Ashu Bhatia on Value Engineering and AI Strategy That Actually Works

Most companies invest millions in AI tools but fail because they skip the foundation that makes AI work.

In this episode, I'm joined by Ashu Bhatia - Global Head of Dexian and Season Technology, AI Transformation Leader with 25+ years at Accenture, Microsoft, American Express, and Siemens, and author of "Value Creation" - who reveals why 60-70% of consulting work still focuses on data foundations, not AI deployment.

We discuss: 

– Why messy data gets exposed (not fixed) by AI
– The three pillars 95% of organizations overlook when building AI readiness 
– Three critical leadership behaviors: clarity, simplification, and prioritization
– How to escape "pilot purgatory" and move AI projects into production 

Connect with Ashu Bhatia: 
LinkedIn: https://www.linkedin.com/in/ashubhatia/
YouTube: https://www.youtube.com/@TechBytesWithAsh
His Book: "Value Creation": https://www.amazon.com/Value-Creation-Information-Technology-Business/dp/1612540368

Connect with Valeriya. 

Need help building AI capability in your organization? Book a call.

Transcript

Valeriya Pilkevich (00:00)
Welcome to AI Made Simple, the transformation series. I'm Valeria Pilkevich, and I talk with global leaders, innovators, and practitioners who are shaping the future of work in the age of AI. I'm joined today by Ashu Bhatia, Global Head of DexSense Technology and AI Transformation Practice with over 25 years of leadership experience at Accenture, Avanade, American Express, and Siemens.

He's also the author of Value Creation, a technology strategy book, which I, by the way, highly recommend you to read. We talk about why 60 % of consulting work is still data wrangling, what differentiates the 5 % of AI projects that actually reach production, why clarity, simplification, and prioritization matter more than pilots, and how to think about AI as a thinking partner rather than a magic solution.

Valeriya Pilkevich (00:50)
Hi, Ashu. Thank you for joining me today.

Ashu (00:52)
Thank you for having me, Valeriya.

Valeriya Pilkevich (00:53)
So just to set the scene, we've met in Bangkok, but you've been living in the US, you have very international background. And maybe you can tell us a bit more about where did you grow up and how that environment shape your early aspirations?

Ashu (01:07)
Sure, sure. No, thank you so much for the opportunity. I follow your work and your content very aggressively. So thank you for hosting me. I'm so excited to reach out to your audience. But yeah, I think as they say, right, the journey kind of makes you and for me, it's been a great journey as yet. So to tell you a little bit about my history and how I grew up. So I am originally from India, but I grew up in Africa and then I was doing a lot of work in Germany actually.

and then moved to England and then US for the last 20, 20 plus years. And now I've circled back here to Asia. The way I look at my journey, frankly, Valeria is that every little piece, you know, it kind of shapes you, right? So I think Africa as a childhood, I mean, different cultures, it helps me adapt to different, you know, just immerse myself culturally elsewhere and really work with global teams very seamlessly. I think I did my engineering, which was always as a kid.

I would take things apart and put them together. And then, my mom would always be a little furious about me doing things around the house like that. But I think it was that mindset that still works very well with problem solving, et cetera. And then, of course, I think when I my first job was in Siemens, actually in Germany, I think the quality, the precision that the German technology instills in you is phenomenal. And I still have that in me. Moved to England, I did a lot of financial innovation at scale and I saw a lot of good things there.

And then, you know, with the U.S. and I did my MBA in the U.S. itself. And I think that put everything together with strategy and finance. I really look at the business of technology, not only technology itself. And I think that's in a nutshell, my journey. But as I say, right, you don't make the journey, but the journey makes you. So that's where I am right now.

Valeriya Pilkevich (02:42)
It's very impressive, I would say. And I consider myself international, living in Germany and Asia and Belarus. So we talked, we met at the conference where you were a speaker at the Dexian event. And I talked to you about AI literacy and how I'm doing a lot of AI literacy training and upskilling and educating the audience as well on LinkedIn and via the corporate trainings.

So you work with global enterprises across different industries, banking, manufacturing, retail, and from your vantage point, why do organizations still struggle with AI literacy even though they invest heavily in technology?

