I’m excited to share my first podcast on “The AI Data Fabric Show” with Mihir Shah as my guest today. The AI Data Fabric Show is a conversation with battle-tested CXO/CDOs and AI leaders, entrepreneurs, and investors who make big bets to transform the enterprise. Mihir was previously the CTO/CIO/Head of Data at Fidelity Investments. He has been an influential advisor to leading startups like Snowflake, Skyflow, Reltio and others. Every CXO and entrepreneur must hear his insightful observations with 30 years of transformational leadership.
Here are my top takeaways from my conversation with Mihir:
- LONG beats FAST when it comes to Data and AI
With the crazy pace of AI evolution, enterprises and CXO’s are busy driving high velocity POC’s focused on specific use-cases. Now, more than ever, it is critical to think of building a foundation for data and AI that will cover a wide spectrum of use-cases. This longer view across businesses and functions should result in the right architecture and platform. It reduces tech debt and silos. This approach will win out on longer term yield and cost of scaling AI and data in the enterprise. The CXO’s job is to partner with business leaders and “sell” them this path. Think Data Up, not Function Down! - Design around constancy and simplicity if you want to disrupt for scale
All transformational CXO’s struggle to go through the disruptive cycles of innovation with infrastructure, data platforms, and AI models. Disruption is easier when you build around your enduring IP, which is typically the “data model” that rarely changes. Architecture should be simple and pragmatic. As an example, UI’s and the middle tier are typically stateless, but databases are stateful, so some lock-in is inevitable. Balance this lock-in with sufficient portability. - Huge value from AI will get unlocked with structured data in the enterprise
There is a lot of focus on unstructured data at the moment, but much of the next innovation comes from the bulk of real enterprise business processes in structured data platforms, whether ledger, customer, transactions, trades, etc. Every person/agent and every job function should have data at their fingertips to be able to perform their job better. Self-service should be the goal of any analytics strategy, whether it’s AI or data or BI, or operations. - Self-Service Data is an open problem for AI
Regardless of whether data is distributed or all in one place, realistic enterprise data has the scale of thousands of tables or objects. Every function and agent will want to see the data in a different way. Asking users to navigate and self-service is too complex and a big gap today. There is a need for curated and personalized visibility and insights. New data fabric architectures are needed (beyond the modern data stack) to address this problem. - CDO’s time has arrived!
For a long time, CDO’s had to spend all their energy to bring data to the forefront and articulate the importance of data foundation. Now everybody’s on this bandwagon. CDO’s need to take advantage of this moment and rally their teams and partner with the business teams. Their time is now.
Listen to the full conversation with Mihir Shah on The AI Data Fabric Show to dive deeper into his complete enterprise transformation playbook (or click here to listen on Spotify or Apple Podcast). The full transcript of the episode is available below.
Full Transcript of the Episode
Prat Moghe, CEO, Promethium – 00:00 The business now has become really impatient. Everybody wants access to data, everybody wants to talk to data.
Mihir Shah – 00:05 Business process change all the time. Like every quarter, there’ll be tweaks in the business process. Your data model does not change unless you enter a brand new completely change your business.
Prat Moghe, CEO, Promethium – 00:15 What does a good data architecture look like? To be able to support AI?
Mihir Shah – 00:19 You cannot have an AI strategy without a data strategy. You do need a foundation data strategy. Don’t look at point in time, look at what’s going to happen in the next five years. So every role in an organization needs to actually have self service data to be able to do their job better.
Prat Moghe, CEO, Promethium – 00:39 Welcome to the inaugural AI Data Fabric show podcast. I’m super excited to be hosting it and following up on our Data Fabric podcast show that was hosted by our founder, Kaycee Lai. It’s an exciting time and particularly excited to welcome Mihir Shah today on the show. Mihir and I have known each other for several years. I’ve been fortunate enough to learn a lot of important lessons around data from Mihir. So again, just by way of background, Mihir comes from a deep and long experience being an industry leader in data analytics. On the enterprise side, he just retired from Fidelity after 30 years there. Congrats, Mihir. Thank you. And where he was the CTO, Chief Data Officer, Head of Enterprise Data Architecture. Enterprise architecture. And then again, there’s a lot of transformational things there that he accomplished.
He was also before that started his career at Churchill Insurance where he was a lead architect. This was an innovative startup in the UK in the insurance space. Before that he was at TCS and now he spends time being an advisor to many startups, including Promethium. I’m super excited about that. He’s a thought leader in the space. He’s an advisor in residence at EY. He’s also a venture partner in several investment funds. So welcome to the podcast.
