WEBVTT
00:00:03.279 --> 00:00:07.360
So most organizations are asking how to integrate artificial intelligence.
00:00:07.519 --> 00:00:11.359
The better question, the harder one, is when it actually makes sense to.
00:00:11.519 --> 00:00:14.160
That tension lives on the center of today's conversation.
00:00:14.320 --> 00:00:27.120
My guest has navigating from the inside one of the most complex product environments in B2B systems where artificial intelligence had to survive contact with real users, real teams, and real operation constraints.
00:00:27.280 --> 00:00:34.799
Michelle Brito is a senior product designer at Volkswagen Digital Solutions, working at the intersection of product, strategy, and emerging technology.
00:00:34.960 --> 00:00:40.479
Always grounded in the user research and the kind of operational reality that doesn't show up in the pitch deck.
00:00:40.560 --> 00:00:43.920
So without further ado, welcome to UMANX Intelligent, Michelle.
00:00:44.079 --> 00:00:45.920
Ready, really, really glad that you're here.
00:00:46.240 --> 00:00:48.079
Oh, what amazing introduction.
00:00:48.320 --> 00:00:49.119
Thank you very much.
00:00:49.359 --> 00:00:49.759
Thank you.
00:00:49.840 --> 00:00:52.320
And I'm very, very, very happy to be here with you.
00:00:52.479 --> 00:00:53.200
Thanks for having me.
00:00:53.439 --> 00:00:55.359
Of course, I am very happy that you're here.
00:00:55.439 --> 00:00:57.359
I'm very excited to see what you will share.
00:00:57.439 --> 00:01:08.000
And let's go right away and talk about the that artificial intelligence is basically dominating almost every strategic conversation right now, but often before the fundamentals are even in place.
00:01:08.319 --> 00:01:12.239
So I want to start start from a slightly uncomfortable place.
00:01:12.400 --> 00:01:18.239
So, Michelle, is AI actually the right starting point for most organizations today?
00:01:18.560 --> 00:01:29.120
I think instead of thinking about what kind of technology we have to use to solve the problem, we have to go back and understand the user's problems.
00:01:29.280 --> 00:01:35.920
I think it's very important to ask answer some questions like is your user familiar with AI solution?
00:01:36.079 --> 00:01:39.359
Is your business aware of all the AI compliance?
00:01:39.519 --> 00:01:45.040
And also as a designers, we need to think like, is this solution solving a user problem?
00:01:45.359 --> 00:01:51.680
And this is another question that should be discussed and analyzing before jumping to the final solution.
00:01:51.920 --> 00:02:03.200
Because I feel like, as you mentioned now, it's pretty common to for for the companies uh choose first the technology instead of understanding the problems, the user problems.
00:02:03.439 --> 00:02:12.960
And let's grab that and make it even more concrete because regardless of readiness, many companies are already trying to implement artificial intelligence.
00:02:13.199 --> 00:02:23.599
What are or what do you see as the most common situations where companies try to introduce artificial intelligence when what they actually need is something much more basic?
00:02:23.840 --> 00:02:24.080
Yeah.
00:02:24.319 --> 00:02:28.960
To be honest, I feel like when they want to make things simpler and faster.
00:02:29.199 --> 00:02:41.919
For example, like uh customer support, they always want to implement an AI assistant, which makes sense because sometimes the interactions are like quicker and we can easily solve uh users' problems.
00:02:42.080 --> 00:02:47.520
And also when the problem is simple and rule-based, it's very common.
00:02:47.680 --> 00:02:49.840
They always want you to implement AI.
00:02:50.000 --> 00:03:00.000
And actually, AI it's in fact is very it's very, very powerful, but it's also very expensive and predictable, and also harder to design well.
00:03:00.240 --> 00:03:21.840
So I feel like companies should avoid uh think about implementing AI when things are already working well, and in other words, like if your product needs to have flows that it's predictable, I think implementing AI could be very, very uh difficult and perhaps it's isn't a good alternative.
00:03:22.080 --> 00:03:33.360
So in those cases, AI is almost acting as a shortcut for deeper operational or even structural issues, because that kind of suggests that the problem isn't just technical, it's more systemic.
00:03:33.520 --> 00:03:35.520
My next question comes hand in hand with that.
00:03:35.680 --> 00:03:42.240
Why do organizations tend to skip these foundational steps and jump straight to artificial intelligence?
00:03:42.639 --> 00:03:52.560
I'd say because they think they have this idea of integration and solving all the users' problems in only one click or one prompt.
00:03:52.719 --> 00:04:06.960
I think at least in my experience, sometimes my stakeholders they come with this idea like, okay, so let's let's integrate everything, let's solve users' problems, that'll be so easy, will be so like uh simple, and also because the market is very, very competitive.
00:04:07.199 --> 00:04:10.400
So you can you can see like AI everywhere.
00:04:10.719 --> 00:04:19.519
And if your new boring uh uh feature doesn't have AI, it seems like your product is out of dated.
00:04:19.600 --> 00:04:27.920
So um I think also because they're assuming that everyone is ready for interacting with an AI feature, probably.
00:04:28.240 --> 00:04:32.879
And it kind of happens in the world because we are we feel very pressured, right?
00:04:33.040 --> 00:04:35.920
And this leads to unintended consequences.
00:04:36.160 --> 00:04:47.519
Have you ever seen situations or heard of situations where introducing artificial intelligence too early created more errors or confusion or even operational risk rather than value?
00:04:47.839 --> 00:04:59.839
Yeah, to be honest, lately I've seen situations where users don't trust the results or when they don't feel like part of the mission uh of the integration with the solution, with the final solutions.
