about what is their business value, right to the assessments to fail fast, the experimentations, the outcomes, everything is documented, tracked, trace, and evaluated, and we call that as a control claim.
00:20:41.247 --> 00:20:47.576
So our control plane becomes a 360 view of what ideas are actually onboarded into a generative AI platform.
00:20:47.846 --> 00:20:51.446
How these ideas are actually converted, evaluated, validated.
00:20:51.717 --> 00:20:58.916
So tomorrow, if there is a decision that we have to take to our a RB saying, Hey, we have gone through the entire hoops.
00:20:59.186 --> 00:21:01.376
We have actually an idea that we want to productionize.
00:21:02.217 --> 00:21:03.686
We can trace that lineage.
00:21:03.686 --> 00:21:08.576
We can understand what all this experimentation has gone in, what kind of scenarios you have tested.
00:21:08.951 --> 00:21:09.761
What is the value?
00:21:09.761 --> 00:21:10.632
What is the ROI?
00:21:10.632 --> 00:21:11.412
What is the cost?
00:21:11.471 --> 00:21:18.731
We can track every single thing in the entire journey, and that journey can now be converted into something meaningful and then put into production.
00:21:19.271 --> 00:21:28.481
So again, every organization, I'm sure has their own process, have their own framework, have their own methodology of taking an idea into production.
00:21:28.481 --> 00:21:32.652
This is ours, but this kind of a methodology is very important and critical.
00:21:32.981 --> 00:21:37.332
So instead of saying, how do we go ahead and put an idea into production?
00:21:38.291 --> 00:21:40.781
Think to it from the lens of multiple stage gates.
00:21:41.142 --> 00:21:46.872
And for us, these stage gates is something that we ha that has evolved through a lots of, lot of lessons learned.
00:21:47.281 --> 00:21:53.971
We also have been stuck into PC land for a long time, and now over a period of time as we are learning our lessons.
00:21:54.557 --> 00:22:04.666
This kind of a factory model has become a standard practice for us, and that is what we have been following for onboarding new ideas and productionizing it, in a much more repeatable.
00:22:05.267 --> 00:22:07.636
Yeah, no and again, like you said, they're repeatable.
00:22:07.696 --> 00:22:08.537
That's you.
00:22:08.626 --> 00:22:10.156
If you're constantly doing it, you're gonna keep getting.
00:22:10.997 --> 00:22:14.477
You're gonna get results in the day'cause you're following a the same process every time.
00:22:14.797 --> 00:22:15.426
Which I love that.
00:22:15.696 --> 00:22:19.317
So then so you mentioned around like adopting generative ai.
00:22:19.797 --> 00:22:27.317
So I guess when it comes to what's the, I guess what's the biggest misconception businesses have when they start adopting generative ai?
00:22:27.497 --> 00:22:29.507
That, and that would catch them off guard.
00:22:30.106 --> 00:22:37.696
So again, when I have been talking to business internal and external, I've seen three kinds of ideologies coming through, right?
00:22:38.297 --> 00:22:44.057
So one is, there are business who thinks that generative AI can be a silver bullet, right?
00:22:44.057 --> 00:22:48.106
So when we have talked to them and understood what kind of use cases do you want to focus on?
00:22:48.886 --> 00:22:50.866
Do you have a well established process?
00:22:50.866 --> 00:22:52.396
Can we look into the documentation?
00:22:52.396 --> 00:22:54.467
Can we look into your existing process?
00:22:55.067 --> 00:23:00.856
So to augment, to reinvent your process, we have to understand your process.
00:23:00.856 --> 00:23:02.146
So where is your process?
00:23:02.146 --> 00:23:04.636
Who can help us understand and document those processes?
00:23:05.116 --> 00:23:06.376
Where is that tribal knowledge?
00:23:06.376 --> 00:23:08.987
People have been working on this process for 20 years.
00:23:09.586 --> 00:23:10.906
How are you getting it done?
00:23:10.906 --> 00:23:13.217
Is there a day in a life that we can actually focus on?
00:23:13.817 --> 00:23:14.116
Sorry.
00:23:14.942 --> 00:23:16.231
Where is the data coming from?
00:23:16.382 --> 00:23:18.422
What kind of data sources we are talking about?
00:23:18.422 --> 00:23:19.592
How ready is the data?
00:23:20.132 --> 00:23:27.487
And when we start asking these kind of questions, a lot of time businesses will come back with a very blank face and say, we have not thought about it.
00:23:28.426 --> 00:23:32.186
We believe that generative AI can actually, get everything done.
00:23:32.237 --> 00:23:34.547
We may or may not have all the processes established.
00:23:34.547 --> 00:23:36.616
We do not know where all the documentations are.
00:23:36.916 --> 00:23:39.166
We do not understand where our data comes from.
00:23:39.406 --> 00:23:44.717
We do not know how clean and how ready our generative AI use case data can be.
00:23:44.957 --> 00:23:48.616
But our thought is that is what generative AI is for.
00:23:48.616 --> 00:23:53.717
It's going to automatically fix all the broken things and come up with magical outcome.
00:23:54.271 --> 00:24:02.132
So I've seen people thinking of generative AI as a magic wand where it can fix things where people do not know what is broken.
00:24:02.731 --> 00:24:13.551
The other end of the spectrum is absolutely opposite, where we have gone and talked to our business partners and they're like, no, we are not going to adopt generative ai because every time I talk to you guys talk about augmentation.
00:24:13.642 --> 00:24:18.781
You do not talk about a hundred percent automation, which means, for them it becomes a binary game.
00:24:19.771 --> 00:24:28.382
If something, if people still need to invest some time as a human in the loop and look into the outcome generated by generative ai, for them it is not good enough.
00:24:28.832 --> 00:24:30.811
They're like, oh, we still have people.
00:24:30.872 --> 00:24:32.642
People have been here for 20 years.
00:24:32.971 --> 00:24:34.561
They know the process in and out.
00:24:34.832 --> 00:24:37.261
We do not think generative AI can do anything about it.
00:24:37.862 --> 00:24:44.811
It's going to be too much of a, effort for us, generative ai, instead of making our lives easier, it's going to add more work for us.
00:24:45.021 --> 00:24:50.902
So you have seen those kind of scenarios in the business where they're very apprehensive about changes.
00:24:51.231 --> 00:25:01.582
They feel that, they are subject matter experts and for them, generative AI becomes yet another tool and burden, which may or may not prove that value that they imagine it would.
00:25:01.701 --> 00:25:03.501
So these are the two ends of the spectrum.
00:25:04.011 --> 00:25:12.791
Where either, people think of this as a magic wand or people think of this as yet another tool that would add more chaos and value.
00:25:13.392 --> 00:25:22.632
So as technology partners, I think what our biggest challenge is to bring these kind of discussions into a common board, talk about what is real, right?
00:25:23.092 --> 00:25:27.221
And the reality is, it is, augmentation will come before automation.
00:25:28.152 --> 00:25:30.192
We will find out the right challenges.
00:25:30.461 --> 00:25:32.382
We might not automate everything.
00:25:32.471 --> 00:25:45.031
We can start with 60% automation, where a lot of these repeatable, redundant manual things can be automated and it can be just 60% and, it's okay to start with 60%, right?
00:25:45.031 --> 00:25:50.491
So the incremental value of generative ai, not everything is going to be fixed in day one.
00:25:50.932 --> 00:25:55.192
It takes some kind of an effort to get the data readiness done, platform readiness done.
00:25:55.701 --> 00:25:59.842
Technology is going to be an enabler, but somebody still has to define those processes.
00:26:00.741 --> 00:26:08.372
Somebody has to look into those business process with a fresh sense and a fresh lens and say, all right, I see your business process.
00:26:08.372 --> 00:26:17.211
This business process is too cumbersome because, we have been doing this for 20 years, but now with technology, we can actually streamline the process.
00:26:17.511 --> 00:26:19.551
We can fast track the decision making, right?
00:26:19.942 --> 00:26:23.382
So I think, ensuring that we can bring people into the same mindset.
00:26:23.892 --> 00:26:27.432
Giving them a true sense that it's going to be a very hydrating journey.
00:26:28.332 --> 00:26:31.961
It's not going to be a magic wand where in day one we can change everything.
00:26:32.561 --> 00:26:33.701
Data readiness is a thing.
00:26:33.912 --> 00:26:47.832
Understanding the business problem is a thing as long as we are actually talking in the right sense, giving them the right approach and educating them about what exactly is going to follow and what is that 30, 60, 90 date program.
00:26:48.356 --> 00:26:58.326
How these 30, 60, 90 day program is going to show them the incremental value and how over a period of time they're gonna see a lot more, value coming out of these initiatives.
00:26:58.537 --> 00:27:09.156
That is important because day one, if people are thinking that it's going to magically change the workforce or immediately show some ROI, I think that's a sticker problem.
00:27:09.906 --> 00:27:12.757
And it has happened not because of the organizational shift.
00:27:13.086 --> 00:27:14.586
I think that is in general, right?
00:27:14.586 --> 00:27:15.576
Generative ai.
00:27:16.537 --> 00:27:26.247
Is the first technology in my mind, in human history where it has been personally adopted first, then, being adopted in a corporate world, right?
00:27:26.257 --> 00:27:36.636
People have been using chat GPT before they started building their agent solution, which means the thing that I can ask any questions and get the right answer.
00:27:36.636 --> 00:27:40.817
So in their mind, generative AI was already something magical.
00:27:40.997 --> 00:27:44.152
So now that we are incorporating those things in into a corporate world.