Mihir Shah – 02:09 Thank you. Prat. Really, really excited to have this conversation.
Prat Moghe, CEO, Promethium – 02:13 So I’m gonna ask you a couple of questions that again I’ve been super intrigued about. First thing is how did you get started in your career and anything where you can sort of what led to that spark where you. Because I think one thing that I’ve been super impressed about with you is you’ve been able to do transformative things in large companies with disruptive technologies and that’s not easy. That transformational CXO role is not easy, but you had that speed and scale point of view. So where did that start?
Mihir Shah – 02:49 So I was never a computer science graduate. I did Mechanical engineering. I did my MBA after that and then I decided after my MBA that I wanted to work in the software industry because I thought that’s going to take off at some point. I was right. I didn’t know what I would do. But as. So that’s when I joined TCS. They had a fantastic training program to take anybody and put them through a boot camp and you start programming. And they asked me to go into the management consulting division. And I said no, let me learn programming hardcore. And that’s what I did initially. And TCS was a great platform to learn all of that. But I think the big break came, as you mentioned, got an opportunity to be employed by Churchill Insurance. I was one of the 10 people on the team.
Prat Moghe, CEO, Promethium – 03:38 This was in London or this was.
Mihir Shah – 03:40 In London in 1980. In the 80s, late 80s actually. 88, 89 timeframe. If you wanted to build a large system, the only way to do that was on the mainframe. Your x86 Unix boxes, Oracle, they were just starting to come up. And Churchill was pretty innovative from a business perspective. The experience a customer has in buying motor insurance in those days was you go to a broker, you fill out a form, that form goes through the mothership where the underwriting algorithms then crunch the data and return the quote. And then the broker tells you the quote next day. And then you start the whole process and takes about a week to get motor insurance.
And what we figured out was how to do exactly the same underwriting process and crunch billions and billions of rows of data with the underwriting model while you’re waiting on the phone in split seconds. And that was a breakthrough. And just based on that breakthrough, we launched a company that you call a 800 number, you answer 10 questions, you get a quote, you answer five more, you get a policy, you want to refine it, you can do it on the phone. So literally from days we actually had a business model where you could get a policy in seconds. It was wildly successful, as you can imagine. Probably one of the first insurtech fintech companies. And this was in 89. That really launched my career.
I mean every single job that I have done, consulting assignment and since then, including my recruitment at Fidelity was because, oh, you worked at Churchill, we want you here to do xyz. And it was an OLTP system highly. We had thousands of telesales operators fielding calls coming in and it was all the transaction processing. So that was my initial foray into databases. But more importantly, I think I was 27 at that time. And then you go into the corporate world, you just don’t want to do the regular, keep the place running. How do you top what you just come from? So you try to replicate that in the corporate world by obviously thinking big, coming out with big ideas. And if you step back a little bit, Churchill took almost 14 months to build a foundation. Then we launched the business.
Mihir Shah – 06:16 And that discipline stayed with me, that you want to do something big. It takes time to build a foundation, then you launch it, and you need to find the right employer. And actually, to be honest, Fidelity, being private really helped that they invest in the long term. They have enough patience if they believe in the idea. And so at Fidelity, I was able to do a lot of big things and three or four major transformations.
Prat Moghe, CEO, Promethium – 06:44 So at Fidelity, like coming back to your point, you’ve probably seen this transition on technologies. You saw clearly, like you said, this idea of speed scale. Yeah. So talk to us a little bit about those lessons because I think that would be really interesting to other Chief Data Officers, CXOs, CTOs, Chief AI Officers now, because they’re all like struggling, I think, with the same question.
Mihir Shah – 07:08 So my last before I retired, one of the things we did was brought all of enterprise data into one platform.
Prat Moghe, CEO, Promethium – 07:17 Which is unheard of. Right.
Mihir Shah – 07:18 Which is unheard of. Right. And they’re all catalog modeled. So the final resting place for all data is the analytics platform. Now trying to do that. The execution is fine. There’s a lot of technological challenges, et cetera. So that’s good. But I think the biggest challenge is how do you convince your business partners to have the patience or have the courage to build that foundation? Because I think what I’ve seen is typically people think in terms of use cases. I have five use cases. Let’s get them done. And then somebody else in some other department has their priorities that they have 10 use cases, let’s get those done. So you end up chasing the use cases, trying to create value very quickly for the business, which is a good thing, but at the cost of building a foundation.