00:05:00.079 --> 00:05:06.560
Sometimes they don't feel like they are kind of doing their jobs, especially when it's an internal tool, for example.
00:05:06.800 --> 00:05:13.360
I think the best outcomes come from like a human AI collaboration, not a substitution.
00:05:13.600 --> 00:05:20.720
How do you usually solve these problems with your stakeholders, with your clients, to make them feel more at ease?
00:05:20.879 --> 00:05:29.040
Because if you introduce these are solutions to your product, there's always going to be these problems because there's no way around it.
00:05:29.199 --> 00:05:33.519
So, how do you usually introduce or solve these situations?
00:05:33.839 --> 00:05:39.680
Well, I feel like uh it's very hard to change people's minds when they have a solution in their minds.
00:05:40.240 --> 00:05:54.720
So for me, now my main challenge is like explain for my stakeholders that it's very important to do a deep research instead of like solve the users' problems with new AI features because we already have a lot of new AI features.
00:05:54.879 --> 00:05:58.000
So I think it's boring, but it's very basic.
00:05:58.160 --> 00:05:59.199
We have to research.
00:05:59.360 --> 00:06:02.879
So we have to investigate, we need to understand the users' problems.
00:06:03.040 --> 00:06:07.199
It's basic, it's old, but in the end, it's the best solutions.
00:06:07.439 --> 00:06:18.000
So if many organizations are getting this wrong, for me, at least, the real question would be if how do you tell when AI actually makes sense versus when it doesn't?
00:06:18.319 --> 00:06:27.279
So for me, when it's something that works, like works very well, I think it doesn't make sense to implement something completely new and complex.
00:06:27.439 --> 00:06:28.959
So for me, it's a big no.
00:06:29.120 --> 00:06:42.879
But when we have the opportunity to introduce, uh, for example, like some very specific AI interactions with the features and the users in a very well consolidated product, I think in this case it makes sense.
00:06:43.199 --> 00:07:00.720
But I don't like when the the stakeholders and the business they want to like implement a 100% AI product, let's say, because sometimes it's very, very risky for the users to interact with this completely new product.
00:07:00.879 --> 00:07:06.639
And also in terms of like in my case, because I have most of my products, they are like internal products.
00:07:06.800 --> 00:07:10.319
Sometimes the users they don't trust, they don't know how to interact.
00:07:10.480 --> 00:07:15.360
So they don't use our products, they continue like doing the same old tests.
00:07:15.519 --> 00:07:19.120
So I think it's very tricky because we are still going and like it.
00:07:19.279 --> 00:07:30.399
This is also my challenge, like to understand the process and to evaluate when and how to apply and to interact for the users to interact with this AI uh features.
00:07:30.800 --> 00:07:35.360
Do you have any tips for people that are in the similar situation as you?
00:07:35.680 --> 00:07:40.639
Because of course people are going to have problems, but if they are not using, how do you solve it?
00:07:40.879 --> 00:07:53.279
Yeah, so I try to ask some questions for myself and also with my PMs and with my business team is like, is AI actually better than simple automation?
00:07:53.519 --> 00:07:59.920
Or something like what specific workflow is uh slow, expensive, or error prone?
00:08:00.079 --> 00:08:03.600
So we should focus on this kind of flows.
00:08:03.920 --> 00:08:12.079
And probably the tasks with the high volume and low tolerance for delay, they are the best for implementing AI.
00:08:12.240 --> 00:08:24.720
So I think we have to be aware of like for introducing AI in your product, you need to have time, you need to have a good research, you need to work close with metrics.
00:08:25.040 --> 00:08:40.960
And how do you recognize when a company is adopting adopting artificial intelligence because it's a strategic necessity versus when it's mostly driven by narrative uh pressure or fear of being left behind?
00:08:41.200 --> 00:08:52.320
Well, I think as I mentioned before, it's very clear when, for example, us as a designer, when we see like a very complex flow that could be just three three three steps.
00:08:52.480 --> 00:08:54.159
A flow with three steps, for example.
00:08:54.320 --> 00:09:00.879
Um, a good example is like sometimes I have to send money to Brazil and I use uh Brazilian bank.
00:09:01.120 --> 00:09:02.879
So basically it's a PIX.
00:09:03.120 --> 00:09:04.720
I don't know if you heard about PIX.
00:09:04.960 --> 00:09:14.960
So we have to press, at least I I used to have to press two buttons and only add the name or the number and the value.
00:09:15.200 --> 00:09:18.960
But now they are introducing a set uh AI assistant.
00:09:19.200 --> 00:09:28.720
So instead of just adding some information, some numbers, I have to interact with uh AI uh assistant, um, saying like the name, the number.
00:09:28.960 --> 00:09:37.600
For me, to be honest, it was way, way uh easier to interact only with my inputs than with an AI assistant.
00:09:37.759 --> 00:09:49.360
So I feel like they are trying to be innovative instead of being like uh um introducing a helpful uh help um flows for the user.
00:09:49.840 --> 00:09:59.519
So I feel like when it's something that is like 10 times uh better, I think it makes sense to to add this AI interaction in your product.
00:09:59.679 --> 00:10:04.320
But sometimes when it makes the user feel like more confused, doesn't make sense at all.
00:10:04.720 --> 00:10:11.440
Are there any specific signals or patterns that immediately stand out uh to you?
00:10:11.759 --> 00:10:28.399
Well, in terms of it stand out for me when we are like gathering and exploring a possible AI solution, it's like when, as I mentioned before, when the technology comes first, before the to understanding like the users' uh problems.
00:10:28.559 --> 00:10:38.399
So I think it's very common because as I mentioned, if you don't have AI now, nowadays, if you don't have an AI feature, you are like outdated.
00:10:38.639 --> 00:10:48.000
So for me, and learning from my past experience as well, because I I also implemented an AI assistant one year and a half ago.
00:10:48.480 --> 00:10:56.720
And so basically, we are not solving the users' problems, we are not understanding the users' problems.
00:10:56.879 --> 00:11:04.799
So I think we sometimes it's very boring to understand and to collect all this information, unfortunately.
00:11:05.120 --> 00:11:12.080
But if you need to make sure that this uh solution will be 10 times better than you have.
00:11:12.240 --> 00:11:20.639
So, for example, if your business team comes with something ready, be careful because it could be like a problem in the problem in the future.
00:11:20.879 --> 00:11:44.240
So sometimes we need to just like do babe steps and try to understand how if your team are in the same page in terms of outcomes and do like very, very small changes, very, very small interactions, for example, with uh users and AI, then to have like a very big and complex feature with AI.
00:11:44.559 --> 00:11:44.879
Yeah.
00:11:45.039 --> 00:11:52.480
And I imagine that these distinctions or these situations become even more critical in certain environments.
00:11:53.120 --> 00:12:00.799
I guess my next question to you is how should organizations think about introducing artificial intelligence responsibly?
00:12:00.960 --> 00:12:08.399
You already kind of talked a little bit about it, but specifically right now when they operate in regulated environments?
00:12:08.960 --> 00:12:12.960
I would say they should start with risky classification.
00:12:13.120 --> 00:12:32.399
At least I was trying to do this like to create a very list with the risk classification and not with the the use case, because we have to be aware, like and also our company and our team, they need to be aware about the all the risks that is implementing AI.
00:12:32.799 --> 00:12:49.120
So and also uh answering some questions like what's the impact of being wrong here, for example, and also who is affected if something happens, for example, like financially, legally, or physically.
00:12:49.360 --> 00:13:15.679
So, and also being very, very intentional about data boundaries because it's something that's for example, for AI assistants, we need to be aware about data boundaries, and last but not least, I think we should to be aware about how in the end the users like the human can reveal the final results, how the users can interact and reveal the final results.
00:13:16.399 --> 00:13:16.720
Okay.
00:13:17.039 --> 00:13:29.200
I guess this brings us to what happens internally because even with the right intent, execution, and also talking with stakeholders depends also on ownership, right?
00:13:29.360 --> 00:13:37.039
So inside of an organization, who should actually be accountable for how artificial intelligence systems are used?
00:13:37.440 --> 00:13:49.919
I think in the end, the responsibility should be distributed because once everyone knows what we are doing, like as a team, I think the responsibility should be very well distributed.
00:13:50.159 --> 00:14:02.159
And I feel like sometimes that tension is exactly where many organizations fail because the outcomes sometimes are so unpredictable and it's hard to prevent errors.
00:14:02.320 --> 00:14:05.519
So, as I mentioned, it's very, very hard to prevent errors.
00:14:05.679 --> 00:14:14.799
So, my suggestion is like maybe you can try to create a racy model with the the roles and also some responsibilities.
00:14:14.960 --> 00:14:16.159
I think it could work.
00:14:16.480 --> 00:14:24.240
Should that responsibility sit in like a specific function, or does it need to be shared across leadership and teams?
00:14:24.480 --> 00:14:25.840
What is your opinion on this?
00:14:26.159 --> 00:14:32.720
To be honest, I think it should be shared because everyone needs to be aware about the impact of errors.
00:14:32.960 --> 00:14:40.240
And once they are aware, of course, the system, the tool, like the product will have some specific information for the user.
00:14:40.399 --> 00:14:49.200
So of course, the leaders they they will be like the most likely the responsibles, but I think it should be shared also with the team.
00:14:49.519 --> 00:14:56.399
And basically accountability is one thing, but how it shows up on the day-to-day is something completely different.
00:14:56.559 --> 00:15:08.000
And I think it's also important for us to introduce this topic to the conversation because what does a healthy and responsible introduction of artificial intelligence inside teams actually look like?
00:15:08.159 --> 00:15:10.879
Because I don't think many people are talking about this.
00:15:10.960 --> 00:15:18.080
I've been hearing a lot of people saying that their like leadership is just saying, you have to have AI, you have to use AI.
00:15:18.240 --> 00:15:20.639
Tell me basically tell me where to spend my money.
00:15:20.799 --> 00:15:22.559
But what is your opinion on this?
00:15:22.799 --> 00:15:25.600
What would you say it's the healthy and responsible way?
00:15:25.840 --> 00:15:37.840
Yeah, I think a healthy introduction uh inside of the teams doesn't have to be like a big lounge or with uh like something like a bold AI transformation announcement, for example.
00:15:38.000 --> 00:15:39.360
I think this is very dangerous.
00:15:39.519 --> 00:15:49.120
It should be like very gradual, like like full of sharing experiences, testing fails, and of course, like sharing also the fails because it's very important.
00:15:49.279 --> 00:15:59.759
The teams that do this well move carefully, so slowly at the beginning, and so they can move like faster and later without breaking trusts, I think.
00:16:00.080 --> 00:16:00.320
Okay.
00:16:00.480 --> 00:16:04.639
So let's assume that an organization wants to move in that direction.
00:16:04.879 --> 00:16:13.120
Where should organizations start if they need to start slow, like you're saying, if they want to do this properly, what would you say they need to introduce first?
00:16:13.600 --> 00:16:18.639
I think first of all, they need to put everyone everyone's expectation at the same page about it.
00:16:18.799 --> 00:16:26.639
When I say everyone, it's everyone, like the the the stakeholders, the business team, and everyone same, same at the same page.
00:16:26.720 --> 00:16:31.200
So also like start by identifying the right kind of problem.
00:16:31.519 --> 00:16:35.039
This is very hard to identify the right kind of problem.
00:16:35.279 --> 00:16:49.200
It might be like the high volume or the time consuming, and or also like the problem that involves like a messy or unstructured data, probably, because they are the most likely to be solved by AI.
00:16:49.440 --> 00:16:53.440
So, what's the most overlooked first step that tends to get ignored?
00:16:53.759 --> 00:17:01.279
I think we ignore the most basic, that it's this AI feature will solve the user's problems.
00:17:01.519 --> 00:17:23.440
So I think we are always jumping to the final solution, and also I feel like we are ignoring all this comp those like compliance uh risks that could be like very problematic in the future, and we are ignoring the data, the connection with all the data, and we have good data to connect in our system.
00:17:23.759 --> 00:17:29.519
So I feel like we are like constantly ignoring this very important uh aspects.
00:17:29.759 --> 00:17:36.640
What enables um that kind of decision making specifically under external pressures?
00:17:36.880 --> 00:17:54.160
Well, I I think if you have a good team, like uh very aware and very well trained about the risks, I think it's way, way more easy to like brainstorm together and find the right uh the right solutions.
00:17:54.480 --> 00:18:13.200
Because when you are under pressure, like as a designer, and you have to design the best flow, it could be very, very hard for you because you don't have like the right tools, the right people, and also a team to help you to create this structure and also to deliver the best solutions for the user.
00:18:13.279 --> 00:18:28.160
So you have to receive this support from your developers, your product owner, your product manager, um to understand the user's problem and also to be able to deliver the best AI solution.
00:18:28.400 --> 00:18:41.599
Do you think that one of the biggest misunderstandings right now is that artificial intelligence adoption is about adding this kind of complexity that you were talking about when in reality it often requires simplifying systems first?
00:18:41.839 --> 00:18:43.200
Definitely, yes.
00:18:44.079 --> 00:18:55.599
And and I think like AI only works well when the product goals are clear, structured, and are very, very intentional.
00:18:55.759 --> 00:18:59.440
And otherwise you are just like it's scaling the confusion.
00:18:59.680 --> 00:19:13.440
So if we zoom out from the from everything we've discussed, what advice would you give teams who want to integrate artificial intelligence responsibly without falling into the trap of premature adoption?
00:19:13.680 --> 00:19:18.960
Yeah, I think the biggest trap is like the sentence we need to use AI.
00:19:19.359 --> 00:19:25.440
I feel like uh we instead of we should like change for we have a problem worth solving.
00:19:25.680 --> 00:19:36.240
I think we have to to change the sentence, the the affirmation, and also like answering some questions like is AI actually better than simpler automation?
00:19:36.480 --> 00:19:41.839
Or what specific workflow is slow, expensive, or error prone?
00:19:42.000 --> 00:19:50.400
I also like um we should focus on um tests with high volume and low tolerance delay.
00:19:50.480 --> 00:19:53.680
Well, it's like it's gold if you find this this test.
00:19:53.920 --> 00:20:00.160
And as I mentioned before, uh we have to define a rule that is 10 times or nothing.
00:20:00.480 --> 00:20:17.200
Because if your um AI solution is not like 10 times better than what you have, or 10 times better than something like very simple automation, I think it doesn't make sense to implement AI in your project.
00:20:17.599 --> 00:20:18.799
100% agree.
00:20:19.039 --> 00:20:20.160
Thank you for sharing.
00:20:20.319 --> 00:20:26.960
And before we finish our conversation, I would like to ask if there's anything else you would like to share with the audience.
00:20:27.279 --> 00:20:35.119
Yeah, I think you know it's very it's very like common uh have AI in our day-to-day conversations.
00:20:35.279 --> 00:20:39.039
And for some people, it's like, okay, I can't like hear about AI anymore.
00:20:39.119 --> 00:20:43.279
From other people, it's like, okay, I really want to implement AI in my products.
00:20:43.440 --> 00:20:59.519
But I I think we have to understand that AI it's something that we are still learning and we are still trying to understand how the users will interact, and probably they will love, probably in five years, everything will be about AI.
00:20:59.599 --> 00:21:02.160
And also it could be like, okay, it's just a hype.
00:21:02.400 --> 00:21:10.799
Let's like do babe steps and try to understand what is the best, what is simple, and what makes people solve people's problems.
00:21:10.960 --> 00:21:12.559
And well, I think that's it.
00:21:12.799 --> 00:21:13.920
Thank you so much, Michelle.
00:21:14.000 --> 00:21:15.920
This has been a really great conversation.
00:21:16.079 --> 00:21:36.799
And I guess what it stands out to me is that responsible artificial responsible adoption of artificial intelligence isn't really about starting with the technology, it's about the user flows, the understanding of the systems, the people and the timing, like you were saying, and maybe even most importantly, having the discipline to build the right foundations before adding intelligence on top.
00:21:36.960 --> 00:21:38.559
Thank you so much for your perspective.
00:21:38.640 --> 00:21:46.799
And to everyone listening, if there's one take, it's this AI doesn't fix uh broken systems, it actually amplifies them.
00:21:46.960 --> 00:21:52.000
So before moving forward, make sure what you're building one is actually ready for it.
00:21:52.240 --> 00:21:53.440
Thank you again, Michelle.
00:21:53.519 --> 00:21:54.559
It has been a pleasure.
00:21:54.799 --> 00:21:55.920
Thank you very much.
00:00:03.279 --> 00:00:07.360
So most organizations are asking how to integrate artificial intelligence.
00:00:07.519 --> 00:00:11.359
The better question, the harder one, is when it actually makes sense to.
00:00:11.519 --> 00:00:14.160
That tension lives on the center of today's conversation.
00:00:14.320 --> 00:00:27.120
My guest has navigating from the inside one of the most complex product environments in B2B systems where artificial intelligence had to survive contact with real users, real teams, and real operation constraints.
00:00:27.280 --> 00:00:34.799
Michelle Brito is a senior product designer at Volkswagen Digital Solutions, working at the intersection of product, strategy, and emerging technology.
00:00:34.960 --> 00:00:40.479
Always grounded in the user research and the kind of operational reality that doesn't show up in the pitch deck.
00:00:40.560 --> 00:00:43.920
So without further ado, welcome to UMANX Intelligent, Michelle.
00:00:44.079 --> 00:00:45.920
Ready, really, really glad that you're here.
00:00:46.240 --> 00:00:48.079
Oh, what amazing introduction.
00:00:48.320 --> 00:00:49.119
Thank you very much.
00:00:49.359 --> 00:00:49.759
Thank you.
00:00:49.840 --> 00:00:52.320
And I'm very, very, very happy to be here with you.
00:00:52.479 --> 00:00:53.200
Thanks for having me.
00:00:53.439 --> 00:00:55.359
Of course, I am very happy that you're here.
00:00:55.439 --> 00:00:57.359
I'm very excited to see what you will share.
00:00:57.439 --> 00:01:08.000
And let's go right away and talk about the that artificial intelligence is basically dominating almost every strategic conversation right now, but often before the fundamentals are even in place.
00:01:08.319 --> 00:01:12.239
So I want to start start from a slightly uncomfortable place.
00:01:12.400 --> 00:01:18.239
So, Michelle, is AI actually the right starting point for most organizations today?
00:01:18.560 --> 00:01:29.120
I think instead of thinking about what kind of technology we have to use to solve the problem, we have to go back and understand the user's problems.
00:01:29.280 --> 00:01:35.920
I think it's very important to ask answer some questions like is your user familiar with AI solution?
00:01:36.079 --> 00:01:39.359
Is your business aware of all the AI compliance?
00:01:39.519 --> 00:01:45.040
And also as a designers, we need to think like, is this solution solving a user problem?
00:01:45.359 --> 00:01:51.680
And this is another question that should be discussed and analyzing before jumping to the final solution.
00:01:51.920 --> 00:02:03.200
Because I feel like, as you mentioned now, it's pretty common to for for the companies uh choose first the technology instead of understanding the problems, the user problems.
00:02:03.439 --> 00:02:12.960
And let's grab that and make it even more concrete because regardless of readiness, many companies are already trying to implement artificial intelligence.
00:02:13.199 --> 00:02:23.599
What are or what do you see as the most common situations where companies try to introduce artificial intelligence when what they actually need is something much more basic?
00:02:23.840 --> 00:02:24.080
Yeah.
00:02:24.319 --> 00:02:28.960
To be honest, I feel like when they want to make things simpler and faster.
00:02:29.199 --> 00:02:41.919
For example, like uh customer support, they always want to implement an AI assistant, which makes sense because sometimes the interactions are like quicker and we can easily solve uh users' problems.
00:02:42.080 --> 00:02:47.520
And also when the problem is simple and rule-based, it's very common.
00:02:47.680 --> 00:02:49.840
They always want you to implement AI.
00:02:50.000 --> 00:03:00.000
And actually, AI it's in fact is very it's very, very powerful, but it's also very expensive and predictable, and also harder to design well.
00:03:00.240 --> 00:03:21.840
So I feel like companies should avoid uh think about implementing AI when things are already working well, and in other words, like if your product needs to have flows that it's predictable, I think implementing AI could be very, very uh difficult and perhaps it's isn't a good alternative.
00:03:22.080 --> 00:03:33.360
So in those cases, AI is almost acting as a shortcut for deeper operational or even structural issues, because that kind of suggests that the problem isn't just technical, it's more systemic.
00:03:33.520 --> 00:03:35.520
My next question comes hand in hand with that.
00:03:35.680 --> 00:03:42.240
Why do organizations tend to skip these foundational steps and jump straight to artificial intelligence?
00:03:42.639 --> 00:03:52.560
I'd say because they think they have this idea of integration and solving all the users' problems in only one click or one prompt.
00:03:52.719 --> 00:04:06.960
I think at least in my experience, sometimes my stakeholders they come with this idea like, okay, so let's let's integrate everything, let's solve users' problems, that'll be so easy, will be so like uh simple, and also because the market is very, very competitive.
00:04:07.199 --> 00:04:10.400
So you can you can see like AI everywhere.
00:04:10.719 --> 00:04:19.519
And if your new boring uh uh feature doesn't have AI, it seems like your product is out of dated.
00:04:19.600 --> 00:04:27.920
So um I think also because they're assuming that everyone is ready for interacting with an AI feature, probably.
00:04:28.240 --> 00:04:32.879
And it kind of happens in the world because we are we feel very pressured, right?
00:04:33.040 --> 00:04:35.920
And this leads to unintended consequences.
00:04:36.160 --> 00:04:47.519
Have you ever seen situations or heard of situations where introducing artificial intelligence too early created more errors or confusion or even operational risk rather than value?
00:04:47.839 --> 00:04:59.839
Yeah, to be honest, lately I've seen situations where users don't trust the results or when they don't feel like part of the mission uh of the integration with the solution, with the final solutions.
00:05:00.079 --> 00:05:06.560
Sometimes they don't feel like they are kind of doing their jobs, especially when it's an internal tool, for example.
00:05:06.800 --> 00:05:13.360
I think the best outcomes come from like a human AI collaboration, not a substitution.
00:05:13.600 --> 00:05:20.720
How do you usually solve these problems with your stakeholders, with your clients, to make them feel more at ease?
00:05:20.879 --> 00:05:29.040
Because if you introduce these are solutions to your product, there's always going to be these problems because there's no way around it.
00:05:29.199 --> 00:05:33.519
So, how do you usually introduce or solve these situations?
00:05:33.839 --> 00:05:39.680
Well, I feel like uh it's very hard to change people's minds when they have a solution in their minds.
00:05:40.240 --> 00:05:54.720
So for me, now my main challenge is like explain for my stakeholders that it's very important to do a deep research instead of like solve the users' problems with new AI features because we already have a lot of new AI features.
00:05:54.879 --> 00:05:58.000
So I think it's boring, but it's very basic.
00:05:58.160 --> 00:05:59.199
We have to research.
00:05:59.360 --> 00:06:02.879
So we have to investigate, we need to understand the users' problems.
00:06:03.040 --> 00:06:07.199
It's basic, it's old, but in the end, it's the best solutions.
00:06:07.439 --> 00:06:18.000
So if many organizations are getting this wrong, for me, at least, the real question would be if how do you tell when AI actually makes sense versus when it doesn't?
00:06:18.319 --> 00:06:27.279
So for me, when it's something that works, like works very well, I think it doesn't make sense to implement something completely new and complex.
00:06:27.439 --> 00:06:28.959
So for me, it's a big no.
00:06:29.120 --> 00:06:42.879
But when we have the opportunity to introduce, uh, for example, like some very specific AI interactions with the features and the users in a very well consolidated product, I think in this case it makes sense.
00:06:43.199 --> 00:07:00.720
But I don't like when the the stakeholders and the business they want to like implement a 100% AI product, let's say, because sometimes it's very, very risky for the users to interact with this completely new product.
00:07:00.879 --> 00:07:06.639
And also in terms of like in my case, because I have most of my products, they are like internal products.
00:07:06.800 --> 00:07:10.319
Sometimes the users they don't trust, they don't know how to interact.
00:07:10.480 --> 00:07:15.360
So they don't use our products, they continue like doing the same old tests.
00:07:15.519 --> 00:07:19.120
So I think it's very tricky because we are still going and like it.
00:07:19.279 --> 00:07:30.399
This is also my challenge, like to understand the process and to evaluate when and how to apply and to interact for the users to interact with this AI uh features.
00:07:30.800 --> 00:07:35.360
Do you have any tips for people that are in the similar situation as you?
00:07:35.680 --> 00:07:40.639
Because of course people are going to have problems, but if they are not using, how do you solve it?
00:07:40.879 --> 00:07:53.279
Yeah, so I try to ask some questions for myself and also with my PMs and with my business team is like, is AI actually better than simple automation?
00:07:53.519 --> 00:07:59.920
Or something like what specific workflow is uh slow, expensive, or error prone?
00:08:00.079 --> 00:08:03.600
So we should focus on this kind of flows.
00:08:03.920 --> 00:08:12.079
And probably the tasks with the high volume and low tolerance for delay, they are the best for implementing AI.
00:08:12.240 --> 00:08:24.720
So I think we have to be aware of like for introducing AI in your product, you need to have time, you need to have a good research, you need to work close with metrics.
00:08:25.040 --> 00:08:40.960
And how do you recognize when a company is adopting adopting artificial intelligence because it's a strategic necessity versus when it's mostly driven by narrative uh pressure or fear of being left behind?
00:08:41.200 --> 00:08:52.320
Well, I think as I mentioned before, it's very clear when, for example, us as a designer, when we see like a very complex flow that could be just three three three steps.
00:08:52.480 --> 00:08:54.159
A flow with three steps, for example.
00:08:54.320 --> 00:09:00.879
Um, a good example is like sometimes I have to send money to Brazil and I use uh Brazilian bank.
00:09:01.120 --> 00:09:02.879
So basically it's a PIX.
00:09:03.120 --> 00:09:04.720
I don't know if you heard about PIX.
00:09:04.960 --> 00:09:14.960
So we have to press, at least I I used to have to press two buttons and only add the name or the number and the value.
00:09:15.200 --> 00:09:18.960
But now they are introducing a set uh AI assistant.
00:09:19.200 --> 00:09:28.720
So instead of just adding some information, some numbers, I have to interact with uh AI uh assistant, um, saying like the name, the number.
00:09:28.960 --> 00:09:37.600
For me, to be honest, it was way, way uh easier to interact only with my inputs than with an AI assistant.
00:09:37.759 --> 00:09:49.360
So I feel like they are trying to be innovative instead of being like uh um introducing a helpful uh help um flows for the user.
00:09:49.840 --> 00:09:59.519
So I feel like when it's something that is like 10 times uh better, I think it makes sense to to add this AI interaction in your product.
00:09:59.679 --> 00:10:04.320
But sometimes when it makes the user feel like more confused, doesn't make sense at all.
00:10:04.720 --> 00:10:11.440
Are there any specific signals or patterns that immediately stand out uh to you?
00:10:11.759 --> 00:10:28.399
Well, in terms of it stand out for me when we are like gathering and exploring a possible AI solution, it's like when, as I mentioned before, when the technology comes first, before the to understanding like the users' uh problems.
00:10:28.559 --> 00:10:38.399
So I think it's very common because as I mentioned, if you don't have AI now, nowadays, if you don't have an AI feature, you are like outdated.
00:10:38.639 --> 00:10:48.000
So for me, and learning from my past experience as well, because I I also implemented an AI assistant one year and a half ago.
00:10:48.480 --> 00:10:56.720
And so basically, we are not solving the users' problems, we are not understanding the users' problems.
00:10:56.879 --> 00:11:04.799
So I think we sometimes it's very boring to understand and to collect all this information, unfortunately.
00:11:05.120 --> 00:11:12.080
But if you need to make sure that this uh solution will be 10 times better than you have.
00:11:12.240 --> 00:11:20.639
So, for example, if your business team comes with something ready, be careful because it could be like a problem in the problem in the future.
00:11:20.879 --> 00:11:44.240
So sometimes we need to just like do babe steps and try to understand how if your team are in the same page in terms of outcomes and do like very, very small changes, very, very small interactions, for example, with uh users and AI, then to have like a very big and complex feature with AI.
00:11:44.559 --> 00:11:44.879
Yeah.
00:11:45.039 --> 00:11:52.480
And I imagine that these distinctions or these situations become even more critical in certain environments.
00:11:53.120 --> 00:12:00.799
I guess my next question to you is how should organizations think about introducing artificial intelligence responsibly?
00:12:00.960 --> 00:12:08.399
You already kind of talked a little bit about it, but specifically right now when they operate in regulated environments?
00:12:08.960 --> 00:12:12.960
I would say they should start with risky classification.
00:12:13.120 --> 00:12:32.399
At least I was trying to do this like to create a very list with the risk classification and not with the the use case, because we have to be aware, like and also our company and our team, they need to be aware about the all the risks that is implementing AI.
00:12:32.799 --> 00:12:49.120
So and also uh answering some questions like what's the impact of being wrong here, for example, and also who is affected if something happens, for example, like financially, legally, or physically.
00:12:49.360 --> 00:13:15.679
So, and also being very, very intentional about data boundaries because it's something that's for example, for AI assistants, we need to be aware about data boundaries, and last but not least, I think we should to be aware about how in the end the users like the human can reveal the final results, how the users can interact and reveal the final results.
00:13:16.399 --> 00:13:16.720
Okay.
00:13:17.039 --> 00:13:29.200
I guess this brings us to what happens internally because even with the right intent, execution, and also talking with stakeholders depends also on ownership, right?
00:13:29.360 --> 00:13:37.039
So inside of an organization, who should actually be accountable for how artificial intelligence systems are used?
00:13:37.440 --> 00:13:49.919
I think in the end, the responsibility should be distributed because once everyone knows what we are doing, like as a team, I think the responsibility should be very well distributed.
00:13:50.159 --> 00:14:02.159
And I feel like sometimes that tension is exactly where many organizations fail because the outcomes sometimes are so unpredictable and it's hard to prevent errors.
00:14:02.320 --> 00:14:05.519
So, as I mentioned, it's very, very hard to prevent errors.
00:14:05.679 --> 00:14:14.799
So, my suggestion is like maybe you can try to create a racy model with the the roles and also some responsibilities.
00:14:14.960 --> 00:14:16.159
I think it could work.
00:14:16.480 --> 00:14:24.240
Should that responsibility sit in like a specific function, or does it need to be shared across leadership and teams?
00:14:24.480 --> 00:14:25.840
What is your opinion on this?
00:14:26.159 --> 00:14:32.720
To be honest, I think it should be shared because everyone needs to be aware about the impact of errors.
00:14:32.960 --> 00:14:40.240
And once they are aware, of course, the system, the tool, like the product will have some specific information for the user.
00:14:40.399 --> 00:14:49.200
So of course, the leaders they they will be like the most likely the responsibles, but I think it should be shared also with the team.
00:14:49.519 --> 00:14:56.399
And basically accountability is one thing, but how it shows up on the day-to-day is something completely different.
00:14:56.559 --> 00:15:08.000
And I think it's also important for us to introduce this topic to the conversation because what does a healthy and responsible introduction of artificial intelligence inside teams actually look like?
00:15:08.159 --> 00:15:10.879
Because I don't think many people are talking about this.
00:15:10.960 --> 00:15:18.080
I've been hearing a lot of people saying that their like leadership is just saying, you have to have AI, you have to use AI.
00:15:18.240 --> 00:15:20.639
Tell me basically tell me where to spend my money.
00:15:20.799 --> 00:15:22.559
But what is your opinion on this?
00:15:22.799 --> 00:15:25.600
What would you say it's the healthy and responsible way?
00:15:25.840 --> 00:15:37.840
Yeah, I think a healthy introduction uh inside of the teams doesn't have to be like a big lounge or with uh like something like a bold AI transformation announcement, for example.
00:15:38.000 --> 00:15:39.360
I think this is very dangerous.
00:15:39.519 --> 00:15:49.120
It should be like very gradual, like like full of sharing experiences, testing fails, and of course, like sharing also the fails because it's very important.
00:15:49.279 --> 00:15:59.759
The teams that do this well move carefully, so slowly at the beginning, and so they can move like faster and later without breaking trusts, I think.
00:16:00.080 --> 00:16:00.320
Okay.
00:16:00.480 --> 00:16:04.639
So let's assume that an organization wants to move in that direction.
00:16:04.879 --> 00:16:13.120
Where should organizations start if they need to start slow, like you're saying, if they want to do this properly, what would you say they need to introduce first?
00:16:13.600 --> 00:16:18.639
I think first of all, they need to put everyone everyone's expectation at the same page about it.
00:16:18.799 --> 00:16:26.639
When I say everyone, it's everyone, like the the the stakeholders, the business team, and everyone same, same at the same page.
00:16:26.720 --> 00:16:31.200
So also like start by identifying the right kind of problem.
00:16:31.519 --> 00:16:35.039
This is very hard to identify the right kind of problem.
00:16:35.279 --> 00:16:49.200
It might be like the high volume or the time consuming, and or also like the problem that involves like a messy or unstructured data, probably, because they are the most likely to be solved by AI.
00:16:49.440 --> 00:16:53.440
So, what's the most overlooked first step that tends to get ignored?
00:16:53.759 --> 00:17:01.279
I think we ignore the most basic, that it's this AI feature will solve the user's problems.
00:17:01.519 --> 00:17:23.440
So I think we are always jumping to the final solution, and also I feel like we are ignoring all this comp those like compliance uh risks that could be like very problematic in the future, and we are ignoring the data, the connection with all the data, and we have good data to connect in our system.
00:17:23.759 --> 00:17:29.519
So I feel like we are like constantly ignoring this very important uh aspects.
00:17:29.759 --> 00:17:36.640
What enables um that kind of decision making specifically under external pressures?
00:17:36.880 --> 00:17:54.160
Well, I I think if you have a good team, like uh very aware and very well trained about the risks, I think it's way, way more easy to like brainstorm together and find the right uh the right solutions.
00:17:54.480 --> 00:18:13.200
Because when you are under pressure, like as a designer, and you have to design the best flow, it could be very, very hard for you because you don't have like the right tools, the right people, and also a team to help you to create this structure and also to deliver the best solutions for the user.
00:18:13.279 --> 00:18:28.160
So you have to receive this support from your developers, your product owner, your product manager, um to understand the user's problem and also to be able to deliver the best AI solution.
00:18:28.400 --> 00:18:41.599
Do you think that one of the biggest misunderstandings right now is that artificial intelligence adoption is about adding this kind of complexity that you were talking about when in reality it often requires simplifying systems first?
00:18:41.839 --> 00:18:43.200
Definitely, yes.
00:18:44.079 --> 00:18:55.599
And and I think like AI only works well when the product goals are clear, structured, and are very, very intentional.
00:18:55.759 --> 00:18:59.440
And otherwise you are just like it's scaling the confusion.
00:18:59.680 --> 00:19:13.440
So if we zoom out from the from everything we've discussed, what advice would you give teams who want to integrate artificial intelligence responsibly without falling into the trap of premature adoption?
00:19:13.680 --> 00:19:18.960
Yeah, I think the biggest trap is like the sentence we need to use AI.
00:19:19.359 --> 00:19:25.440
I feel like uh we instead of we should like change for we have a problem worth solving.
00:19:25.680 --> 00:19:36.240
I think we have to to change the sentence, the the affirmation, and also like answering some questions like is AI actually better than simpler automation?
00:19:36.480 --> 00:19:41.839
Or what specific workflow is slow, expensive, or error prone?
00:19:42.000 --> 00:19:50.400
I also like um we should focus on um tests with high volume and low tolerance delay.
00:19:50.480 --> 00:19:53.680
Well, it's like it's gold if you find this this test.
00:19:53.920 --> 00:20:00.160
And as I mentioned before, uh we have to define a rule that is 10 times or nothing.
00:20:00.480 --> 00:20:17.200
Because if your um AI solution is not like 10 times better than what you have, or 10 times better than something like very simple automation, I think it doesn't make sense to implement AI in your project.
00:20:17.599 --> 00:20:18.799
100% agree.
00:20:19.039 --> 00:20:20.160
Thank you for sharing.
00:20:20.319 --> 00:20:26.960
And before we finish our conversation, I would like to ask if there's anything else you would like to share with the audience.
00:20:27.279 --> 00:20:35.119
Yeah, I think you know it's very it's very like common uh have AI in our day-to-day conversations.
00:20:35.279 --> 00:20:39.039
And for some people, it's like, okay, I can't like hear about AI anymore.
00:20:39.119 --> 00:20:43.279
From other people, it's like, okay, I really want to implement AI in my products.
00:20:43.440 --> 00:20:59.519
But I I think we have to understand that AI it's something that we are still learning and we are still trying to understand how the users will interact, and probably they will love, probably in five years, everything will be about AI.
00:20:59.599 --> 00:21:02.160
And also it could be like, okay, it's just a hype.
00:21:02.400 --> 00:21:10.799
Let's like do babe steps and try to understand what is the best, what is simple, and what makes people solve people's problems.
00:21:10.960 --> 00:21:12.559
And well, I think that's it.
00:21:12.799 --> 00:21:13.920
Thank you so much, Michelle.
00:21:14.000 --> 00:21:15.920
This has been a really great conversation.
00:21:16.079 --> 00:21:36.799
And I guess what it stands out to me is that responsible artificial responsible adoption of artificial intelligence isn't really about starting with the technology, it's about the user flows, the understanding of the systems, the people and the timing, like you were saying, and maybe even most importantly, having the discipline to build the right foundations before adding intelligence on top.
00:21:36.960 --> 00:21:38.559
Thank you so much for your perspective.
00:21:38.640 --> 00:21:46.799
And to everyone listening, if there's one take, it's this AI doesn't fix uh broken systems, it actually amplifies them.
00:21:46.960 --> 00:21:52.000
So before moving forward, make sure what you're building one is actually ready for it.
00:21:52.240 --> 00:21:53.440
Thank you again, Michelle.
00:21:53.519 --> 00:21:54.559
It has been a pleasure.
00:21:54.799 --> 00:21:55.920
Thank you very much.