Ashu (03:18)
Yeah, no, that's a billion dollar question to be honest with you. As a professional services person doing a lot of consulting, helping clients understand this journey for AI and data, I get that question a lot. I would say if I were to net it out, right? Like people, I mean, we all know that we wake up to AI, we wake up and read about LinkedIn or watch the...

television or listen to the radio, you drive to work, you're still using AI, you're using maps, et cetera. Sometimes when you want to kind of settle in in your evening, you open entertainment television, still AI shows up, right? So there's definitely an overdose of AI. There's a little bit of AI fatigue, but that's for a reason, right? People, to your point, have invested enterprises across financial services, manufacturing, you name any industry sector, have invested very wisely in data and analytics and now AI, but

what AI literacy means and where the challenges are. think if I were to net it out, three things, right? The first is data foundation. No matter what we do, and I used to preach that not to date myself, that when it comes to data, if I were to take all of this and boil it down to all the business exec, data or information management has three pillars, right? You bring data in, that's the first one. You store data in your own organization, and then you use that data or information to make some decisions. That's it, three things, right?

So that whole assembly line of data starts with the right data platforms, clean data, data hygiene, making sure that you know who produced that data, whether these days it's inside the organization or outside. And then that discipline around governance, around cleanliness, around data management, around definition still is very important for AI to use that data to make more informed decisions, if you will. So that's the first where I feel AI literacy needs to be focused on making sure that they understand the

the foundations are important. I think the second is, if you look at it, AI, as we just discussed, the hype is phenomenal. It's everywhere, and you wake up and it's there, but it leads to some expectations, right? So there could be some misaligned expectations. Now, AI is definitely very important. It's a great technology. It's evolving as you and I are speaking here, but is it the answer or the panacea for every problem? Maybe not, right? You need to understand

where AI is gonna be implemented. Like when softwares and computers came to the being many, many years ago, you put something and you put some input, some output would come in. They call it deterministic output. Now people think AI is similar, but AI is more probabilistic. What that means is because it's based on a technology that is kind of understanding the next word in some connections through all the technology, that it will give you some probability and then it will give you some answers.

So the human judgment becomes even more important in this technology. And that's the expectation that leadership has to set. And I think the third, frankly, is, and we all talk about it, culture, right? They say, right, like culture eats strategy for breakfast, and then of course meets them in the town hall for lunch. But the key is that as a company, as an organization, we're all made within, it's all about the people. So how much is your risk appetite?

Are you willing to do some of this technology at the back end pieces and kind of experiment, or are you bold enough to take it to your consumers on the front end piece of your enterprise? How you determine that is going to be based on how much risk and culture and our people truly data aligned. And I think those three things, you know, in a nutshell, actually make some challenge around where we are in this journey. you know, leaders have to think through this holistically, step back and say, this is the three pillars that I really need to work.

Valeriya Pilkevich (06:43)
I hear also many leaders, many people are now talking about the next step. So agentic future, also agent management skills. And I think what you mentioned about data literacy is so crucial because you cannot possibly work with agents if we think about the future organization where the employees might be working alongside agents.

And they need all of the skills like agent orchestration and understanding the quality of the outputs, right? And data management. what data goes into agents. I really liked the idea that data literacy must come before AI literacy because AI cannot exist without the data.

since we are talking about the data in your speech, you mentioned that everybody's interested in AI, all the companies and want to jump on this wave, but actually the 70 % of where you as a consulting firm get money from is still data.

So can you give us a bit more insight about it? Why do you think it's the case?

Ashu (07:39)
Sure, sure. No, that's a great question. And I do remember that statistic because being in a consulting firm, 60 to 70 % of our work still comes from data wrangling, cleaning the data, putting the platforms ready. So the three pillars I mentioned, right? Data in, data storage, and then using that for cleaning insights and putting it, you know, business users to use something. The analogy I typically say is let's just imagine this, right? You walk into a restaurant, you're right. Beautiful restaurant, beautiful music.

Valeriya Pilkevich (07:47)
Mm-hmm.