In fact, it’s almost the opposite of the foundation. You’re actually creating technical debt when you’re executing in silos. So what people need to do is aggregate all the use cases and aggregate all the current states at the enterprise level, or at least at the business unit level, and then come up with a foundational strategy. And that strategy, or your blueprint then becomes how you execute or what you move towards. I think that is the thing I’ve seen missing is that if you’re a chief data officer, you’re expected to produce results very quickly and that’s how you stay in the job. Which means the only way to do that is very quickly execute vertically. Every use case that come in terms of the priorities of the business and you lose focus on the fact that you also need to build a foundation.
Mihir Shah – 08:58 So that’s really important. I think my biggest lesson is that have the patience and the time to build the foundation first before you execute on the use case.
Prat Moghe, CEO, Promethium – 09:07 I think what you’re saying is interesting, but the key point is that you’re somehow being able to combine that. Step back, look at the aggregated thing and execute on it, but do it with speed because like you said, most CDOs run out of time and then they’re not able to demonstrate it. So how do you convince, when you have this thing of multiple use cases across different groups, how do you convince people to work with you in building that foundation? Right, like getting the data right is to me like a little bit like laying the road out that will support traffic from different parts of the town. Anything you can share, who do you bring on as an ally? How do you convince that?
Obviously you are probably a great sales guy, but you also have probably a great record in delivering and building trust. And so anything you can share?
Mihir Shah – 10:02 I think, I wouldn’t say there’s one particular way. Circumstances are different in different companies, but I think the most generic way is to kind of not look in silos and look as a chief architect or as a CTO, you’re one of the few people in the company who have a cross business unit, cross functional view. So leverage that view. And that’s really the job is to kind of say, I’m not working silos, I’m going to actually build an aggregated view for people to see. Now there are many times business priorities say that’s too risky, we have other priorities. We’re not going to do that, but at least give it a shot and you’ll be surprised. When you build a business case, when you paint the big picture, you will create a scenario that they haven’t seen before because everybody’s working on the silos.
But when they see the big picture, everybody wants to do the right thing. So it takes as an architect, as a chief architect, it takes time to do an entire current state. How many databases do we have in the analytics space? And the entire company, what does it cost? What hardware does it run, what is the technical debt involved? And then the time value is. You don’t look at point in time, look at what’s going to happen in next five years? Right. So when we did this, we had actually in the next five years would have ended up spending the same amount of money that we spent on building the foundation because there were three new data warehouses being planned. Hardware was being replaced in the five year span because were on premises.
And there are many other maintenance costs and modernization costs that each one, every business unit is planning. So do you aggregate all that? And that’s a big number. And you can actually make an argument saying with the same number you can actually modernize, meet your business case needs and we can actually build a much more solid foundation for the long term. The only compromise is that not everybody will get their use cases exactly at the time that they want. And are you willing to compromise on that? And that’s a give and take that you do with the business?
Prat Moghe, CEO, Promethium – 12:15 Yeah, because in the time you were at Fidelity, for example, there was like a whole transition in terms of technologies and timescale. So in general, as technologies have moved, I’ve always seen that you always manage to jump to the next technology, which is not easy. Disruptive technologies are not easy. As you do that, what’s kind of been the mantra for you? Where is the IP in all of this? And how do you. With so many things changing, what is constant?
Mihir Shah – 12:44 So I think it’s easy in the data space, to be honest, because the first principles of data have not really fundamentally changed from in the last 30 years. It’s the same thing. At the end of the day, you have a business data model, you have persistent data that is stored on disk, and then you have an engine that is able to transact with that storage unit, whether it’s retrieval or so at the most fundamental level, nothing has changed. And maybe the technology has changed in terms of the topology of how things are deployed, the separation of compute versus the storage. So there are many variations. But if you really look at data, nothing has changed. The second thing is at the end of the day, technologies may change, but your core IP is in your business data model.
And if you have a great business data model, as you know, data models are really. Data models have a lifetime. I mean they last for their lifetime unless you business. Yeah, they don’t variant unless they’re not. Right. So unlike business processes, which can change when management changes, when business process change all the time, like every quarter, there’ll be tweaks in the business process. Your data model does not change unless you enter a brand new, completely change your business. So if you build all your systems Based on a solid data model foundation, then it’s very easy to move from technology to technology. You’re kind of immune. The second thing is keep it simple, right? I mean, in my last project, we just basically said, our IP is the data model, our injection pipelines are Python and SQL.