00:27:44.751 --> 00:27:52.021
Understanding that it still requires a lot of work in terms of data technology, adoption, business value, business process.
00:27:52.201 --> 00:27:54.402
It comes as a sticker shock for a lot of our customers.
00:27:55.001 --> 00:27:58.912
So I guess, who needs to own generative AI inside an organization?
00:27:58.912 --> 00:28:02.211
Is it technology, data, product, or the business?
00:28:02.811 --> 00:28:03.442
Absolutely.
00:28:03.442 --> 00:28:04.251
It has to be business.
00:28:04.251 --> 00:28:11.602
We talk about the fact that the accountability of generative AI use cases, it has to be on the use case level.
00:28:11.602 --> 00:28:13.102
It has to be on a business level.
00:28:13.311 --> 00:28:14.902
It cannot be just on committees.
00:28:14.902 --> 00:28:17.852
It cannot be just on, technology enablement teams, right?
00:28:17.852 --> 00:28:23.582
If you put it in the hands of technology, they would just focus on the latest and greatest technology.
00:28:23.582 --> 00:28:25.142
They'll be running behind technology.
00:28:25.142 --> 00:28:29.102
They'll be talking about the right things to do in terms of technology adoption.
00:28:29.162 --> 00:28:31.031
But not ROI, not outcome, not growth.
00:28:31.632 --> 00:28:37.791
If you put it in the hands of data, it'll go into a into a rapid hole where we talk about is my data, right?
00:28:38.392 --> 00:28:42.622
How do I get my data right now, what are the different ways in which the data needs to be onboarded?
00:28:42.981 --> 00:28:48.951
How do you provide the access control, the security, and the, whole line yards of data governance.
00:28:49.551 --> 00:28:54.231
If you put it in the hands of products, they'll think of generative AI yet as another tool, right?
00:28:54.231 --> 00:28:55.402
Again, we'll talk about.
00:28:56.392 --> 00:28:59.092
How we can just convert generative AI into a tool.
00:29:00.051 --> 00:29:02.571
So the right ownership has to be on business.
00:29:02.781 --> 00:29:04.642
Business has to drive the value.
00:29:04.642 --> 00:29:10.551
They have to own the value, and like I said, the data, the technology, and the product teams, they have to enable business.
00:29:10.791 --> 00:29:12.321
At the end of the day, it has to be business.
00:29:12.321 --> 00:29:12.576
Who owns it?
00:29:13.176 --> 00:29:13.626
Okay.
00:29:13.747 --> 00:29:16.626
No, that's interesting And yeah, good to see.
00:29:17.017 --> 00:29:19.057
Going back to obviously looking at adoption again.
00:29:19.116 --> 00:29:23.896
You've got obviously security risk and ethics, often slow adoption.
00:29:24.317 --> 00:29:30.527
So how can you balance moving fast with being responsible without, like paralyzing the organization?
00:29:31.126 --> 00:29:32.257
Yeah, it's a great question.
00:29:32.257 --> 00:29:35.037
Again I hope, and there is a clear answer to this.
00:29:35.307 --> 00:29:37.166
Every organization struggles with this.
00:29:37.166 --> 00:29:41.307
Even our organization struggles, time to line, understanding where to draw the line.
00:29:41.906 --> 00:29:43.196
But one thing is clear, Ben.
00:29:44.112 --> 00:29:50.741
One thing that is clear is we need to separate out the policy from permission, and that has worked for us.
00:29:51.071 --> 00:29:54.432
So what I mean by that is AI policies will evolve.
00:29:55.132 --> 00:29:59.271
We cannot just wait for AI policies to be mature before we productionize anything.
00:29:59.422 --> 00:30:01.071
In that case, we'll never productionize anything.
00:30:01.821 --> 00:30:08.301
So in my mind, as long as we start with the right permissions, which means are we picking up the right use case?
00:30:08.422 --> 00:30:09.201
Does this right?
00:30:09.291 --> 00:30:10.731
Does the use case have.
00:30:11.362 --> 00:30:16.402
The right data governance, the right data auditability that is needed.
00:30:16.832 --> 00:30:24.781
Are we not opening up too much data from a permission point of view when we are actually exposing the data through generative AI use case?
00:30:25.442 --> 00:30:27.061
Do the people have the right permission?
00:30:27.422 --> 00:30:32.192
Are there rule based access control, attribute based access control, system based access control?
00:30:32.551 --> 00:30:34.321
Who gets to access what kind of data?
00:30:34.531 --> 00:30:47.251
So as long as we are providing outcome with a need to know basis, as long as we are providing outcome with what is the bare minimum data we can actually synthesize and put it in the hands of whoever's seeking the data, we should be good.
00:30:47.852 --> 00:30:55.051
So again, what I was trying to put with this is it's important to have technology permissions being very robust and solid.
00:30:55.457 --> 00:31:12.346
That is the reason we talked about data readiness and technology enablement for business, which means when a generative use case is there and this generative AI use case, for example, talks to five different systems, we need to ensure that the right permissions are in place for all these five systems.
00:31:12.586 --> 00:31:15.317
People can interact, access those five systems.
00:31:15.737 --> 00:31:20.416
And again, the amount of data is actually propagated point.
00:31:21.257 --> 00:31:23.836
organization have figured out all the AI policies?
00:31:23.896 --> 00:31:24.557
Probably not.
00:31:24.886 --> 00:31:29.176
Should we be waiting for all the AI policies to be in place before we can productionize not at all.
00:31:29.277 --> 00:31:30.207
So it's a fine line.
00:31:30.207 --> 00:31:40.317
So as long as we are talking about responsible AI incorporated into the design patterns, as long as safety is not an afterthought, as long as the right permissions and access controls are in place.
00:31:40.856 --> 00:31:50.217
As long as we do have AI policies, which is still evolving and we are continuously measuring, monitoring those AI policies, even though they're not profit should be.
00:31:50.217 --> 00:31:50.547
Okay.
00:31:51.116 --> 00:32:04.977
So I'm a firm believer where as long as we are providing the right access, the right permissions, and as long as we are working with the right people within the organization, we talked about security and CISO and forming those policies.
00:32:05.711 --> 00:32:06.582
We should be good.
00:32:06.642 --> 00:32:11.227
No, we should not just wait for something to become perfect because in that case, we just chase our tails.
00:32:11.826 --> 00:32:15.932
Have you seen teams that move too fast and that sort and pay the price?
00:32:16.041 --> 00:32:19.461
Pay the price for bypassing governance or compliance?
00:32:19.652 --> 00:32:20.281
Guardrails.
00:32:20.882 --> 00:32:27.692
I've seen people doing that and I've seen where organizations would have a very quick time to provide an MVP.
00:32:28.096 --> 00:32:36.467
They might actually move into an MVP phase way sooner than any other company, but eventually they'll get stuck in that MVP phase.
00:32:36.517 --> 00:32:40.356
They will not productionize things, and that is where we run the risk of shadow it.
00:32:40.537 --> 00:32:50.616
I've seen many a times where their AI policies, their responsible AI policies, their safety breaks, their security and governance, was nothing in there, was nothing in place.
00:32:51.037 --> 00:32:53.666
They still, pushed for creating those MVPs.
00:32:54.041 --> 00:33:00.311
Eventually the MVPs were success from a business point of view, but nobody was giving them permission to go to production.
00:33:00.402 --> 00:33:10.632
So what happened was it become, yet, it became yet another MVP that was getting stuck in this shadow IT business was using those tools on a sporadic basis.
00:33:10.902 --> 00:33:12.551
It was never going into production.
00:33:12.551 --> 00:33:13.692
There was no adoption.
00:33:13.842 --> 00:33:16.632
But then there will be small fragments, there will be small.
00:33:17.531 --> 00:33:22.902
Groups of business who are using it on their day-to-day basis, because now, MVP became a thing.
00:33:23.201 --> 00:33:33.311
So again, if you get stuck into this kind of a scenario where there is a shadow, it nobody's making your asset or solution go into production, it's not an ideal thing.
00:33:33.432 --> 00:33:41.201
And when you talk about multiple use cases getting stuck in that MVP and the shadow, it grows and grows, it becomes a nightmare for an organization.
00:33:41.682 --> 00:33:44.771
So again what we talk to our customer is very clear.
00:33:45.386 --> 00:33:47.037
Do not think about security.
00:33:47.037 --> 00:33:49.527
Do not think about AI policies as an afterthought.
00:33:49.977 --> 00:33:52.826
Always ensure that you have the right guardrails in place.
00:33:52.826 --> 00:33:53.666
From day one.
00:33:54.116 --> 00:33:55.497
Your policies will mature.
00:33:55.557 --> 00:34:00.086
We'll come back and reframe some of the policies, but start with those permissions.
00:34:00.086 --> 00:34:06.957
Start with the policies that is approved, and ensure that we are not cutting commerce in terms of governance risk and mitigation at least.
00:34:07.741 --> 00:34:08.487
Okay, brilliant.
00:34:08.516 --> 00:34:09.146
No, I like that.
00:34:09.507 --> 00:34:21.606
And so then if say for a company listening today and they wanted to move from sort of gen AI curiosity to real impact, say in the next six to 12 months, what would be your advice?
00:34:21.637 --> 00:34:22.817
What to prioritize first?
00:34:23.416 --> 00:34:25.396
So I think it should be a dual velocity approach.
00:34:25.396 --> 00:34:30.947
Again we talk about this many times, so we talk about there are two things that needs to happen in parallel.