Ashu (08:04)
great lighting, you get there, there's a great experience, you walk in, they seat you on the table, you open the menu, the pictures look phenomenal, great food, great desserts, great wine, the pictures are all Instagram worthy, that's all great. But let's say you were given the opportunity to kind of walk behind the door and go into the kitchen, right? In that case, if there is so much chaos, like let's say the labels are missing, you don't know how much inventory came in, which vegetables or produce is fresh,

is the fridge having fish that is exactly a couple of hours or a couple of days old, the labeling is on sticky notes and stuff like that, you know that the chef there is struggling, right? And then you suddenly see the owner walk in and say, don't worry about it, we'll fix this, we'll get a Michelin rated chef and this all will be done. You don't where this is going, it cannot happen. We know that some structure is absolutely needed. You absolutely need to understand if you're producing the data within your organization, let's say you're a retail bank.

and you're capturing data and you're selling all these products. Let's say you're selling insurance, credit cards, of course, deposit loans, et cetera. But these days, a lot of retail banks are getting into more margin products like international money transfer or asset management or wealth management, right? You've got all these products and you have a certain set of processes where you go servicing, know your customer, AML, et cetera. But if you don't have this data foundation, if you're not labeling things as to saying, hey,

we found all this consumer data within our company or we bought it from outside and it's not completely around a discipline of how it's stored, you will never get those insights that you're trying to use to make informed decisions. So just like in the kitchen, what AI will not do is if your data is messy, AI cannot fix it. It'll actually expose it faster, right? In the process, say that you are not doing certain things right. So likewise, I think the reason why a lot of, you know,

myself and a lot of consulting firms are doing still a lot of work on the data foundation is that people have understood the power of the technology for sure is going to be great. But they also the companies who are already foreseeing the future, some are more mature than others. And that's what we do in consulting is we try to understand where you are on the maturity curve. But you need to know to your point that the data discipline is going to be so key to start really leveraging insights and then measuring what you're doing so that you

as a company, whether you're a bank, whether you're a manufacturing company, whether you're a supply chain, you know, logistics partner, that you're using some of those data assets into real information, and then some decisions to really make that business impact that a lot of leaders are asking for. Because if you wake up and you say, hey, this technology is great, and I'm reading all the industry folks talk about this new shiny object. It's a great shiny object, great toy, but end of the day, there's a lot behind it that you need to do.

so that you really leverage in the right way. And I think that's where we are with this current technology.

Valeriya Pilkevich (10:50)
Yeah, that restaurant analogy resonates deeply with my experience at Kentar. I would work on marketing mix modeling projects, you know, trying to predict sales based on historical marketing spent and companies struggled to even locate their data. Sometimes it was lost entirely when the agency contracts ended.

Valeriya Pilkevich (11:11)
It really reinforced for me that without data discipline, no amount of AI sophistication will help.

And that brings me to my next question about your DMACC framework.

Valeriya Pilkevich (11:22)
Do you think the company can train a team of AI agents that exactly will follow this process or do they still have to do the work manually?

Ashu (11:30)
Yeah, no, that's a great question because I am seeing a lot of that in the industry. And thank you for alluding to the book that DMACC framework is like a Six Sigma process framework. I'm a Six Sigma black belt. So I was trained to be very process specific, but to your question of saying that, can we actually use this technology itself to make the technology data better? It's a great question, right? Because the answer is somewhat yes. You can deploy AI and you know, because end of the day, what is AI?

Valeriya Pilkevich (11:50)
Mm-hmm.

Ashu (11:57)
the computers crunch numbers or things and patterns much faster than a human brain, right? We're all very smart. We do things. We learn human beings are the best organisms here on this planet. But computers, once it's something that's repetitive and it's patterns or it's something that can be crunching numbers, they're much faster, right? So yes, AI can be used to clean the data. You can try to bring things and saying, is there nulls in this column? Is the address missing or do you have the right letters?

Or do you have a zip code that does not match a certain city? Yes, you can have rules and it can be used to cleanse that. But before that, you have to step back and have a little bit of a bifocal vision to say that end of the day, that data management capability that you're trying to build in the company is also around processes. What that means is data processes is so important.

When you walk into, again, going back a little bit to the retail banking example, I have walked into banks where when they're selling different products in different units, business units within the bank, they sometimes won't share data amongst themselves. They're like, well, we're selling insurance products and this business unit is selling some retirement products. They will sometimes not share the data. Now, what happens in those data silos that we walk into sometimes is

Even if you've done some work here in cleansing the data and put some data quality metrics around no-nulls or real-time data and others have not, AI can take that in, but it's going to waste a lot of cycles and is going to exacerbate the problem that might exist in the company. So the answer is yes, AI can absolutely be used, but you need to step back and say, what is my data management capability? Why do I need to have clean data for certain important functions? So if you're working in healthcare, for example, you have

Valeriya Pilkevich (13:24)
Mm-hmm.

Ashu (13:38)
you know, if let's say it's some injection, some dates or some machine that is monitoring some patients, there is no scope for data lag, right? Or data miss quality. But yes, suppose you're doing something where you're sending flyers for a new credit card to a certain address and it goes to wrong address that that error is permissible, right? So depending on the use case that the company is trying to do, you can deploy these technologies, but it still goes back to the whole management of the data.

Valeriya Pilkevich (13:57)
Mm-hmm.

Ashu (14:06)
and the process behind it before you start applying technology to cleanses better.

Valeriya Pilkevich (14:10)
I think you also mentioned it in the book that data quality initiatives have to encompass the whole enterprise, Like people, processes and People we already before, data literacy, right? Data ownership, clear ownership, processes you just mentioned and technologies that's where we use AI, maybe agentic help at some point when we are clear on the people and the processes part.

Ashu (14:18)
Exactly.

Valeriya Pilkevich (14:32)
if you had to coach a leadership team for 90 days on becoming AI ready, what are the three, behaviors or habits you would focus on first? And maybe you can also tell us what is AI ready, because we always hear about AI first organizations,

Ashu (14:45)
So now that's great because I'm in consulting, I help a lot of clients and there's a lot of jokes on consultants that they create all these terms. So AI first or AI ready, it used to be client server ready, cloud first, and now it's AI first. So we the way that the technology is evolving.

I think the three things, the three behaviors that I would really recommend is first is clarity, being very clear as to what kind of things are you going after. And that's very tough sometimes, because at the leadership level, absolutely there is, they know, right? That there is, this is my shareholder value tree, it's more revenue.

Valeriya Pilkevich (15:12)
Mm-hmm.

Ashu (15:20)
lower cost or reducing my risk. If you unfold that whole shareholder value tree, where am I going to be leveraging this technology? That clarity needs to be percolated down to the teams, right? So it's all about the communication. It's knowing some of that and they say, right, like if your communication is very little, it's like a tree fell in the forest. No one knows about it. So you want to make sure that once you're clear about what you're pursuing, that it should percolate down to the employee level, not just the middle management level.

Valeriya Pilkevich (15:28)
Mm-hmm.

Mm hmm. Yeah.

Ashu (15:48)
or senior director level, it should be at the nth employee level so that people are clear that this is what they're working towards. So that's, I would say, a behavior that leadership needs clarity. The second is simplification. What I mean by that is there's so much going on that it's very easy to get overwhelmed, right? You can't be going after everything, but you have to say, OK, if this is my use, because I'll give you an example. We just talked in your example of marketing and how people use data to do marketing push.

And sales forecasting is a big problem statement that I work with retailers sometimes, as well as supply chain people. I was working on some sales forecasting tool for a client and they're saying that we should go after all these use cases and that all made sense. But I asked them one thing. said, we can go after all these use cases, but before we do everything, like this was like a, you know, a drinks company trying to order in which stores will the inventory need to be changed? I said, every Monday, who decides the inventory to be reordered?

What do they look at? Start with something simple. Where is the human trying to make that decision? From there, AI will evolve itself, right? Rather than us saying, hey, this is the value chain. These are the use cases. This is the potential, or this is the 30%, 40%, 50 % productivity gains. All that is great. It gives you a starting point. But start from where the decisions are being made in the field. And once you pick that up, then you unfold it. It simplifies how you deploy your AI solutions.

And I think the third would be the most important where a lot of companies I think struggle with is prioritization. Because everyone knows the power of this technology. Everyone wants something going on, right? I talked to chief analytics officer in a bank one day and he made a very funny statement. He says, I have more pilots than Delta Airlines or Lufthansa. And I was like, ⁓ then it made sense to me that they're doing a lot of experiments all across.

And no one is able to say, which things do we really need to go after, right? So if the leadership understands some of the potential, needs to step back and say that if you're the CEO of a company, of course, you will not be completely hands on with the potential of it. So talk to whoever is making that decision, whether it's a chief technology officer, CIO, chief data officer, whatever, understand the potential and say, these are things that we need to kind of prioritize, work, have a working group to help understand, as I say, right, if everything is important,

nothing is important. So help your company understand which pieces to go after because from the initial momentum, you will reap some success, whether it's external facing strategy, what your competitors might be doing or where your industry is going, but have some of those patterns going. That's when your employees and your whole organization can kind of really say, okay, we're all rowing in the same direction and this is where we need to get to.

Valeriya Pilkevich (18:25)
So clarity, simplicity, prioritization. I love it. Regarding clarity in your book, you also mentioned when you talk about culture, you also talk about vision trap, right? Where optimizing individual processes or parts does not necessarily lead to a big optimization on the company level, company-wide. So I think it's also very much aligned when trying to, do different pilots or different AI initiatives that are not aligned with the

general vision and general understanding of where value is generated, and speaking about pilots, like you speak a lot, as you mentioned, with CIOs, CTOs, CEOs across different companies, different industries, multiple regions. what do you say differentiates organizations or companies that actually get value from AI instead of just, doing another pilot? We also know all of the statistics from the recent MIT report that 95 % of

all the projects, AI generative AI projects fail and only 5 % actually get into production and generating actual value So what do you say differentiates those companies or those 5 %?

Ashu (19:28)
Yeah, no, that's a great

question. You know, it's that statistic from that study came up recently in another thing that we were talking about, because it's scary from a leadership vantage point to say that I'm investing so much capex, so much op-ex the power of this technology. And if I don't get the benefits and a lot of companies are struggling right because of that. So I would say the companies that are trying to win are doing two or three things right. And sometimes you have to look at it from a long play and I'll explain what that means.

The first thing is, if you step back, and you think about this technology, if you just look at technology in general, most of the technology transformation would come typically sometimes from the military, from organizations, and then come to consumers, and then it would go, right, whether it was cell phones, you know, or it was internet, for example. But GenAI was actually the opposite, right? So OpenAI came in and said, hey, there's ChatGPT, let's roll it out. The consumers are already playing with it.

The enterprises said, let's be careful. Let's see if they can do it. And now the companies in the boardrooms are saying, this thing is real, right? So they say that AI is more adopted in the cubicles than in the boardrooms, right? there was another study that said 67 % of the employees are using some tools like this to make their jobs better or make their day better. So the challenge for the companies is to give them tools inside the company within the security bounds.

so that they can have the same experience and still leverage this technology. And that's a big challenge because you want security, you want firewalls, you want proprietary data to stay within your firm. So that's something that leaderships are thinking through, So coming back to your question on saying, what are the behaviors that are making some of the people win? The first is understanding that what is it that the employees can use? Where are those productivity slash, revenue gains or risk reduction gains going to be working? And

at least incorporating a digital lab or an AI lab. And what I mean by that is a lot of my clients will say even at the senior level, hey, that because things are evolving so fast, should we just wait a little bit and see where all this goes before we invest anything? And our answer typically is you've got to be doing things right now. So you have to be doing the experiments as the technology evolves. Not that you can deploy it right away, but if you have an AI lab or a digital lab, I would say have three lengths to that, right?

have a telescopic lens, look at the telescope, see where this is coming in a year or two, is it merging with quantum, what is happening, have some innovation lab, look at where the art of the possible is, and intermingle that with your shareholder value. The second is have a little bit of a binocular lens saying, hey, something near shore, what have I already deployed? Like the sales forecasting. If like, I'll give you the example where we're working with this retailer, they built this sales forecasting modules, now comes in the CMO and says, I want to now use some of this data

to see if I launch a new product, how much will it cannibalize my existing products, and what is my forecast for this thing? And then can I mix things like events or weather data? So all those things evolve, but you have to have that binocular lens and you have to have things ready. Otherwise you will miss some of the decision criteria right there. The third is you have to have a microscopic lens. When you're actually deploying something for the real use for employees or consumers, you need to make sure the data quality testing, et cetera, has been already deployed.

So the winners here are the ones who are already doing these kinds of experiments with the right mindset and the right lens. And the third thing that's coming today these days is, and I was talking to someone, the way this technology is evolving is litigation is never keeping up with it because the technology is advancing so much. There's not enough regulation, not enough rules, because no one knows where this technology is going. So there's a term in the industry which we call responsible AI, right? So R-A-I.

So the companies that are doing it right have realized that to win in the long-term, they should better put responsible AI or AI by design rather than thinking about it later or through reaction, if you will, through something. Because if the technology is going at lightning speed and your company is developing things at speed of sound, which is fast, it's still slow. So it's better to understand and step back and say, what should I be doing in terms of explainability, transparency, and all those realms

so that I am not doing anything wrong for my consumers, because then it leads to so much risks, whether it's reputation, lost revenue, whatever. You want to kind put the experiments, the responsible AI by design, and then doing things slowly to say which use cases you deploy. So the companies that are kind of focused on all of that are a little bit winning. And I think those are the companies that will really be able to win in the long term.

Valeriya Pilkevich (23:54)
I appreciate the perspective on responsible AI. Here in Europe, we take a very deliberate governance first approach, which has real benefits, but sometimes can slow our pace compared to ⁓ USA or Asia. So it's an interesting tension between innovation and responsibility, I would say.

Valeriya Pilkevich (24:13)
While we are talking about pilots, actually want to ask you another question, and I want to pose a quote from your book. You're saying when you talk about value creation, about processes, that whether a company manufactures physical products, develops software, or provides services, the thought process of who creates value needs to be understood from the customer's perspective. So do you think it's the same when we think about use cases for AI pilots or AI projects?

Ashu (24:41)
Yeah, no, that's a great question. First of all, a lot of people differ on the term value, right? What exactly is value, right? If your company is a little bit of a black box, there's things coming in, value is supposed to be going out. So you've done something to the inputs by which the consumer is willing to give you more, right?

So as they say, the value is to be measured from within the company and the stakeholders could be shareholders of the company as well as the consumers. So the reason I put that code to saying that everything should be looked at from the consumer standpoint, right? So if you even take a look at insurance, for example, property casual insurance, let's say you have a car and you buy some insurance. They say that no one wakes up in the morning and says, ⁓ today's a great day. I'm going to go and buy some insurance. No.

Valeriya Pilkevich (25:10)
Mm-hmm.

Ashu (25:23)
Yes, you wake up and excited about, buying new shoes or a new suit, know, insurance products that are not as glamorous. So yet you have to be looking at what you have to be doing in this industry. So I talked to a lot of insurance companies would say that for us to look at new products, we are thinking, oh, maybe, you know, you take a Grab ride or Uber ride. Maybe I can buy you some, sell you some micro insurance for that ride, right? At literally, literally at $2 a ride or whatever. So they have to think creative with what they're doing, but

value has to be perceived from, again, the consumer standpoint. What is the consumer doing? If he or she is not as excited about your product, where is he or she participating? For example, if you walk into any big malls of the world, there's people who are trying to sell you credit cards at the time of checkout from your POS terminal, depending on where you're shopping, right? They're saying, hey, can you buy the Citibank card that we're coming? So,

People are always saying that if Valeria is a consumer and she buys things here or she goes hiking there or she consumes this, where else can I integrate my products so that I can make her life experience better? That's how companies need to be thinking. In fact, I was reading a book, it's called, think, Competing in the Age of AI by Lakhani. They said that

Valeriya Pilkevich (26:25)
Mm-hmm.

Ashu (26:32)
With AI, it will be even tougher for companies to maintain their core because people are thinking of completely new business processes, completely new business ideas and out of the box because the technology is providing and enabling that, right? They give an example where they said, for example, you know, E-Trade companies like Amazon or Shopee in Asia, you know, you would do things quickly and within two hour delivery to your home and, you know, you're either capturing a lot of market.

