Engineering Trustworthy Enterprise AI

WEBVTT
00:00:01.439 --> 00:00:19.440
Hey everybody, really interesting chat today on why enterprise AI reliability is becoming a much bigger challenge as companies move from experimentation to autonomous agents running real businesses and business operations with Sushant from TCS.
00:00:19.600 --> 00:00:21.039
Sushant, how are you?
00:00:21.519 --> 00:00:22.399
Very well, Evan.
00:00:22.559 --> 00:00:23.039
Thank you.
00:00:23.120 --> 00:00:24.399
It's a pleasure to be here.
00:00:24.800 --> 00:00:28.640
Well, thanks for having me and uh look forward to this discussion.
00:00:28.879 --> 00:00:32.799
Before that, how would you describe the work you're doing at TCS these days?
00:00:32.960 --> 00:00:34.240
Who is TCS?
00:00:34.799 --> 00:00:42.159
So, TCS, we are a technology services company, technology and business process services company, the second largest in our space.
00:00:42.399 --> 00:00:49.520
And we have taken this bold ambition to be the largest AI-led services company on the planet, and we are well on our way to do that.
00:00:49.920 --> 00:01:03.759
My role in TCS is I lead technology transformation for North America, which is all the cool buzzwords you can think of: AI, digital, including cybersecurity, and some of those older relics like cloud in the AI world.
00:01:03.840 --> 00:01:06.959
So all of that is my remit, and that's what I do for TCS.
00:01:07.040 --> 00:01:08.640
I'm based out of Houston.
00:01:09.120 --> 00:01:09.840
Fantastic.
00:01:09.920 --> 00:01:17.200
And you and your team talk to CEI, uh CIOs, CEOs, CXOs, everyday technology leaders in particular.
00:01:17.439 --> 00:01:25.599
Talk about the conversations you're having today, how have they changed around AI, Egypt AI in particular over the past 12 months?
00:01:26.159 --> 00:01:42.079
I think it's uh a really exciting time, you know, in terms of what's going on in the industry, because there are a lot of fundamental shifts that we're witnessing on a very profound scale in the industry, in our interactions that I have with the C-suite.
00:01:42.319 --> 00:01:52.159
And you're right that uh if I go back a little bit, a lot of the technology conversations tend to be strictly in the CIO, the CTO, the CDO realm.
00:01:52.400 --> 00:01:58.159
Now they've shifted to the CEO, the COO, you know, the board, and that kind of realm.
00:01:58.239 --> 00:02:08.479
Because what's happened is technology, what was considered to be an enabler of automation, has now become an enabler of intelligence for the business, for the enterprise.
00:02:08.639 --> 00:02:15.599
So that's essentially led to maybe three things, I'll say, that we're seeing a lot in the in the market these days.
00:02:15.840 --> 00:02:23.759
One is, you know, every CEO is asking what can technology do to transform my business outcomes.
00:02:24.319 --> 00:02:33.520
The second is from an AI paradigm, people have done enough of experimentation, you know, little things in the crucibles, use cases, and so on.
00:02:33.599 --> 00:02:37.439
They all want to scale it now to the enterprise level to get those benefits.
00:02:37.680 --> 00:02:44.159
And then the third thing is when that's happening, the obvious question is can we rely, is AI reliable?
00:02:44.319 --> 00:02:48.400
Can we rely on the outcomes, the business outcomes produced by AI?
00:02:48.560 --> 00:02:51.840
So that's what what we're witnessing in the marketplace these days.
00:02:52.479 --> 00:02:53.199
Fascinating.
00:02:53.280 --> 00:03:09.520
And you you commissioned some research in which uh TCS and HFS found that only 35% of these executives you you talk about believe AI consistently delivers the software, I'm sorry, the business outcomes that they anticipate or expect or promise.
00:03:09.759 --> 00:03:11.759
What's going wrong here today?
00:03:12.400 --> 00:03:14.159
I think a couple dimensions there.
00:03:14.240 --> 00:03:24.080
You're right, we commissioned a study very recently with HFS, and the sample size was about 101 executives, if I remember right, all of them C-suite executives.
00:03:24.400 --> 00:03:31.360
And, you know, the funny thing in this world that's been going on is everybody talks about ROI, but AI was predominantly on investments.
00:03:31.520 --> 00:03:34.560
Nobody was thinking about a return on those investments.
00:03:34.719 --> 00:03:41.439
They were heavily, heavily indexed on the eye, but now people are coming back and asking the question where's my return on those investments?
00:03:41.680 --> 00:03:47.680
And what is moving the needle in terms of the business outcome, in terms of you know, these initiatives.
00:03:47.840 --> 00:03:52.800
So that's what the executives are looking for, and that's where the conversation is changing these days.
00:03:53.199 --> 00:03:53.919
Fascinating.
00:03:54.000 --> 00:03:56.879
I can't read, wait to read through that report.
00:03:57.120 --> 00:04:07.439
And of course, you you're experts in IT, and over decades, you've you've measured IT reliability, you know, thinking about uptime and availability.
00:04:07.520 --> 00:04:09.919
We all know what happens when IT goes wrong.
00:04:10.000 --> 00:04:13.039
We see it in the headlines in the Wall Street Journal.
00:04:13.199 --> 00:04:18.639
But why doesn't that definition, the old definition, work for AI?
00:04:20.160 --> 00:04:28.240
So I sound like an academic here, but if you if it humor me for a moment, and then I'll go back to the core question that you asked.
00:04:28.560 --> 00:04:35.839
Reliability is about consistency and stability at a very, very fundamental definition level.
00:04:36.240 --> 00:04:44.720
Now, how the definition of reliability has morphed and layers have gotten added to it, especially from an IT standpoint, you're absolutely right.
00:04:44.959 --> 00:04:47.519
Back in the day, it was about system availability.
00:04:47.759 --> 00:04:52.079
Is this a system available whatever time or is this a latency that is expected?
00:04:52.240 --> 00:04:54.879
And then we started designing the resilient systems.
00:04:55.120 --> 00:05:02.160
And then as we got into the boom of the cloud era around the pandemic, you know, everybody was embracing cloud.
00:05:02.399 --> 00:05:08.000
I saw reliability move from what I call as component reliability to service reliability.
00:05:08.079 --> 00:05:13.839
And without sounding too technical, let me kind of elaborate on that with a metaphor and then I'll take it to the AI realm.
00:05:14.000 --> 00:05:15.519
Think about an orchestra.
00:05:15.839 --> 00:05:22.240
You have the band with different instruments, you know, you may have a guitar, you may have a piano, you may have the drums, timbales, what have you.
00:05:22.319 --> 00:05:24.720
And each instrument may be perfectly tuned.
00:05:24.879 --> 00:05:30.639
That is, each component is reliable and it's supposed to do what it's and it's doing what it's supposed to do.
00:05:30.800 --> 00:05:40.240
But in terms of service reliability, when they start playing the orchestra, if they're not in sync, the audience, the hearer, the listener doesn't enjoy the music.
00:05:40.399 --> 00:05:41.759
And it's about that experience.
00:05:41.839 --> 00:05:43.680
So that's the outcome that you're trying to deliver.
00:05:43.759 --> 00:05:50.720
So service reliability does not look at each component, but as a whole, what service is being delivered and is that reliable?
00:05:50.800 --> 00:05:54.720
So that's the morphing that happened with the cloud coming in.
00:05:54.879 --> 00:05:57.120
And then AI took it up another notch.
00:05:57.279 --> 00:06:00.800
A couple other aspects kind of got layered on top of that.
00:06:00.959 --> 00:06:13.120
That is trust, dependability, because now AI is taking the main stage in terms of like you initially described, not just giving suggestions or advice, but also going to the outcomes.
00:06:13.279 --> 00:06:21.199
So if I may put it this way, and then probably I'll make it more real with an example and a business outcome tied to that.
00:06:22.000 --> 00:06:26.319
Back in the day, it was is the system working the way it was designed to work?
00:06:27.040 --> 00:06:37.199
The early days of AI, a couple of years ago, if I go back, and I'm talking more gen AI than the traditional AI, it became can we trust the answer that AI is giving?
00:06:37.439 --> 00:06:50.480
And now the question is if you're making it autonomous agency system, can we trust the outcome that AI is kind of, you know, steering you towards or the actions that AI is taking?
00:06:50.639 --> 00:06:54.639
Let me make it a little bit more real and then I'll pause for your thoughts, your reflections.
00:06:54.879 --> 00:06:59.360
One of our clients, it's a very, very large uh medical distribution company.
00:06:59.519 --> 00:07:02.240
They are like Fortune 10 class company.
00:07:02.879 --> 00:07:09.120
And they have a bunch of systems and different companies, and they've grown through acquisitions over the years.
00:07:09.600 --> 00:07:12.959
Now, in terms of system reliability, one of their firewalls went down.
00:07:13.120 --> 00:07:16.000
This was in the public domain, so I'm okay to talk about it.
00:07:16.160 --> 00:07:22.480
One of their firewalls went down, and they did not have the wiring to be able to find out where exactly things broke.
00:07:22.720 --> 00:07:29.839
But what this led to is the medications that were supposed to reach terminal patients or whatever in the supply chain could not reach on time.
00:07:30.000 --> 00:07:31.439
Huge reputational damage.
00:07:31.680 --> 00:07:39.759
Also, the cost of downtime to a business, something like that, in their own words, is about a million dollars a minute if it's unplanned downtime.
00:07:40.000 --> 00:07:47.680
Now, what AI does, there's a traditional AI, is finds out what's broken where with the observability and the wiring and all of that.
00:07:47.839 --> 00:07:50.560
And then the next level is so can you trust that answer?
00:07:50.639 --> 00:07:52.800
Will you go and take that action with the human in the loop?
00:07:52.959 --> 00:07:59.519
Or the next level with agenc or autonomous operations is will you allow AI to go and fix that?
00:07:59.759 --> 00:08:12.879
So essentially it may be something as pedestrian as a firewall or a system going down, but the implications of the business outcome in terms of the loss damage, million dollars a minute, like I said, or you know, the reputational damage, that could be huge.
00:08:12.959 --> 00:08:20.160
So that's fundamentally how AI is changing, you know, how we look at system reliability and the business outcomes in the AI era.
00:08:20.399 --> 00:08:21.279
Does that make sense?
00:08:21.600 --> 00:08:22.240
Yeah, totally.
00:08:22.319 --> 00:08:24.399
It reminds me of an anecdote early in my career.
00:08:24.480 --> 00:08:32.159
I won't name names, but the customer was 1-800 flowers, you know, you call your for your flowers for your wife or mother, etc.
00:08:32.639 --> 00:08:40.240
And you know, our system, our network went down on Mother's Day, which, as you can imagine, not a good experience.
00:08:40.480 --> 00:08:46.639
Companies like that aren't in the flowers business, they're actually in the IT and phones and phone taking business.
00:08:46.879 --> 00:08:55.600
So, yeah, I understand that sort of reputational damage and how quickly things can get out of pocket, out of hand.
00:08:56.399 --> 00:09:09.279
Now, the exciting news for techies like me is that their research also showed 66% of enterprises are scaling AI, and we love to talk about which models and which AI and where it lives.
00:09:09.360 --> 00:09:11.600
And you know, technologists love this stuff.
00:09:11.919 --> 00:09:16.799
But what's not so exciting is only about a quarter have controls in place.
00:09:17.039 --> 00:09:21.679
So for the CFOs, the board, the CEO, that's kind of concerning.
00:09:21.840 --> 00:09:24.000
Are companies moving too fast?
00:10:09.970 --> 00:10:11.649
I I think so at some level.
00:10:11.730 --> 00:10:14.129
I think there are multiple, let's unpack that a little bit.
00:10:14.289 --> 00:10:16.370
So, first of all, thanks for sharing that story.
00:10:16.529 --> 00:10:17.730
That resonates very well.
00:10:17.889 --> 00:10:27.889
And secondly, the point that you just made the companies are moving too fast because people are embracing, people are adopting AI for the individual productivity.
00:10:28.129 --> 00:10:36.529
And the enterprises are lagging behind and putting the controls, the guardrails, and the HFS study that you alluded to also reflects that.
00:10:36.689 --> 00:10:40.929
In fact, I am reminded of another data point we saw from that study.
00:10:41.330 --> 00:10:45.730
If you, and and sorry, I'll go at a tangent for a moment and then bring it back.
00:10:45.889 --> 00:10:56.450
So the TCS AI framework has five levels of maturity, which is you know, you're using AI just as a tool or an assistant, or you know, something far more advanced, which we'll talk about later.
00:10:56.689 --> 00:11:08.210
But fundamentally, um, in terms of copilot, something that people are using, it was only 8% of the people surveyed said that they would not trust co-pilot results.
00:11:08.370 --> 00:11:11.090
92% are willing to trust co-pilot results.
00:11:11.330 --> 00:11:21.649
But the moment you take it to an autonomous domain, only 20% said that they will trust AI to take the actions, which is 10x.
00:11:22.289 --> 00:11:36.769
So it goes back to the point you're making because the foundational elements, the guardrails, the eval sets, uh, without getting too technical, essentially, and it's not just about the models, because there's been far too much focus on the models.
00:11:37.090 --> 00:11:42.210
You know, you're creating capacity with the models, but capacity does not equate to dependability.
00:11:43.730 --> 00:11:48.129
You have to think about the underlying data, how is the data structure in the enterprise?
00:11:48.210 --> 00:11:58.929
You have to think about the interfaces, you have to think about the integrations, and then you have to think about are the right guardrails put in place for AI to take action?
00:11:59.250 --> 00:12:01.570
And then is it safe?
00:12:01.889 --> 00:12:02.769
Is it secure?
00:12:02.929 --> 00:12:05.330
And all of the yada yada yada, I can go on and on.
00:12:05.490 --> 00:12:12.370
But fundamentally, what I'm trying to say is it also goes into the dimensions of accountability, transparency.
00:12:12.689 --> 00:12:16.850
You know, can you trace the actions down to why did it happen?
00:12:17.090 --> 00:12:30.210
And in a heavily regulated environment, in a lot of industries that we work in, it's very hard to make peace with something like that unless you put the right foundations, the right guardrails, the right enablers, or you need to have a human in the loop.
00:12:30.370 --> 00:12:35.730
That's that's only when people are willing to stand behind the outcomes that AI produces.
00:12:36.769 --> 00:12:37.250
Interesting.
00:12:37.330 --> 00:12:38.370
Yeah, it's funny.
00:12:38.529 --> 00:12:42.610
Life we runs on trust, but so does business and so does technology.
00:12:42.850 --> 00:12:45.730
And, you know, it's such an important question.
00:12:45.970 --> 00:12:57.250
In fact, many companies simply block access to employees when it comes to Chat GPP or Claude and simply block them on the firewall or what have you.
00:12:57.490 --> 00:13:03.250
So we don't necessarily even trust humans or employees all the time to use these tools.
00:13:03.330 --> 00:13:08.129
But when it comes to the agents themselves, which are providing tremendous value.
00:13:08.210 --> 00:13:12.850
I have two agents running for me right now doing work as I'm on this discussion.
00:13:13.009 --> 00:13:15.169
There's no question of the value it provides.
00:13:15.330 --> 00:13:23.090
But how do you decide how much freedom and action and decision making to give these agents?
00:13:24.210 --> 00:13:29.970
I think that's a very interesting question because trust is at the core of it.
00:13:30.289 --> 00:13:40.689
And you know, even when you put human in the loop, does the human have the right understanding, the right context, the right authority, the right data?
00:13:40.850 --> 00:13:45.169
I mean, all of those elements also come into play before you trust the outcome.
00:13:45.330 --> 00:13:49.970
But fundamentally, where I'm going with that even is trust is a human perception.
00:13:51.009 --> 00:14:07.090
I would take it a level beyond, especially from the TCS paradigm, from the engineering paradigm that we come with as a company, I would elevate it to a call of trustworthiness, which I think is an engineering responsibility.
00:14:07.570 --> 00:14:27.090
So what I'm what I'm trying to say with that is if you can engineer the systems, the agents in a way, with all the controls that we talked about, and the whole ecosystem in a way that the evidence and the consistency is always there for humans to stand behind.
00:14:27.409 --> 00:14:29.090
That's when you can trust the system.
00:14:29.409 --> 00:14:35.169
So to me, trust is not just a compliance paradigm, it's a business enabler.
00:14:35.490 --> 00:14:50.610
And to your point, you cannot unlock the power of these agents at the enterprise scale if that trust has inherently not been built in, those foundations have not been built in to unlock that value for the enterprise.
00:14:51.169 --> 00:14:52.289
Yeah, really interesting.
00:14:52.370 --> 00:14:53.970
And things fail, things break.
00:14:54.049 --> 00:14:58.210
We know this from IT, we know this from our personal experiences.
00:14:58.610 --> 00:15:09.169
But while I think the data that you commissioned showed most companies can detect AI failures, like 60% or so, only 23% can routinely stop them.
00:15:09.409 --> 00:15:10.289
You know, why is that?
00:15:10.370 --> 00:15:12.850
Why is stopping a bad AI decision so difficult?
00:15:13.009 --> 00:15:20.289
There is, I guess there's no break glass in case of emergency button or or you know, kill switch kind of thing.
00:15:20.450 --> 00:15:24.769
How do you stop these AIs uh going bad?
00:15:25.330 --> 00:15:30.450
So I think it's a maturity journey, and we're talking about the AI maturity framework of TCS.
00:15:30.529 --> 00:15:35.730
So let me kind of use that as an anchor and then address a question that you called out.
00:15:37.250 --> 00:15:43.090
At a very basic level, you're using AI either as a tool or as an assistant.
00:15:43.250 --> 00:15:45.649
I think co-pilot, say, for example.
00:15:45.970 --> 00:15:48.689
And there's always a human in the loop in that scenario.
00:15:48.850 --> 00:15:50.370
So that's level one and two.
00:15:50.529 --> 00:15:56.129
And if AI does something that you don't like, you don't act on it, as in essentially if you get an answer.
00:15:56.289 --> 00:16:13.570
But as you move to the next three levels of maturity, which is either it's a supervised agent, which again is significant human oversight, or even the next level could be autonomous workforce or the holy grail, which most enterprises have not gotten to, is the agentic enterprise.
00:16:13.809 --> 00:16:23.889
Now, in those two scenarios, levels four and five, if you've not put the right controls and right guardrails, the agent can be mischievous.
00:16:24.289 --> 00:16:27.809
I mean, the agent can do things to meet its goal.
00:16:27.889 --> 00:16:35.649
There was a salacious article about hugging face and open AI not very long back, say for example, because the agents were trying to meet their goals.
00:16:36.450 --> 00:16:40.289
And the right controls were not given to the agents.
00:16:41.330 --> 00:16:55.409
Again, it was a test environment, but basically, what I'm trying to say is so if you don't put the right checks and balances on when the agent can act, then the agent can run rogue and you can have all of these challenges.
00:16:55.490 --> 00:16:59.809
And that's the 23% who talk about they cannot control what's going to happen.
00:16:59.970 --> 00:17:06.529
So, long story short, and I'll make it real perhaps with uh with an example, with an industry example.
00:17:07.090 --> 00:17:26.210
But it's a maturity journey, and you have to curate that and you have to make sure the agent has the right context, the right guardrails, the right checks and balances, and the kill switch that you talked about should be involved or should happen if it tries to break the mold, essentially, because that requires policy as code.
00:17:26.529 --> 00:17:41.250
That requires, you know, and uh there are multiple technologies to do that, the hardnesses that we've built, say, for example, in these scenarios, or there are scenarios wherein you have an agent checking or an AI checking an AI's job before it's taken into mainstream.
00:17:41.330 --> 00:17:49.330
So, again, the ways to do that checks and balances, but if you don't do the foundations right, if you don't do the basics right, those kind of challenges are likely to happen.
00:17:49.490 --> 00:17:59.250
But one of the large banks we were working with, we were working with, uh, you know, this was about credit card disputes.
00:17:59.650 --> 00:18:02.610
They decided to go down the agentic path.
00:18:02.850 --> 00:18:16.050
Now, if somebody disputes a credit card transaction, it's very easy for the agent to say, okay, you know, what's what's the transaction, where it comes from, what's the data required for it, and so on.
00:18:16.210 --> 00:18:19.330
Now, the agent will tell you this is what went wrong.
00:18:19.490 --> 00:18:33.090
Now, but the question is, have you reached a place in your maturity of the foundation that you'll let the agent take the corrective action and you know pay back that money or do whatever needs to be done to resolve that dispute?
00:18:33.250 --> 00:18:37.890
To get to that level of autonomy, you essentially have to build all the right foundations.
00:18:39.090 --> 00:18:46.130
Yes, and speaking of foundations for AI autonomy and reliability, who who owns that inside a large organization?
00:18:46.290 --> 00:18:47.730
Is it the board?
00:18:47.890 --> 00:18:51.650
Is it customer service, the CIO, all the above?
00:18:51.810 --> 00:18:55.730
It's it must get pretty complicated in the large enterprises that you work with.
00:18:56.290 --> 00:18:57.730
It is a complex question.
00:18:57.810 --> 00:19:13.810
And in fact, uh, you know, I I want to steal a page from our CIO's playbook in TCS, because when we try to do this for ourselves as customer zero, the whole question was that can you stand behind an agentic decision as you're moving to an autonomous enterprise?
00:19:13.970 --> 00:19:21.890
So essentially, to curate these policies and guardrails, you need a small group of people representing the various disciplines that are involved, essentially.
00:19:21.970 --> 00:19:29.010
Like you said, it could be customer service, but you also need legal, you need risk, you need compliance, uh, you need finance.
00:19:29.170 --> 00:19:40.130
So a think tank, a small group of these people is formed, which defines the policies, and the policies should be the gateway before you allow the agents to act on these things.
00:19:40.370 --> 00:19:45.890
That's the way the hardness that I was talking about, or the other uh controls that I was talking about come into play.
00:19:46.050 --> 00:19:49.090
So have you wired it right with these people?
00:19:49.250 --> 00:19:52.450
And essentially the enterprise, the organization has to own that.
00:19:52.530 --> 00:20:04.290
You cannot come back and say, I'm a bank, I don't know why my agent did this, because some decisions are reversible, some decisions are inconsequential, but some nations could have huge repercussions.
00:20:04.530 --> 00:20:14.130
So, where do you give what level of autonomy to agents and have you essentially built those controls with the right team, with the right accountability, is what it boils down to.
00:20:14.930 --> 00:20:15.730
Fantastic.
00:20:15.890 --> 00:20:27.490
So I'm a huge proponent of agents, I use agents, and I've seen tremendous personal productivity gains from AI, you know, 10x, perhaps even for myself professionally, personally.
00:20:27.730 --> 00:20:39.810
And I think most companies are seeing productivity gains from AI, just if you if you see some of the data from your research and just anecdotally, but that hasn't necessarily turned into revenue growth.
00:20:39.970 --> 00:20:47.330
Uh, how do you close that gap and show the real ROI beyond the productivity benefits?
00:20:47.810 --> 00:20:48.690
Yeah, you're right.
00:20:48.770 --> 00:20:53.010
There's been a lot of focus on activity and individual productivity.
00:20:53.490 --> 00:21:04.210
And part of it is because that behavior was being rewarded, and it was a great way to bring about organizational change, people embracing AI ways of working.
00:21:04.370 --> 00:21:12.290
But when you have to take it to enterprise scale, you have to build the right foundations and you know you have to link it to the right metrics.
00:21:12.530 --> 00:21:21.890
So essentially, again, going back to that three-part framework that we talked about reliability, is the system doing what it's designed to do?
00:21:23.010 --> 00:21:30.610
Can I trust the answer that AI gives me, or can I, you know, trust AI to do that whole business outcome?
00:21:30.930 --> 00:21:33.410
That's essentially where you need to go with it.
00:21:33.490 --> 00:21:35.330
So let me make it real with an example.
00:21:35.490 --> 00:21:42.690
Again, this is a very large bank that we're working with, and one of their lines of business is commercial lending.
00:21:43.650 --> 00:21:57.410
Now, they would receive a lot of checks from the people that they had lent out these loans to, but because of inaccurate information or whatever, these were not applied and they were sitting in the suspense account.
00:21:57.490 --> 00:22:07.970
When they built this whole enterprise grade uh with the end in mind, the AI foundation, and then unlock that use case.
00:22:08.210 --> 00:22:15.490
What they were able to do is get 450 million uptake on the checks that were applied.
00:22:15.730 --> 00:22:20.210
They were able to get 82% reduction in the suspense accounts.
00:22:20.370 --> 00:22:24.290
Likewise, there's another example of revenue that you talked about.
00:22:24.530 --> 00:22:29.810
There's a distributor of electronic parts that's that's a large client of ours.
00:22:30.130 --> 00:22:36.210
And they would get all these components from OEMs, and there would be buyers on their website trying to buy these things.
00:22:36.290 --> 00:22:43.410
And you know, the OEMs keep I mean, thousands and thousands of SKUs, and the OEMs keep coming up with more and more specifications.
00:22:43.570 --> 00:22:47.570
People We're not able to buy, not able to determine what they needed.
00:22:47.890 --> 00:23:03.410
With agentic intervention and autonomous sales, 7% of their sales today, which is north of a billion dollars, and this again in the public domain is being done through agentic autonomous, you know, way in agentic autonomous way.
00:23:03.570 --> 00:23:09.490
So there's a billion dollar plus uptick in revenue for a large enterprise if you do things right.
00:23:09.650 --> 00:23:14.450
So bottom line is what are the KPIs, what are the metrics you're focused on?
00:23:14.610 --> 00:23:20.770
Are those business outcomes or are those, you know, at the enterprise level, or are those individual productivity?
00:23:20.850 --> 00:23:26.850
Can I get my code 10 times faster, or can I, you know, do my meetings quicker, or what have you.
00:23:27.010 --> 00:23:42.610
So you you really have to have the big picture view and the C-suite, the board has to define that this is where we want to go, this is where we want to move the needle on these metrics, could be revenue, could be habitat, could be experience in NPS or whatever quantifiable form you measure, and so on.
00:23:42.690 --> 00:23:45.170
So that's essentially the secret sauce of how to do it.
00:23:45.250 --> 00:23:49.890
But you have to start from the right and then move to the left to how do you go about doing this?
00:23:50.530 --> 00:23:52.130
Yeah, brilliant insight.
00:23:52.290 --> 00:23:56.450
And let's talk about getting to a truly autonomous and AI-driven enterprise.
00:23:56.690 --> 00:24:07.650
You've actually at TCS, I've I've been reading, have been doing a lot of that work internally with the world's biggest AI hackathon and other initiatives to transform yourselves.
00:24:07.810 --> 00:24:09.250
Maybe talk a little bit about that.
00:24:09.330 --> 00:24:18.930
And in the industry, you know, what has to happen before businesses can get to that autonomous enterprise vision that we're so excited about?
00:24:19.650 --> 00:24:32.930
Yeah, so we're very proud of what we have achieved as TCS because uh, you know, we we when we embrace this vision of being the largest AI uh services company, we had to be customer zero for ourselves.
00:24:33.010 --> 00:24:36.610
And that was the first foundational pillar that we put for this.
00:24:36.770 --> 00:24:49.410
And, you know, for a lot of our processes, finance, HR, I mean, everything, uh including the work we deliver for our customers, we have been able to get phenomenal autonomous results essentially using AI.
00:24:49.890 --> 00:24:59.650
Now, while while that's translating into a lot of things, uh, sorry, if you could take me back to your question beyond TCS, you also wanted to know.
00:24:59.810 --> 00:25:02.370
Sorry, if you could repeat the question for me, I won't I lost time.
00:25:02.770 --> 00:25:02.930
Sure.
00:25:03.090 --> 00:25:03.490
No worry.
00:25:03.570 --> 00:25:06.130
Now I'll edit this this part out just for James.
00:25:06.210 --> 00:25:08.210
So I'll start over with the question again.
00:25:08.290 --> 00:25:10.930
So, and we'll uh three, two.
00:25:11.650 --> 00:25:21.410
So, you know, looking ahead of you know, 12, 24 a month, companies are trying to get to this truly autonomous enterprise, AI-driven enterprise.