00:34:31.396 --> 00:34:35.237
So one is from a technology point of view, think about building a common gen.
00:34:35.237 --> 00:34:35.777
AI found.
00:34:36.766 --> 00:34:52.422
Common Gen AI foundation are those repeatable, reusable, standardized governance and ready to monitor, blocks that will ensure that, if you're building one use case or if you're building a hundred use cases, you have that common set of repeatable capability.
00:34:53.211 --> 00:35:07.911
So for example, data onboarding, data readiness, semantically generation, your agents framework, your scaffolding, your responsible ai, your control planes, your governance, and monitoring your testing framework your adoption framework.
00:35:08.181 --> 00:35:16.791
All of these are repeatable, reusable technology enablers where you can build them in a way that these are common across all use cases.
00:35:17.257 --> 00:35:33.516
So the most important thing that I go and talk to my customer is that dual velocity, where at the bottom part of dual velocity, you need to focus on these repeatable, reusable modules that makes generative AI adoption technologically faster and standardized.
00:35:34.056 --> 00:35:39.606
The other part of dual velocity is that business adoption, which means work with business in parallel.
00:35:39.637 --> 00:35:41.887
So again the key word here is dual velocity.
00:35:42.336 --> 00:35:43.387
The word dual, which means.
00:35:43.896 --> 00:35:46.536
We are not waiting on technology to finish their work.
00:35:46.597 --> 00:35:49.567
We are not waiting on technology to start building their foundation.
00:35:50.016 --> 00:35:55.777
But at the same time, while technology is building their foundation, we need to ensure that there are right business owners.
00:35:56.137 --> 00:36:05.197
We work with our business partners, try to understand all the use cases that are important, critical within the organization, and we tiered them up, right?
00:36:05.257 --> 00:36:10.702
Like I said, we find out which are these use cases, the good that go into bucket one, bucket two, and bucket three.
00:36:11.362 --> 00:36:12.652
We try to provide value.
00:36:12.652 --> 00:36:16.851
So value realization, value evaluation of those use cases are very critical.
00:36:17.362 --> 00:36:22.762
And then while the technology enablement is there, we identify the top two.
00:36:22.762 --> 00:36:23.512
Top three, right?
00:36:23.572 --> 00:36:25.942
Do not dilute your generative AI adoption.
00:36:26.331 --> 00:36:27.472
Focus on the top two.
00:36:27.472 --> 00:36:28.972
Focus on the top three first.
00:36:29.302 --> 00:36:33.052
Find out those use cases, which are maybe low hanging fruits.
00:36:33.081 --> 00:36:35.032
Maybe provide better value at low risk.
00:36:35.871 --> 00:36:37.851
Then just start incorporating, right?
00:36:37.851 --> 00:36:43.762
The worst thing that can happen to an organization is to wait on the side and hope things to standardize.
00:36:43.762 --> 00:36:45.681
Instead of that, just dive in.
00:36:46.132 --> 00:36:48.532
Dive in with some low risk use cases.
00:36:48.902 --> 00:36:50.492
Focus on those use cases.
00:36:50.791 --> 00:36:52.351
Do not dilute your portfolio.
00:36:52.351 --> 00:36:55.561
Do not go after 10, 20, 30 different use cases on day one.
00:36:56.072 --> 00:36:57.242
Find those one, two, or three.
00:36:57.242 --> 00:36:57.931
Use case.
00:36:58.322 --> 00:36:59.371
Have a timeline.
00:36:59.581 --> 00:37:01.981
Say that in six months time, I need to productionize it.
00:37:02.416 --> 00:37:06.077
Have the buy-in from business and ensure that technology gonna be supporting.
00:37:06.496 --> 00:37:24.606
So as long as you're doing this dual velocity, as long as you are finding those right use cases, grooming those use cases, ensuring that these use cases are good to go, and technology is building those repeatable, standardized framework and the platform, then if you combine them together, it becomes a solid way of solving multiple use cases.
00:37:24.996 --> 00:37:31.387
And in the six months period, when you build that solution, after six months, you now have created yourself a machine.
00:37:31.822 --> 00:37:42.931
You have created yourself a process where, you know the fourth and the fifth and the sixth use cases, they have no time to market because you already have a solid Gen I foundation.
00:37:43.291 --> 00:37:47.132
There are common, repeatable, standardized building blocks that are already in place.
00:37:47.581 --> 00:37:53.311
You already have the right engineering in place and you also have the generative AI adoption process in place, right?
00:37:53.552 --> 00:37:57.961
So you combine them together, like I said, through dual velocity that becomes a building block.
00:37:57.961 --> 00:37:58.532
So again.
00:37:59.447 --> 00:38:01.456
Are all organizations following them?
00:38:01.547 --> 00:38:02.177
Probably not.
00:38:02.657 --> 00:38:12.507
But again, if I have to provide 2 cents to anybody, again, through you who are listening in my opinion, this kind of a dual velocity methodology and everybody can find their own methodologist.
00:38:12.507 --> 00:38:20.092
But again, in my opinion, this kind of a dual velocity where think from repeatability, think from reusability, things from, standardization from day one.
00:38:21.016 --> 00:38:28.067
Find out those use cases, have business, own it, and deliver through technology is the one approach that has worked.
00:38:28.456 --> 00:38:37.461
And as long as we are putting those approaches in, in, like I said, six months to 12 months timeframe and ensuring that there is something tangible coming out of it, that's the right approach.
00:38:38.061 --> 00:38:38.362
Brilliant.
00:38:38.422 --> 00:38:43.402
That's some great in insights and yeah, listeners listening will find that, yeah, truly your value.
00:38:43.452 --> 00:38:45.612
And so I think we'll leave it there today.
00:38:45.681 --> 00:38:47.331
So thank you for your conversation.
00:38:47.931 --> 00:38:49.461
Thank you so much, Ben, for having me here.
00:38:49.612 --> 00:38:51.922
It hasn't been, it has been an absolute pleasure.
00:38:52.141 --> 00:38:56.461
Again, thank you for all these insightful questions and good luck.
00:38:57.061 --> 00:38:59.371
Hopefully you're all finding these conversations useful.
00:38:59.532 --> 00:39:04.751
Please make sure you follow or subscribe data analytics chat so you don't miss future episodes.
00:39:05.052 --> 00:39:08.411
Thanks for listening, and I'll see you in the next episode.
00:00:00.000 --> 00:00:05.790
We reached a position, we reached a place where there were 3000 experiments running in a particular sandbox at a particular time.
00:00:06.089 --> 00:00:17.710
We have built generative AI solutions that are cutting edge, but eventually we have seen no adoption and if we do not have any adoption, eventually it becomes, yet another experimentation.
00:00:18.234 --> 00:00:26.094
So I've seen people thinking of generative AI as a magic wand where it can fix things where people do not know what is broken.
00:00:26.394 --> 00:00:27.385
Generative ai.
00:00:28.344 --> 00:00:38.054
Is the first technology in my mind, in human history where it has been personally adopted first, then, being adopted in a corporate world, right?
00:00:52.097 --> 00:00:59.777
Almost every organization today is experimenting with generative ai, but very few are seeing meaningful impact at scale.
00:01:00.436 --> 00:01:07.522
The hardest part isn't the technology, it's the work behind the scenes, data foundations, operating models, and how decisions get made.
00:01:08.486 --> 00:01:18.867
In today's episode of Data Analytics Chat, I'm joined by Naan Paul, who's the managing Director for Chief Generative AI Center of Excellence at Accenture.
00:01:19.227 --> 00:01:24.266
And today we'll discuss what it really takes to adopt generative AI inside an organization.
00:01:24.867 --> 00:01:32.996
What separates experimentation from impact where companies get stuck, and what leaders should prioritize over the next six to 12 months.
00:01:33.537 --> 00:01:34.887
Naan great to have you here.
00:01:35.486 --> 00:01:36.596
Thank you so much, Ben.
00:01:36.867 --> 00:01:37.137
Thank you.
00:01:37.167 --> 00:01:39.447
Thank for hosting me and I'm really excited to be here.
00:01:39.807 --> 00:01:40.167
Thank you.
00:01:40.736 --> 00:01:46.557
So I guess, Anne, before we dive in, do you wanna give listeners a quick introduction to who you are and the work you are leading today?
00:01:47.156 --> 00:01:47.966
Absolutely.
00:01:48.337 --> 00:01:49.566
My name is Naan Paul.
00:01:49.566 --> 00:01:55.746
I'm based out of us, and for me, generative AI has been a very organic growth.
00:01:56.496 --> 00:01:58.236
Five years ago, my organization
00:01:58.537 --> 00:01:58.596
to
00:01:59.346 --> 00:02:01.087
opened up the Center of Excellence.
00:02:01.147 --> 00:02:06.216
It's a global entity, and as a part of this COE, as a part of the global organization.
00:02:06.816 --> 00:02:11.347
I supposed to understand help customers build
00:02:12.037 --> 00:02:12.456
when,
00:02:12.997 --> 00:02:13.717
generative ai.
00:02:14.317 --> 00:02:20.312
So for me, I have three major three ways that I contribute as a part of.
00:02:20.912 --> 00:02:23.106
One is being part of the consulting work.
00:02:23.706 --> 00:02:25.206
I work with my customers.
00:02:25.266 --> 00:02:27.247
I try to understand their North Star vision.
00:02:27.306 --> 00:02:28.611
I try to understand their gaps.
00:02:29.211 --> 00:02:36.981
I try to understand what it'll take for them, like you said, to move from an experimentation mindset to a productionized mindset.
00:02:37.581 --> 00:02:41.451
Two, I work with my products teams.
00:02:41.542 --> 00:02:54.502
I work with my partners to understand what it means to build solutions, jumpstart kits, frameworks, the entire scaffolding bit, because when we are talking about enterprise adoption, when we are talking about scalability.
00:02:55.192 --> 00:02:59.331
It is very important to do it right and do it in a repeatable way.
00:02:59.662 --> 00:03:02.481
We talk about building the right foundation.
00:03:02.542 --> 00:03:08.872
We talk about how generative AI needs to be designed in a way that we are not just chasing use cases and demos.
00:03:09.141 --> 00:03:10.581
We are chasing outcomes.
00:03:10.972 --> 00:03:13.942
So what it means to build those foundational capabilities.
00:03:14.331 --> 00:03:18.141
And three, because of how we are working very closely with our partners.
00:03:19.132 --> 00:03:23.301
It's about aligning our roadmap to the roadmap of our partners.
00:03:23.602 --> 00:03:27.891
What are, for example, Microsoft offering AWS offering, GCP offering.
00:03:28.431 --> 00:03:42.106
Now, what is Nvidia offering and how as Accenture, we are incorporating those understandings, incorporating those newness into our storyline, into our adoption, and how can we bring the latest thinking to our customers?
00:03:42.646 --> 00:03:43.787
So this is my day-to-day.
00:03:43.817 --> 00:03:46.967
This is how you know we have approached generative ai.
00:03:47.567 --> 00:03:55.192
From, like you said, change management to having ownership to ensuring that we are about the bigger picture and.
00:03:55.792 --> 00:03:56.092
Brilliant.
00:03:56.092 --> 00:04:02.657
Just so the listeners know, obviously, so are you specialized in one industry Accenture, or is it across the board?
00:04:03.257 --> 00:04:05.356
So within Accenture, we have multiple groups.
00:04:05.687 --> 00:04:09.747
However, my group as I was talking about, is part of Center of Advanced ai.
00:04:09.836 --> 00:04:12.896
So Center of Advanced AI is a very horizontal entity.
00:04:13.706 --> 00:04:21.836
We are supposed to be the tip of the sphere where we are supposed to understand, incorporate, and in vibe what is the latest and greatest technology offering.
00:04:22.242 --> 00:04:26.831
So my team and I, we specialize mostly across domains, across verticals.
00:04:27.132 --> 00:04:29.291
We work with our functional leads.
00:04:29.351 --> 00:04:34.572
We work with our domain experts, the subject matter experts, but we bring in the technology bits.
00:04:34.752 --> 00:04:41.531
We understand what is the rate latest offering, what it means to scale, what it means to secure.
00:04:41.591 --> 00:04:44.891
We talk about responsible AI security safety breaks.
00:04:45.252 --> 00:04:47.651
So we will bring in mostly the horizontal technology.
00:04:48.252 --> 00:04:58.942
We talk and adoption and others that we work with, like our business partners, our SMEs, our subject domain experts, they will bring in the functional knowledge.
00:04:58.971 --> 00:05:00.802
So together we will solve problems for.
00:05:01.401 --> 00:05:01.822
Brilliant.
00:05:01.851 --> 00:05:05.471
So you'll have a good blend of use cases, which would be great for the listeners today.
00:05:06.312 --> 00:05:16.932
And I guess something before we move on to the data topic, has there been a experience that you've gone through that's most shaped how you approach leading AI initiatives in organizations today?
00:05:17.531 --> 00:05:18.172
A hundred percent.
00:05:18.276 --> 00:05:20.346
Like it has been a growth cycle.
00:05:20.406 --> 00:05:21.786
It has been a lot of lessons learned for us.
00:05:22.387 --> 00:05:29.197
I remember five years ago when we started off this journey, our intention was how do we ensure that we are educating people?
00:05:30.007 --> 00:05:32.076
So we created our own sandboxes.
00:05:32.586 --> 00:05:41.997
We actually advertised, we actually, put in a lot of money where we asked our account teams, like we, we have multiple account teams.
00:05:41.997 --> 00:05:46.257
We asked, sorry we asked our account teams to onboard.
00:05:46.812 --> 00:05:48.432
Into these sandboxes.
00:05:48.701 --> 00:05:52.721
We provided, tenancy, we provided Azure services.
00:05:53.081 --> 00:05:56.721
We were early on bringing in, LLM models.
00:05:56.781 --> 00:06:02.791
We were exposing those LLM models and we are requesting all these account teams to go, and experiment.
00:06:03.692 --> 00:06:09.901
So the initial days, it was all about getting familiar, ensuring that people are not looking away.
00:06:10.651 --> 00:06:23.192
Giving them a safe space, giving them all the right tools and technologies, giving them the right support to ensure that people can bring in their thought process, they can bring in their ideas and just experiment, right?
00:06:23.461 --> 00:06:32.851
So we focused on ensuring that, just keep on experimenting, whatever thoughts you have, irrespective of whether these outcomes are going to add value or not.
00:06:33.041 --> 00:06:33.762
Just experiment.
00:06:34.617 --> 00:06:40.406
We reached a position, we reached a place where there were 3000 experiments running in a particular sandbox at a particular time.
00:06:41.007 --> 00:06:51.396
So what it did was we ensured that different account teams who are working with their respective clients, they were able to tell a story, they were able to showcase something.
00:06:51.817 --> 00:06:53.646
And again, showing is believing.
00:06:54.442 --> 00:06:58.567
And for the first one and a half years, we are doing a lot of this experimentation.
00:06:58.927 --> 00:07:02.166
We were not productionizing anything, but that was okay with us.
00:07:02.976 --> 00:07:06.666
Again, as time went by, we realized that we have to shift the gear.
00:07:06.997 --> 00:07:11.526
We have to understand what it means to just stop experimenting and start focusing on value.
00:07:11.627 --> 00:07:20.387
So for the next one, one and a half years we actually promoted, we talked about how we have to change our mindset from being, just experimenting to.
00:07:20.822 --> 00:07:31.531
Identifying, bubbling up the right use cases, focusing on what is important, and use those lessons learned from experimentation to start providing real value outcomes.
00:07:31.862 --> 00:07:35.791
And when we started building those real value outcomes, we saw the real challenges.
00:07:35.791 --> 00:07:37.291
We talked about data readiness.
00:07:37.291 --> 00:07:39.391
We talked about the operating model.
00:07:39.391 --> 00:07:42.182
You talked about the scalability, adoption.
00:07:42.182 --> 00:07:46.396
You talked about the real ROIs, and we've stumbled upon those challenges too.
00:07:47.161 --> 00:07:59.312
As we stumbled upon those challenges, we tried to understand whether it is a technology problem, whether it is an adoption problem, whether it is a responsible problem in terms of ownership, and we started mitigating a lot of those.
00:07:59.672 --> 00:08:00.872
So over a period of time.
00:08:00.872 --> 00:08:03.922
Now, if we look back, we did the right approach in my opinion.
00:08:03.971 --> 00:08:09.041
We didn't just go ahead and start saying, Hey, we are not going to start, or we are not going to enable anybody.
00:08:09.357 --> 00:08:12.206
To use generative AI unless you really provide a value.
00:08:12.266 --> 00:08:16.466
No, we said, don't worry about the first initial phase, just experiment.
00:08:16.586 --> 00:08:17.937
Just get yourself comfortable.
00:08:18.266 --> 00:08:20.487
Just ensure that you are technology aligned.
00:08:20.906 --> 00:08:23.716
And then over a period of time, we learned our lessons.
00:08:23.987 --> 00:08:29.896
We brought in business, we got the right buy-ins, and again, we organically grown into.
00:08:30.257 --> 00:08:32.206
What and how we can deliver values.
00:08:32.506 --> 00:08:36.197
So if you look back into those five years, it has been a very incremental journey for us.
00:08:36.197 --> 00:08:55.386
And now when we talk about business process automation, when we talk about augmentation, when we talk about, identifying those key functional capabilities within an organization and look into those from the lens of automation and augmentation, we understand what it takes, because of how the journey has been propagated for us.
00:08:55.787 --> 00:08:57.767
Long story, Ben, again, to your point.
00:08:58.667 --> 00:08:59.027
Yes.
00:08:59.027 --> 00:09:04.937
Now the important bits are ensuring that we are not just stuck into that POC world.
00:09:04.996 --> 00:09:07.096
We are not just looking into experimentation.
00:09:07.126 --> 00:09:09.937
We are not stuck, getting stuck into that mindset.
00:09:10.297 --> 00:09:16.927
But the journey had to start somewhere, and the journey for us started five years ago, and now it's all about value.
00:09:17.047 --> 00:09:22.606
It's all about looking into, as a company and saying, what is, what are those important things?
00:09:23.162 --> 00:09:27.581
Which are making us lose time, which is making us lose, money or growth opportunities.
00:09:27.581 --> 00:09:36.562
So now we are looking into generative AI opportunities from the lens of saving time, making money or growing as an organization.
00:09:36.772 --> 00:09:41.871
So again, to reach that particular milestone, it was a lot of hard work.
00:09:41.881 --> 00:09:48.287
It was a lot of technology adoption from my team and it was a lot of business alignment that we worked together with.
00:09:48.886 --> 00:09:56.017
Yeah, and I love the fact where you say, now it's about creating value because look, we're going for a transformational period now, isn't it?
00:09:56.017 --> 00:10:02.736
With businesses, and if you've got every business department coming to with you with a use case, you can't do it all, can you?
00:10:02.736 --> 00:10:08.706
You need to be choosing the one that's going to make the biggest ROI or the biggest impact for your business right now.
00:10:08.706 --> 00:10:09.667
You can't be doing everything.
00:10:10.267 --> 00:10:11.101
A hundred percent.
00:10:11.341 --> 00:10:12.812
And I think that's a mind shift.
00:10:12.871 --> 00:10:14.267
Shift change to your point.
00:10:14.866 --> 00:10:25.657
We couldn't have just talked about value from day one to your point, but now when we talk about how we are approaching generative ai, it's all about, who is owning it.
00:10:26.197 --> 00:10:27.817
It cannot own generative ai.
00:10:27.856 --> 00:10:37.037
If you give it the responsibility of owning generative ai, they'll just focus on architecture tools and what is the right technology stack.
00:10:37.831 --> 00:10:39.841
That is not the right measure of generative ai.
00:10:39.841 --> 00:10:43.532
Again, you mentioned it, and again, I think everybody understands it now.
00:10:44.131 --> 00:10:47.792
The real value of generative AI now is how is it transforming business?
00:10:47.851 --> 00:10:49.292
What is that underlying ROI?
00:10:49.892 --> 00:11:03.182
We are not just talking about automation because yes, automation is a good measure, but in my opinion, it's all about augmentation first, which means can we identify those important low risk, high impact use cases?
00:11:03.902 --> 00:11:20.341
Can we find out those business scenarios where the augmentation or automation can work hand in hand and allow businesses to own solution that can reduce, for example, the effort, the time, or even FTEs, right?
00:11:20.942 --> 00:11:32.131
And when we start looking into generative BI use cases from the lens of what can organizations do, how the decision making can be made faster, how the decision making can be de-risked.
00:11:32.731 --> 00:11:38.101
Those are the right measures to pick up generative use cases now, and that is how we are approaching it now.
00:11:38.131 --> 00:11:41.642
So again, the experimentation phase was important.
00:11:41.761 --> 00:11:55.677
Maybe it was important three years ago, four years ago, but now with all the technology changes with the way in which we are actually approaching generative ai, the way in which we should be approaching generative AI is business adoption and the ownership.
00:11:56.326 --> 00:12:01.216
The ownership at any point of time at this day and age is all about business.
00:12:01.317 --> 00:12:02.937
How are they looking into business?
00:12:02.937 --> 00:12:09.397
How are they looking into augmentation on automation and what business processes they feel comfortable, augmenting with generative ai.
00:12:09.846 --> 00:12:11.981
So I think those are the right questions we need to ask right now.
00:12:12.876 --> 00:12:13.386
Brilliant.
00:12:13.657 --> 00:12:17.307
So I guess most leaders feel pressure to do something with generative ai.
00:12:17.996 --> 00:12:20.621
In your experience, what separates organizations that.
00:12:21.221 --> 00:12:23.052
From those that create value.
00:12:23.652 --> 00:12:25.812
It's a multidimensional question to be honest.
00:12:25.812 --> 00:12:27.761
And you are, you're so right to point it out.
00:12:28.302 --> 00:12:32.172
So I have seen organizations where there is a lot of IT dependency.
00:12:32.772 --> 00:12:39.522
A lot of focus has gone into the CTO office where the intention is, do you have the right infrastructure?
00:12:39.971 --> 00:12:41.831
Do we have the right technology stack?
00:12:42.131 --> 00:12:42.881
What is that right?
00:12:42.881 --> 00:12:43.662
Technology stack.
00:12:44.501 --> 00:12:51.761
Can we set up the right technology foundation and can we onboard business and allow them to come and experiment?
00:12:52.152 --> 00:12:59.922
And that is a methodology where a lot of companies have gone into, and for the last two, two and a half years, they have been stuck into this POC loop, right?
00:13:00.522 --> 00:13:02.351
Technology is driving Gen ai.
00:13:02.351 --> 00:13:04.392
There is no adoption.
00:13:04.721 --> 00:13:06.552
Business is trying to prove a value.
00:13:06.581 --> 00:13:11.022
But even if there are certain values that are getting proved, there is no adoption because.
00:13:11.711 --> 00:13:13.392
It's not coming from the business.
00:13:13.991 --> 00:13:17.231
So what has happened right now is there's an organizational shift.
00:13:17.322 --> 00:13:18.822
We talked about change management.
00:13:19.302 --> 00:13:22.241
We talked about how it is an important management decision.
00:13:22.542 --> 00:13:30.032
So we are seeing and talking to a lot of our customers where we are educating them about, strategic way of adopting generated ai.
00:13:30.302 --> 00:13:33.542
And that strategic way of adopting generative AI is business first.
00:13:34.142 --> 00:13:37.142
So we have to bring business and technology in the same table.
00:13:37.591 --> 00:13:51.932
We have to talk to them, together, we have to talk to them about what value, like I talked about, as organizations, if you're looking into our organization and understanding why we are losing time, money, and growth opportunities, it has to come from business.
00:13:52.381 --> 00:13:57.182
And business can then tier down and tier up what kind of use cases they're looking at.
00:13:57.961 --> 00:14:02.322
And typically, again, technology can augment and help those decisions.
00:14:02.591 --> 00:14:04.272
They can help technologically.
00:14:05.201 --> 00:14:12.192
To tier down and say, oh, in my tier one I have these kind of use cases, which are high impact, but these are high risk.
00:14:12.461 --> 00:14:24.131
In the middle tier, we do have those use cases where we can have pretty significant opportunities of growth and we do not need, high volume, high risk speeder.
00:14:24.761 --> 00:14:35.042
We can go into a, tier three where the value of generative value use cases might not be significant, but we are also not looking into significant risks.
00:14:35.282 --> 00:14:47.486
So again, it's a measure of what is that fine line, what as an organization we are comfortable with a lot of time when we tear down and look into how we are going to.
00:14:48.432 --> 00:14:51.221
Position our use cases into these buckets.
00:14:51.491 --> 00:14:58.751
And what does it mean to onboard a new project and what does it mean to have ownership of the project in terms of outcome?
00:14:58.932 --> 00:15:11.432
But how does technology enable them through, like you said, data readiness, platform readiness technology adoption ensuring that the right guardrails, the right responsible ai, the safety nets are put in there.
00:15:11.761 --> 00:15:14.971
So again, technology will bring in those as an augmentation.
00:15:15.152 --> 00:15:23.611
But businesses driving the value by identifying the right use cases and technology can help them identify the right use cases in terms of risk profile.
00:15:24.152 --> 00:15:30.121
So once you do that, once you marry up the business and technology, then it becomes an easy, smoother, right?
00:15:30.121 --> 00:15:32.312
Then it becomes a game of adoption.
00:15:32.912 --> 00:15:36.152
So you have seen it more than again, in multiple ways.
00:15:36.182 --> 00:15:37.652
We have seen this multiple ways.
00:15:37.951 --> 00:15:43.081
A lot of time we have built brilliant, wonderful outcomes.
00:15:43.081 --> 00:15:54.701
We have built generative AI solutions that are cutting edge, but eventually we have seen no adoption and if we do not have any adoption, eventually it becomes, yet another experimentation.
00:15:55.226 --> 00:16:00.576
And it becomes yet another effort where, the organization is not getting a lot of value outta it.
00:16:01.027 --> 00:16:02.886
So again, like I said, it's a combination.
00:16:03.096 --> 00:16:15.277
But to my point, what I was trying to emphasize was it is a function of business adoption, business ownership, augmented by technology enablement.
00:16:15.876 --> 00:16:23.392
So I guess is there is there an early signal where this experiment is heading towards value rather than a proof of concept?
00:16:23.991 --> 00:16:24.741
Absolutely.
00:16:24.802 --> 00:16:27.621
And again, no, I come from a technology background.
00:16:27.621 --> 00:16:29.692
I can talk about how we have done it, right?
00:16:30.172 --> 00:16:36.392
So what we have done is we have created a framework and we call it a funnel framework, right?
00:16:36.392 --> 00:16:38.631
Again, the analogy is very simple.
00:16:39.231 --> 00:16:59.221
We have hundreds of ideas that actually come into our generative AI platform at any point of time, and out of hundreds of ideas, probably five of them, them actually go into production so we have created that setup where we have project onboarding and we have the use case onboarding process, which means.
00:16:59.912 --> 00:17:02.552
We work with our ciso, we work with our legal team.
00:17:02.552 --> 00:17:05.071
We work with our business to create a self-service form.
00:17:05.821 --> 00:17:14.402
We document and understand what use cases people want to bring in, and we allow our business to come and submit those forms.
00:17:15.182 --> 00:17:20.612
So as business owners, they can actually justify and discuss what is their value, what they want to solve for.
00:17:21.211 --> 00:17:27.852
What we do internally is we have actually converted those business request forms into technology workflows.
00:17:28.281 --> 00:17:38.821
We worked with our c cso, legal our engineering department, and we start providing weightage to these kind of ideas, and these weightage is actually convert to something meaningful.
00:17:38.852 --> 00:17:41.071
We can automatically bucket and tier those.
00:17:41.926 --> 00:17:48.237
So we can automatically know if a particular idea is high risk, low risk, high value, low value, and we create a rub rubric.
00:17:48.836 --> 00:17:56.517
So if a particular project is worth investing on, it actually gets onboarded into something we call it as a fail fast sandbox.
00:17:57.027 --> 00:18:00.926
And it is very important to your point, to have a concept of fail fast.
00:18:01.527 --> 00:18:08.067
So where we are differentiating ourselves from a technology point of view, we do not want to get stuck in that field loop forever.
00:18:08.666 --> 00:18:10.287
So we have technology enablement.
00:18:10.346 --> 00:18:14.487
Now we think of building generative AI solutions through a common foundation.
00:18:14.787 --> 00:18:23.067
When we onboard a particular use case into that platform, we will automatically see whether that use case is going somewhere.
00:18:23.436 --> 00:18:26.287
If we are providing value, success is good.
00:18:26.287 --> 00:18:28.487
If not we need to fail fast.
00:18:28.487 --> 00:18:30.267
We need to stop that buck right there.
00:18:30.866 --> 00:18:33.207
We do that through assessments.
00:18:33.207 --> 00:18:37.586
We have functional assessments where it can be human in the loop, it can be experts.
00:18:37.946 --> 00:18:42.416
We have technical assessments where we evaluate, scalability cost.
00:18:42.537 --> 00:18:49.977
We look into responsible AI biasness that comes out of the answers, the accuracy measures, the ground truth evaluations.
00:18:50.817 --> 00:18:57.646
We look into the finops we look into the observability, metricses, we look into the, all the logs and try to understand whether.
00:18:58.156 --> 00:19:03.616
That particular idea that we are trying to prove in that fail fast sandbox, is it worth investing on?
00:19:04.217 --> 00:19:13.426
So it has become like a standard for us where we are not going to keep an idea into a sandbox, which we call it as a fail fast sandbox for a long period of time.
00:19:13.666 --> 00:19:15.106
We have to prove a value out of it.
00:19:15.707 --> 00:19:19.156
And the value, like I said, is proven through technology assessment and human.
00:19:20.116 --> 00:19:35.057
If we can prove that value goes into scaled ideation, that is where we can start putting in the belts and whistles and we can convert that idea, which was a POC into something more meaningful like an MVP, and then we hand it over to the hands of business for a very small period of time.
00:19:35.146 --> 00:19:36.467
Again, shadow it and all.
00:19:37.186 --> 00:19:43.846
So we want to ensure that we are restricting business to have a good look and feel into what we have built, because again, they are the owners.
00:19:44.251 --> 00:19:52.602
In the scale ideation phase, we are understanding whether the outcome is, aligned to what the initial thought experiment was.
00:19:52.991 --> 00:20:01.092
And if it is, then again, we go through those entire assessment phase, and if everything looks good, then we start talking about packaging and putting into production through automation.
00:20:01.692 --> 00:20:04.301
So for us, it's a very stage gated process.
00:20:04.301 --> 00:20:11.201
There is a demand funnel, there is a fail fast sand sandbox where we try to establish value quickly.
00:20:11.741 --> 00:20:17.652
We do assessments, we do scaled ideation, we do reassessments, and then we decide whether it has to go into production.
00:20:17.981 --> 00:20:19.182
So it becomes a factory.
00:20:19.511 --> 00:20:40.271
And more importantly, we are continuously monitoring those factor, which means whoever is putting an idea into this funnel, whoever is talking about what is their business value, right to the assessments to fail fast, the experimentations, the outcomes, everything is documented, tracked, trace, and evaluated, and we call that as a control claim.
00:20:41.247 --> 00:20:47.576
So our control plane becomes a 360 view of what ideas are actually onboarded into a generative AI platform.
00:20:47.846 --> 00:20:51.446
How these ideas are actually converted, evaluated, validated.
00:20:51.717 --> 00:20:58.916
So tomorrow, if there is a decision that we have to take to our a RB saying, Hey, we have gone through the entire hoops.
00:20:59.186 --> 00:21:01.376
We have actually an idea that we want to productionize.
00:21:02.217 --> 00:21:03.686
We can trace that lineage.
00:21:03.686 --> 00:21:08.576
We can understand what all this experimentation has gone in, what kind of scenarios you have tested.
00:21:08.951 --> 00:21:09.761
What is the value?
00:21:09.761 --> 00:21:10.632
What is the ROI?
00:21:10.632 --> 00:21:11.412
What is the cost?
00:21:11.471 --> 00:21:18.731
We can track every single thing in the entire journey, and that journey can now be converted into something meaningful and then put into production.
00:21:19.271 --> 00:21:28.481
So again, every organization, I'm sure has their own process, have their own framework, have their own methodology of taking an idea into production.
00:21:28.481 --> 00:21:32.652
This is ours, but this kind of a methodology is very important and critical.
00:21:32.981 --> 00:21:37.332
So instead of saying, how do we go ahead and put an idea into production?
00:21:38.291 --> 00:21:40.781
Think to it from the lens of multiple stage gates.
00:21:41.142 --> 00:21:46.872
And for us, these stage gates is something that we ha that has evolved through a lots of, lot of lessons learned.
00:21:47.281 --> 00:21:53.971
We also have been stuck into PC land for a long time, and now over a period of time as we are learning our lessons.
00:21:54.557 --> 00:22:04.666
This kind of a factory model has become a standard practice for us, and that is what we have been following for onboarding new ideas and productionizing it, in a much more repeatable.
00:22:05.267 --> 00:22:07.636
Yeah, no and again, like you said, they're repeatable.
00:22:07.696 --> 00:22:08.537
That's you.
00:22:08.626 --> 00:22:10.156
If you're constantly doing it, you're gonna keep getting.
00:22:10.997 --> 00:22:14.477
You're gonna get results in the day'cause you're following a the same process every time.
00:22:14.797 --> 00:22:15.426
Which I love that.
00:22:15.696 --> 00:22:19.317
So then so you mentioned around like adopting generative ai.
00:22:19.797 --> 00:22:27.317
So I guess when it comes to what's the, I guess what's the biggest misconception businesses have when they start adopting generative ai?
00:22:27.497 --> 00:22:29.507
That, and that would catch them off guard.
00:22:30.106 --> 00:22:37.696
So again, when I have been talking to business internal and external, I've seen three kinds of ideologies coming through, right?
00:22:38.297 --> 00:22:44.057
So one is, there are business who thinks that generative AI can be a silver bullet, right?
00:22:44.057 --> 00:22:48.106
So when we have talked to them and understood what kind of use cases do you want to focus on?
00:22:48.886 --> 00:22:50.866
Do you have a well established process?
00:22:50.866 --> 00:22:52.396
Can we look into the documentation?
00:22:52.396 --> 00:22:54.467
Can we look into your existing process?
00:22:55.067 --> 00:23:00.856
So to augment, to reinvent your process, we have to understand your process.
00:23:00.856 --> 00:23:02.146
So where is your process?
00:23:02.146 --> 00:23:04.636
Who can help us understand and document those processes?
00:23:05.116 --> 00:23:06.376
Where is that tribal knowledge?
00:23:06.376 --> 00:23:08.987
People have been working on this process for 20 years.
00:23:09.586 --> 00:23:10.906
How are you getting it done?
00:23:10.906 --> 00:23:13.217
Is there a day in a life that we can actually focus on?
00:23:13.817 --> 00:23:14.116
Sorry.
00:23:14.942 --> 00:23:16.231
Where is the data coming from?
00:23:16.382 --> 00:23:18.422
What kind of data sources we are talking about?
00:23:18.422 --> 00:23:19.592
How ready is the data?
00:23:20.132 --> 00:23:27.487
And when we start asking these kind of questions, a lot of time businesses will come back with a very blank face and say, we have not thought about it.
00:23:28.426 --> 00:23:32.186
We believe that generative AI can actually, get everything done.
00:23:32.237 --> 00:23:34.547
We may or may not have all the processes established.
00:23:34.547 --> 00:23:36.616
We do not know where all the documentations are.
00:23:36.916 --> 00:23:39.166
We do not understand where our data comes from.
00:23:39.406 --> 00:23:44.717
We do not know how clean and how ready our generative AI use case data can be.
00:23:44.957 --> 00:23:48.616
But our thought is that is what generative AI is for.
00:23:48.616 --> 00:23:53.717
It's going to automatically fix all the broken things and come up with magical outcome.
00:23:54.271 --> 00:24:02.132
So I've seen people thinking of generative AI as a magic wand where it can fix things where people do not know what is broken.
00:24:02.731 --> 00:24:13.551
The other end of the spectrum is absolutely opposite, where we have gone and talked to our business partners and they're like, no, we are not going to adopt generative ai because every time I talk to you guys talk about augmentation.
00:24:13.642 --> 00:24:18.781
You do not talk about a hundred percent automation, which means, for them it becomes a binary game.
00:24:19.771 --> 00:24:28.382
If something, if people still need to invest some time as a human in the loop and look into the outcome generated by generative ai, for them it is not good enough.
00:24:28.832 --> 00:24:30.811
They're like, oh, we still have people.
00:24:30.872 --> 00:24:32.642
People have been here for 20 years.
00:24:32.971 --> 00:24:34.561
They know the process in and out.
00:24:34.832 --> 00:24:37.261
We do not think generative AI can do anything about it.
00:24:37.862 --> 00:24:44.811
It's going to be too much of a, effort for us, generative ai, instead of making our lives easier, it's going to add more work for us.
00:24:45.021 --> 00:24:50.902
So you have seen those kind of scenarios in the business where they're very apprehensive about changes.
00:24:51.231 --> 00:25:01.582
They feel that, they are subject matter experts and for them, generative AI becomes yet another tool and burden, which may or may not prove that value that they imagine it would.
00:25:01.701 --> 00:25:03.501
So these are the two ends of the spectrum.
00:25:04.011 --> 00:25:12.791
Where either, people think of this as a magic wand or people think of this as yet another tool that would add more chaos and value.
00:25:13.392 --> 00:25:22.632
So as technology partners, I think what our biggest challenge is to bring these kind of discussions into a common board, talk about what is real, right?
00:25:23.092 --> 00:25:27.221
And the reality is, it is, augmentation will come before automation.
00:25:28.152 --> 00:25:30.192
We will find out the right challenges.
00:25:30.461 --> 00:25:32.382
We might not automate everything.
00:25:32.471 --> 00:25:45.031
We can start with 60% automation, where a lot of these repeatable, redundant manual things can be automated and it can be just 60% and, it's okay to start with 60%, right?
00:25:45.031 --> 00:25:50.491
So the incremental value of generative ai, not everything is going to be fixed in day one.
00:25:50.932 --> 00:25:55.192
It takes some kind of an effort to get the data readiness done, platform readiness done.
00:25:55.701 --> 00:25:59.842
Technology is going to be an enabler, but somebody still has to define those processes.
00:26:00.741 --> 00:26:08.372
Somebody has to look into those business process with a fresh sense and a fresh lens and say, all right, I see your business process.
00:26:08.372 --> 00:26:17.211
This business process is too cumbersome because, we have been doing this for 20 years, but now with technology, we can actually streamline the process.
00:26:17.511 --> 00:26:19.551
We can fast track the decision making, right?
00:26:19.942 --> 00:26:23.382
So I think, ensuring that we can bring people into the same mindset.
00:26:23.892 --> 00:26:27.432
Giving them a true sense that it's going to be a very hydrating journey.
00:26:28.332 --> 00:26:31.961
It's not going to be a magic wand where in day one we can change everything.
00:26:32.561 --> 00:26:33.701
Data readiness is a thing.
00:26:33.912 --> 00:26:47.832
Understanding the business problem is a thing as long as we are actually talking in the right sense, giving them the right approach and educating them about what exactly is going to follow and what is that 30, 60, 90 date program.
00:26:48.356 --> 00:26:58.326
How these 30, 60, 90 day program is going to show them the incremental value and how over a period of time they're gonna see a lot more, value coming out of these initiatives.
00:26:58.537 --> 00:27:09.156
That is important because day one, if people are thinking that it's going to magically change the workforce or immediately show some ROI, I think that's a sticker problem.
00:27:09.906 --> 00:27:12.757
And it has happened not because of the organizational shift.
00:27:13.086 --> 00:27:14.586
I think that is in general, right?
00:27:14.586 --> 00:27:15.576
Generative ai.
00:27:16.537 --> 00:27:26.247
Is the first technology in my mind, in human history where it has been personally adopted first, then, being adopted in a corporate world, right?
00:27:26.257 --> 00:27:36.636
People have been using chat GPT before they started building their agent solution, which means the thing that I can ask any questions and get the right answer.
00:27:36.636 --> 00:27:40.817
So in their mind, generative AI was already something magical.
00:27:40.997 --> 00:27:44.152
So now that we are incorporating those things in into a corporate world.
00:27:44.751 --> 00:27:52.021
Understanding that it still requires a lot of work in terms of data technology, adoption, business value, business process.
00:27:52.201 --> 00:27:54.402
It comes as a sticker shock for a lot of our customers.
00:27:55.001 --> 00:27:58.912
So I guess, who needs to own generative AI inside an organization?
00:27:58.912 --> 00:28:02.211
Is it technology, data, product, or the business?
00:28:02.811 --> 00:28:03.442
Absolutely.
00:28:03.442 --> 00:28:04.251
It has to be business.
00:28:04.251 --> 00:28:11.602
We talk about the fact that the accountability of generative AI use cases, it has to be on the use case level.
00:28:11.602 --> 00:28:13.102
It has to be on a business level.
00:28:13.311 --> 00:28:14.902
It cannot be just on committees.
00:28:14.902 --> 00:28:17.852
It cannot be just on, technology enablement teams, right?
00:28:17.852 --> 00:28:23.582
If you put it in the hands of technology, they would just focus on the latest and greatest technology.
00:28:23.582 --> 00:28:25.142
They'll be running behind technology.
00:28:25.142 --> 00:28:29.102
They'll be talking about the right things to do in terms of technology adoption.
00:28:29.162 --> 00:28:31.031
But not ROI, not outcome, not growth.
00:28:31.632 --> 00:28:37.791
If you put it in the hands of data, it'll go into a into a rapid hole where we talk about is my data, right?
00:28:38.392 --> 00:28:42.622
How do I get my data right now, what are the different ways in which the data needs to be onboarded?
00:28:42.981 --> 00:28:48.951
How do you provide the access control, the security, and the, whole line yards of data governance.
00:28:49.551 --> 00:28:54.231
If you put it in the hands of products, they'll think of generative AI yet as another tool, right?
00:28:54.231 --> 00:28:55.402
Again, we'll talk about.
00:28:56.392 --> 00:28:59.092
How we can just convert generative AI into a tool.
00:29:00.051 --> 00:29:02.571
So the right ownership has to be on business.
00:29:02.781 --> 00:29:04.642
Business has to drive the value.
00:29:04.642 --> 00:29:10.551
They have to own the value, and like I said, the data, the technology, and the product teams, they have to enable business.
00:29:10.791 --> 00:29:12.321
At the end of the day, it has to be business.
00:29:12.321 --> 00:29:12.576
Who owns it?
00:29:13.176 --> 00:29:13.626
Okay.
00:29:13.747 --> 00:29:16.626
No, that's interesting And yeah, good to see.
00:29:17.017 --> 00:29:19.057
Going back to obviously looking at adoption again.
00:29:19.116 --> 00:29:23.896
You've got obviously security risk and ethics, often slow adoption.
00:29:24.317 --> 00:29:30.527
So how can you balance moving fast with being responsible without, like paralyzing the organization?
00:29:31.126 --> 00:29:32.257
Yeah, it's a great question.
00:29:32.257 --> 00:29:35.037
Again I hope, and there is a clear answer to this.
00:29:35.307 --> 00:29:37.166
Every organization struggles with this.
00:29:37.166 --> 00:29:41.307
Even our organization struggles, time to line, understanding where to draw the line.
00:29:41.906 --> 00:29:43.196
But one thing is clear, Ben.
00:29:44.112 --> 00:29:50.741
One thing that is clear is we need to separate out the policy from permission, and that has worked for us.
00:29:51.071 --> 00:29:54.432
So what I mean by that is AI policies will evolve.
00:29:55.132 --> 00:29:59.271
We cannot just wait for AI policies to be mature before we productionize anything.
00:29:59.422 --> 00:30:01.071
In that case, we'll never productionize anything.
00:30:01.821 --> 00:30:08.301
So in my mind, as long as we start with the right permissions, which means are we picking up the right use case?
00:30:08.422 --> 00:30:09.201
Does this right?
00:30:09.291 --> 00:30:10.731
Does the use case have.
00:30:11.362 --> 00:30:16.402
The right data governance, the right data auditability that is needed.
00:30:16.832 --> 00:30:24.781
Are we not opening up too much data from a permission point of view when we are actually exposing the data through generative AI use case?
00:30:25.442 --> 00:30:27.061
Do the people have the right permission?
00:30:27.422 --> 00:30:32.192
Are there rule based access control, attribute based access control, system based access control?
00:30:32.551 --> 00:30:34.321
Who gets to access what kind of data?
00:30:34.531 --> 00:30:47.251
So as long as we are providing outcome with a need to know basis, as long as we are providing outcome with what is the bare minimum data we can actually synthesize and put it in the hands of whoever's seeking the data, we should be good.
00:30:47.852 --> 00:30:55.051
So again, what I was trying to put with this is it's important to have technology permissions being very robust and solid.
00:30:55.457 --> 00:31:12.346
That is the reason we talked about data readiness and technology enablement for business, which means when a generative use case is there and this generative AI use case, for example, talks to five different systems, we need to ensure that the right permissions are in place for all these five systems.
00:31:12.586 --> 00:31:15.317
People can interact, access those five systems.
00:31:15.737 --> 00:31:20.416
And again, the amount of data is actually propagated point.
00:31:21.257 --> 00:31:23.836
organization have figured out all the AI policies?
00:31:23.896 --> 00:31:24.557
Probably not.
00:31:24.886 --> 00:31:29.176
Should we be waiting for all the AI policies to be in place before we can productionize not at all.
00:31:29.277 --> 00:31:30.207
So it's a fine line.
00:31:30.207 --> 00:31:40.317
So as long as we are talking about responsible AI incorporated into the design patterns, as long as safety is not an afterthought, as long as the right permissions and access controls are in place.
00:31:40.856 --> 00:31:50.217
As long as we do have AI policies, which is still evolving and we are continuously measuring, monitoring those AI policies, even though they're not profit should be.
00:31:50.217 --> 00:31:50.547
Okay.
00:31:51.116 --> 00:32:04.977
So I'm a firm believer where as long as we are providing the right access, the right permissions, and as long as we are working with the right people within the organization, we talked about security and CISO and forming those policies.
00:32:05.711 --> 00:32:06.582
We should be good.
00:32:06.642 --> 00:32:11.227
No, we should not just wait for something to become perfect because in that case, we just chase our tails.
00:32:11.826 --> 00:32:15.932
Have you seen teams that move too fast and that sort and pay the price?
00:32:16.041 --> 00:32:19.461
Pay the price for bypassing governance or compliance?
00:32:19.652 --> 00:32:20.281
Guardrails.
00:32:20.882 --> 00:32:27.692
I've seen people doing that and I've seen where organizations would have a very quick time to provide an MVP.
00:32:28.096 --> 00:32:36.467
They might actually move into an MVP phase way sooner than any other company, but eventually they'll get stuck in that MVP phase.
00:32:36.517 --> 00:32:40.356
They will not productionize things, and that is where we run the risk of shadow it.
00:32:40.537 --> 00:32:50.616
I've seen many a times where their AI policies, their responsible AI policies, their safety breaks, their security and governance, was nothing in there, was nothing in place.
00:32:51.037 --> 00:32:53.666
They still, pushed for creating those MVPs.
00:32:54.041 --> 00:33:00.311
Eventually the MVPs were success from a business point of view, but nobody was giving them permission to go to production.
00:33:00.402 --> 00:33:10.632
So what happened was it become, yet, it became yet another MVP that was getting stuck in this shadow IT business was using those tools on a sporadic basis.
00:33:10.902 --> 00:33:12.551
It was never going into production.
00:33:12.551 --> 00:33:13.692
There was no adoption.
00:33:13.842 --> 00:33:16.632
But then there will be small fragments, there will be small.
00:33:17.531 --> 00:33:22.902
Groups of business who are using it on their day-to-day basis, because now, MVP became a thing.
00:33:23.201 --> 00:33:33.311
So again, if you get stuck into this kind of a scenario where there is a shadow, it nobody's making your asset or solution go into production, it's not an ideal thing.
00:33:33.432 --> 00:33:41.201
And when you talk about multiple use cases getting stuck in that MVP and the shadow, it grows and grows, it becomes a nightmare for an organization.
00:33:41.682 --> 00:33:44.771
So again what we talk to our customer is very clear.
00:33:45.386 --> 00:33:47.037
Do not think about security.
00:33:47.037 --> 00:33:49.527
Do not think about AI policies as an afterthought.
00:33:49.977 --> 00:33:52.826
Always ensure that you have the right guardrails in place.
00:33:52.826 --> 00:33:53.666
From day one.
00:33:54.116 --> 00:33:55.497
Your policies will mature.
00:33:55.557 --> 00:34:00.086
We'll come back and reframe some of the policies, but start with those permissions.
00:34:00.086 --> 00:34:06.957
Start with the policies that is approved, and ensure that we are not cutting commerce in terms of governance risk and mitigation at least.
00:34:07.741 --> 00:34:08.487
Okay, brilliant.
00:34:08.516 --> 00:34:09.146
No, I like that.
00:34:09.507 --> 00:34:21.606
And so then if say for a company listening today and they wanted to move from sort of gen AI curiosity to real impact, say in the next six to 12 months, what would be your advice?
00:34:21.637 --> 00:34:22.817
What to prioritize first?
00:34:23.416 --> 00:34:25.396
So I think it should be a dual velocity approach.
00:34:25.396 --> 00:34:30.947
Again we talk about this many times, so we talk about there are two things that needs to happen in parallel.
00:34:31.396 --> 00:34:35.237
So one is from a technology point of view, think about building a common gen.
00:34:35.237 --> 00:34:35.777
AI found.
00:34:36.766 --> 00:34:52.422
Common Gen AI foundation are those repeatable, reusable, standardized governance and ready to monitor, blocks that will ensure that, if you're building one use case or if you're building a hundred use cases, you have that common set of repeatable capability.
00:34:53.211 --> 00:35:07.911
So for example, data onboarding, data readiness, semantically generation, your agents framework, your scaffolding, your responsible ai, your control planes, your governance, and monitoring your testing framework your adoption framework.
00:35:08.181 --> 00:35:16.791
All of these are repeatable, reusable technology enablers where you can build them in a way that these are common across all use cases.
00:35:17.257 --> 00:35:33.516
So the most important thing that I go and talk to my customer is that dual velocity, where at the bottom part of dual velocity, you need to focus on these repeatable, reusable modules that makes generative AI adoption technologically faster and standardized.
00:35:34.056 --> 00:35:39.606
The other part of dual velocity is that business adoption, which means work with business in parallel.
00:35:39.637 --> 00:35:41.887
So again the key word here is dual velocity.
00:35:42.336 --> 00:35:43.387
The word dual, which means.
00:35:43.896 --> 00:35:46.536
We are not waiting on technology to finish their work.
00:35:46.597 --> 00:35:49.567
We are not waiting on technology to start building their foundation.
00:35:50.016 --> 00:35:55.777
But at the same time, while technology is building their foundation, we need to ensure that there are right business owners.
00:35:56.137 --> 00:36:05.197
We work with our business partners, try to understand all the use cases that are important, critical within the organization, and we tiered them up, right?
00:36:05.257 --> 00:36:10.702
Like I said, we find out which are these use cases, the good that go into bucket one, bucket two, and bucket three.
00:36:11.362 --> 00:36:12.652
We try to provide value.
00:36:12.652 --> 00:36:16.851
So value realization, value evaluation of those use cases are very critical.
00:36:17.362 --> 00:36:22.762
And then while the technology enablement is there, we identify the top two.
00:36:22.762 --> 00:36:23.512
Top three, right?
00:36:23.572 --> 00:36:25.942
Do not dilute your generative AI adoption.
00:36:26.331 --> 00:36:27.472
Focus on the top two.
00:36:27.472 --> 00:36:28.972
Focus on the top three first.
00:36:29.302 --> 00:36:33.052
Find out those use cases, which are maybe low hanging fruits.
00:36:33.081 --> 00:36:35.032
Maybe provide better value at low risk.
00:36:35.871 --> 00:36:37.851
Then just start incorporating, right?
00:36:37.851 --> 00:36:43.762
The worst thing that can happen to an organization is to wait on the side and hope things to standardize.
00:36:43.762 --> 00:36:45.681
Instead of that, just dive in.
00:36:46.132 --> 00:36:48.532
Dive in with some low risk use cases.
00:36:48.902 --> 00:36:50.492
Focus on those use cases.
00:36:50.791 --> 00:36:52.351
Do not dilute your portfolio.
00:36:52.351 --> 00:36:55.561
Do not go after 10, 20, 30 different use cases on day one.
00:36:56.072 --> 00:36:57.242
Find those one, two, or three.
00:36:57.242 --> 00:36:57.931
Use case.
00:36:58.322 --> 00:36:59.371
Have a timeline.
00:36:59.581 --> 00:37:01.981
Say that in six months time, I need to productionize it.
00:37:02.416 --> 00:37:06.077
Have the buy-in from business and ensure that technology gonna be supporting.
00:37:06.496 --> 00:37:24.606
So as long as you're doing this dual velocity, as long as you are finding those right use cases, grooming those use cases, ensuring that these use cases are good to go, and technology is building those repeatable, standardized framework and the platform, then if you combine them together, it becomes a solid way of solving multiple use cases.
00:37:24.996 --> 00:37:31.387
And in the six months period, when you build that solution, after six months, you now have created yourself a machine.
00:37:31.822 --> 00:37:42.931
You have created yourself a process where, you know the fourth and the fifth and the sixth use cases, they have no time to market because you already have a solid Gen I foundation.
00:37:43.291 --> 00:37:47.132
There are common, repeatable, standardized building blocks that are already in place.
00:37:47.581 --> 00:37:53.311
You already have the right engineering in place and you also have the generative AI adoption process in place, right?
00:37:53.552 --> 00:37:57.961
So you combine them together, like I said, through dual velocity that becomes a building block.
00:37:57.961 --> 00:37:58.532
So again.
00:37:59.447 --> 00:38:01.456
Are all organizations following them?
00:38:01.547 --> 00:38:02.177
Probably not.
00:38:02.657 --> 00:38:12.507
But again, if I have to provide 2 cents to anybody, again, through you who are listening in my opinion, this kind of a dual velocity methodology and everybody can find their own methodologist.
00:38:12.507 --> 00:38:20.092
But again, in my opinion, this kind of a dual velocity where think from repeatability, think from reusability, things from, standardization from day one.
00:38:21.016 --> 00:38:28.067
Find out those use cases, have business, own it, and deliver through technology is the one approach that has worked.
00:38:28.456 --> 00:38:37.461
And as long as we are putting those approaches in, in, like I said, six months to 12 months timeframe and ensuring that there is something tangible coming out of it, that's the right approach.
00:38:38.061 --> 00:38:38.362
Brilliant.
00:38:38.422 --> 00:38:43.402
That's some great in insights and yeah, listeners listening will find that, yeah, truly your value.
00:38:43.452 --> 00:38:45.612
And so I think we'll leave it there today.
00:38:45.681 --> 00:38:47.331
So thank you for your conversation.
00:38:47.931 --> 00:38:49.461
Thank you so much, Ben, for having me here.
00:38:49.612 --> 00:38:51.922
It hasn't been, it has been an absolute pleasure.
00:38:52.141 --> 00:38:56.461
Again, thank you for all these insightful questions and good luck.
00:38:57.061 --> 00:38:59.371
Hopefully you're all finding these conversations useful.
00:38:59.532 --> 00:39:04.751
Please make sure you follow or subscribe data analytics chat so you don't miss future episodes.
00:39:05.052 --> 00:39:08.411
Thanks for listening, and I'll see you in the next episode.