Now they're coming up with saying that, let's say you wake up on Monday early morning before you head out to work or start your podcast and you open the front door and there's already a box from amazon.com there based on all your history. They know that you're going to be running out of, let's say the milk or the cereal last month or whatever it may be. And because AI is used for that, it may not be a hundred percent accurate. And out of the 10 items, you take the nine because they make sense and you leave the one item.

and a drone comes and picks it up for reverse logistics because you don't need that one item, right? So all of that system has to be integrated to your business process to make sure that you're leveraging data, AI, and the robotics to make sure that the consumer has that experience. So if you think through that lens, then a lot of new solutions are going to keep coming. And that's what is going to be really the power of this technology, I think.

Valeriya Pilkevich (27:48)
It's a great perspective.

Ashu (27:50)
Absolutely. It opens

the eyes, right, for all of us living in this age and day of AI and a lot of people told me you should start watching some of the old sci-fi movies because that's where the ideas are going to come from.

Valeriya Pilkevich (28:01)
Ashu, for teams and organizations feeling overwhelmed by the base of AI development, and I think many of us in this space feel that pressure, what's your advice for building confidence and maintaining an experiment-driven mindset without getting swept up in hype?

Ashu (28:16)
Yeah, no, that's as you said, right? All of us, despite at the crossroads of all these tools and the processes, sometimes you say there's no way you can keep up with so much happening because the big firms, the FAANG as they're called, they're billions of dollars. Every time we wake up, there's new research happening, right? New LLMs, new hardware, new data centers, et cetera. So I think if, you know, and again, life has to go on and you still have to do what you have to do. The way I look at it is if you were to really

do these experiments and kind of not be overwhelmed and still seek the value that the technology will give you. I think there's two or three things that you can do. The first is always anchor AI to one real decision. Like I said, like if you're doing sales forecasting, ask on a Monday, where do you start? Who is ordering this new inventory? What information do they use? What are the data sources? What are the confidence levels? So come up with that. So that's the number one thing. The second thing

Valeriya Pilkevich (28:56)
Mm-hmm.

Ashu (29:08)
time box your experiments. You need to have some experimentation, but if you let it go on, it will consume a lot of your resources. So whatever it may be, four to six weeks, define the success criteria, define why that kernel or that shell makes sense before it can be deployed or taken towards, ⁓ prove the value before it can be going to scale the value phase, as they call it. And then the third is, I would say that Always think of AI

as a thinking partner. They say that, and you mentioned that in the beginning, that the more you ask, the better answers you'll get. And I realized that the more, because this technology is still kind of always about, it has some understanding and the patterns, but it started as a tool which was doing just the next best word, if you will, the simplest core. So it's going to provide some answers.

Valeriya Pilkevich (29:36)
Mm-hmm.

Ashu (29:57)
But the best thing is your hard work is going to come by asking and being more curious, saying, hey, if this is what happened, why did this happen? Can you go back and give me another perspective? Can you crawl the web? Or sometimes people are saying to reduce hallucinations or the effect of it, sometimes use two different engines, right? And let them argue and let them debate on the same topic where you can remove some of those blind spots. So the idea is that the managers who are truly have taken that whole Socratic kind of thing to ask more questions.

are the ones who are going to eventually win because no doubt about the fact that management has to explain their decisions, has to explain why they did that, but they have to be asking way more questions in these experiments to really reap the benefits of this technology. So one of the things for AI training, I think, is the habit to ask more questions, more deep level. If you're really doing root cause analysis to go to that hole, they say the five layers of why, why, why.

Valeriya Pilkevich (30:26)
Mm-hmm.

Ashu (30:52)
and really understand some of the root causes of certain behaviors. And those are the organizations or the behaviors or the experiments that people are going to use to really kind of win with this technology.

Valeriya Pilkevich (31:05)
Thank you for being on this podcast. It was very eye-opening and I'm sure will bring a lot of value to our audiences.

Ashu (31:11)
Thanks so much, Valeria, for having me. Appreciate it.

Valeriya Pilkevich (31:12)
You can find Ashu Bhatia on LinkedIn, follow his YouTube channel, and learn more about his book Value Creation. All links are in the show notes. If you enjoyed this episode, follow AI Made Simple, the transformation series for more conversations with practitioners shaping how AI is actually adopted inside organizations. Thanks for listening.