Then if you want to lift and shift all of that, you could do it. Now. It’s going to be costly, but you can at least do it. It’s not a complete rewrite. The one thing about data is that any databases in your stack, your middle tier is stateless, your UIs are stateless, your databases are stateful. So no matter what you do, you’re going to get locked in. So you got to remember that and then pick your technology saying, how much do you want to get locked in? And to whom and what partners? And one of the principles I have followed in the past is because you got to lock in all our data technologies. We wanted to be CSP independent, so portable across. Portable across. Whether it’s Azure or GCP or AWS Oracle, we want that portability across CSP vendors.
So when it comes to database technologies, we look for, right from transaction processing to caching to analytics, we look for technologies that are portable, but you’re still locked into that technology. That’s one particular. There’s no way out of that.
Prat Moghe, CEO, Promethium – 15:33 Got it, Got it. Great. Let’s switch to AI and data now, because that’s a super interesting topic that as AI comes into the enterprise, one of our observations has been that the business now has become really impatient. Everybody wants access to data, everybody wants to talk to data. Everybody’s essentially imagining that AI will drive value out of data faster. Can you comment a little bit about this whole question of data architecture? What does a good data architecture look like to be able to support AI? Because our observation has been that the modern data stack, which is the modern data platforms and pipelines, and they’re great for batch processing, but when it comes to ad hoc and agents wanting access to data, how do you get that data ready for AI?
Mihir Shah – 16:24 So I mean, just back to basics. I mean, you cannot have an AI strategy without a data strategy. You do need a foundational data strategy. The observation I’ve had with actual very practical kind of experience is that when you sell a foundational data initiative, AI is one of the biggest drivers. You sell it. It’s saying, hey, you need to be ready for AI. So we need to do a foundational data strategy. When you build foundational data strategy, your value is coming from, not from AI. But the fact that you got clean data and you have facts about your business at your fingertips at any given point in time. Right. How many customers do I have? What were the sales in this region yesterday? Those kind of basic questions comes from the data foundation and doesn’t come from AI.
So there’s a byproduct of AI drives investment data foundation. And then the value comes from actually just basic facts that you weren’t able to get. Right now I’m not. So you immediately get value. Now from an AI perspective, obviously, you know, being able to combine data sets. So let me back up a little bit. Right? So my observation is that you can get, you either have a better model than somebody else or you have unique data that nobody else has. Right? So there are really two ways to get competitive advantage. So the better the data that your first party data that you have is actually a big competitive advantage. And to organize that and have your AI models, agentic models, machine learning, whatever you’re doing, have access to your first party data is really important because your third party public data, everybody else.
The other observation is that recently the focus has been on LLMs. Everybody’s talking about gen AI and LLMs and which is great because there was this massive corpus of data which was locked inside documents and unstructured data and videos and stuff like that and you are able to now process it. But I think we’ve gone too far in the sense that we just completely ignored the fact that your traditional intelligence is in your structured data. It’s in your general ledger, your sales, your transactions, your trades, et cetera. So we need to kind of strike a balance between yes, we need to leverage the unstructured data, but what about our structured data? How do we actually leverage that using agenting AI? I think that’s a big area.
I would say that one of the most important things is to have every Persona or every job function. If you are sitting and doing a specific job, you should have data at your fingertips to be able to perform their job better. Right? Very simple. If you’re a project manager, you shouldn’t have to go to ask a finance person saying is my project on track?
Prat Moghe, CEO, Promethium – 19:29 Right.
Mihir Shah – 19:29 Okay, you should. Or if you’re a finance guy and you’re looking at red, green, yellow, which project are on track or not.
Prat Moghe, CEO, Promethium – 19:37 You shouldn’t have to go to the.
Mihir Shah – 19:38 Project manager, you have to go to the project manager. So every role in an organization needs to actually have self service data to be able to do their job better. And I think that should be the goal of any analytics strategy, whether it’s AI or data or BI, et cetera.
Prat Moghe, CEO, Promethium – 19:56 What we are seeing is that many people feel like you can point these LLMs and models to your structured data and then you are done. But the reality is that there is a big gap in terms of context between the business people and the data that exists. Like you said that data model itself is in many cases it’s in people’s head. Right? Like people interpret it differently, whether it’s semantics, whether it’s technical context, business context. So how meaningful the question really is? Like how meaningful is it this gap to self service where anybody can get access to data, ask the questions that are relevant to them and get the answers that are relevant to them.
Mihir Shah – 20:40 It is really important, Prat. Because look, I’m coming from a technology side. I used to think that your traditional data model, fully documented in a catalog is good enough.
Prat Moghe, CEO, Promethium – 20:51 Good enough, yeah.