00:25:21.490 --> 00:25:22.930
And it's no small feat.
00:25:23.010 --> 00:25:37.250
I know even you at TCS with five, six hundred thousand employees, have managed to really reinvent yourself towards an AI-driven workforce with like the world's biggest hackathon and other initiatives like that.
00:25:37.650 --> 00:25:48.370
So, talk about a little bit about yourself and your transformation, but also you know what needs to happen for your customers for businesses can trust AI to run major operations.
00:25:49.090 --> 00:25:51.410
Yeah, so you're absolutely right.
00:25:51.570 --> 00:25:58.610
Very proud of what we've been able to achieve at TCS and a lot of you know our processes run in an autonomous fashion, and we're getting there.
00:25:58.850 --> 00:26:05.570
I think for large enterprises to be able to do that, very simplistically, there are three sets of things.
00:26:05.890 --> 00:26:09.330
I would say tool set, skill set, and mindset.
00:26:10.610 --> 00:26:12.130
You can provide all the tools.
00:26:12.290 --> 00:26:17.650
Obviously, we talked about the foundation and all of those things, and you can, you know, engineer it right.
00:26:17.730 --> 00:26:20.610
So that is the first foundational block that needs to happen.
00:26:21.090 --> 00:26:25.650
The second is a talent, because it's not just technology talent, i.e., the skill set.
00:26:25.730 --> 00:26:28.930
It's not just a technology talent that is leveraging that.
00:26:29.490 --> 00:26:37.170
You need even the business folks to be equally comfortable with AI to move to that autonomous phase.
00:26:37.330 --> 00:26:48.850
And finally, the mindset, because the biggest change that AI is having is it's not sprinkling the AI pixie dust on your workflows or processes, it's reimagining things and how they can be done all over again.
00:26:49.010 --> 00:26:53.810
And that's where the ways of working, the mindset, the whole operating model needs to be looked at.
00:26:53.970 --> 00:27:01.890
So essentially, organizations are taking a three, uh maybe a couple dimensions, I would say, in which they're moving towards autonomy.
00:27:02.130 --> 00:27:04.050
One is persona-based.
00:27:04.290 --> 00:27:08.210
Which are the personas that would benefit the most from full autonomy?
00:27:08.450 --> 00:27:15.650
And they would start out with that, build it, and they will come kind of a model, or enable them with the right tools and the skill set and the mindset.
00:27:15.810 --> 00:27:17.970
The other could be function-based.
00:27:18.130 --> 00:27:23.330
Okay, AP or AR or whatever could be a function, or it could be even customer service.
00:27:23.490 --> 00:27:25.890
Is that something that I want to move towards autonomy?
00:27:26.050 --> 00:27:36.290
So again, organizations are taking different approaches, but bottom line is you need the tool set, the skill set and the mindset, and it's a journey to embrace full autonomy.
00:27:36.450 --> 00:27:37.970
And there are some journey.
00:27:38.450 --> 00:27:47.410
Sorry, and there are some functions like compliance and regulation, wherein you need to be a little bit more thoughtful about it, a little bit more deliberate about it when you go to that stage.
00:27:47.970 --> 00:27:58.690
So true, and it is a journey, and uh, I'm learning so much from these conversations, and uh, congratulations on the journey so far with a long and interesting road ahead.
00:27:59.010 --> 00:28:02.130
Thanks very much for joining and sharing your insights.
00:28:02.450 --> 00:28:03.890
Thank you very much for having me.
00:28:03.970 --> 00:28:05.090
The pleasure is all ours.
00:28:05.250 --> 00:28:06.210
Thank you, everyone.
00:28:06.530 --> 00:28:08.610
Thank you, and thanks everyone for listening and watching.
00:28:08.690 --> 00:28:14.690
Also, check out our TV show on Bloomberg Television and Fox Business at techmact.tv.
00:28:14.850 --> 00:28:15.730
Thanks, Rushant.
00:28:15.890 --> 00:28:16.450
Thanks, everyone.
00:28:16.770 --> 00:28:17.170
Thank you.
00:28:17.330 --> 00:28:18.290
Bye bye.