Mihir Shah – 20:52 And then you kind of actually sit down with the business partners and then you realize it’s not right and they need to see the data in their own way and the way they see it in real life in their day to day working environment. So even though your foundation data is based on a very solid data modeling foundation, every function in Persona would like to see it in a different way. So that’s one problem. The second one is when you bring all the data together in one place, which is a lot of companies are doing it as part of their analytics strategy. That’s great because now you’ve got rid of all the silos and distributed databases all over the place. But then now you have analytics warehouse with maybe 5,000 tables in it.
So if you just give access to individuals to say now navigate these tables and find self service yourself, it’s not possible because it’s too complex, even with the catalog. So you got to give people small curated views. In fact, the best thing would be they should be able to create the small curated views themselves. And maybe there’s a power user in a department, that person can go and create a view for finance or create a view for risk, create a view for security or compliance or whatever that function may be. And that I think is still, we are still lacking a good way for users to interact with your schemas. And I think that’s a huge opportunity. I know Promethium is working on your whole idea of a fabric and being able to create these views self service.
I think that’s a really big space that needs to happen.
Prat Moghe, CEO, Promethium – 22:33 Thanks Mihir. We are super excited about it and I think the key is, like you said, leveraging all the past innovations that have happened, including great data models, modernized data stack catalogs that exist and tools that exist. So where are you heading to next? I know that it’s been probably a super rewarding journey after you’ve kind of gone through all these challenges. I know you’re part of many different startups, you’re part of funds, what are you seeing and what’s next.
Mihir Shah – 23:03 So you know, I’ve worked in large enterprises last 37 years and then obviously decided to spend more time with the family, more time on hobbies and traveling and stuff like that, which is all good. But you’re not going to stop working. I mean, you need that stimulation. You want to be out there on the edge knowing about what’s coming next and helping out young entrepreneurs with their product ideas.
Prat Moghe, CEO, Promethium – 23:33 And you got involved in quite a few formative startups, right? Including Snowflake, including.
Mihir Shah – 23:38 Exactly right. So during my career as a chief architect and CTO, I’ve worked with many startups and helping them cross their product roadmaps. But anyway, so I came across a new term somebody introduced me saying it’s your post corporate career. When you quit, you’re not really retired, you move on to a portfolio career. So what you do is you build a portfolio and it’s up to you how you want to build that portfolio and how many hours would that take. But essentially my portfolio is a couple of VC firms which keeps me engaged with what’s coming next, where the money is flowing, plus a lot of young people that you have to interact with. I work with EY as a resident advisor that directly leverages my 30 years of experience as we go to clients on different projects and things like that.
Then I work with a bunch of startups including Promethium. Thanks for having me on as an advisor. And that’s where you get really engaged with the product and a little bit in more detail than you would as a VC. So.
Prat Moghe, CEO, Promethium – 24:50 And you’ve been helping startups. I’ve noticed like there are cutting edge, best of breed in different parts, right, Including Reltio, including Skyflow.
Mihir Shah – 24:59 That’s right. They’re all Skyflow from Lithium. There’s a bunch of companies that I’ve been working with. I would say that this is still, it’s only been 12 months, right. So we’ll see where this goes.
Prat Moghe, CEO, Promethium – 25:12 But it’s a super exciting time, I think. I’m glad that you’re leveraging your experience. Any last parting words of advice to particularly to chief data officer CTOs that we are seeing that particularly there is a group of chief data and analytics officers Obviously there are CIOs and CTOs but then there’s Chief AI Officers. Anything you can talk about?
Mihir Shah – 25:34 All I can say is that in my 30 years bringing data to the forefront and being able to articulate the importance of data foundation we have fought that battle for decades now everybody’s on this bandwagon. This is our time.
Prat Moghe, CEO, Promethium – 25:52 This is our time.
Mihir Shah – 25:53 Take the advantage. Everybody wants to do this.
Prat Moghe, CEO, Promethium – 25:57 You’re front of the bus for once.
Mihir Shah – 25:59 Exactly in front of us. So this is a great opportunity for CDOs and chief architects and data engineers et cetera this is our time so you have to take this opportunity to.
Prat Moghe, CEO, Promethium – 26:10 Get it right and top do’s versus.
Mihir Shah – 26:13 Don’ts think big think data up not function down that’s a more durable foundation strategy and think across your business units not in silos and use cases and.
Prat Moghe, CEO, Promethium – 26:29 Keep it simple and keep it simple. Awesome. Thanks again, really appreciate you being on.
SHARE THIS:
