Episode: Why Data Modernization Is the Foundation of Enterprise AI

Prasanna Vaidya (0:05)

Welcome to GydeBites, I am Prasanna, your host. You can't build intelligent enterprise on the data that enterprise itself cannot trust or cannot understand. Yet, that's still the reality across many organizations, and as AI is moving into a lot of business workflows, the data gaps are actually becoming harder to ignore. If you don't have good data, the AI will not work for you. And that's what the data modernization, that's why the data modernization becomes a critical foundation for enterprise AI.

It ensures AI can work with right data, right context, knowledge across your business. We are going to explore this today with Sourabh Mukherjee. He is head of AI and data strategy at Accenture. He has nearly three decades of experience in data analytics and AI.

Sourabh has helped global enterprises harness their data for their business growth. He also specializes in AI strategy, data modernization, and large-scale transformation initiatives. He is a published author, he is a keynote speaker, and an academic advisor. Sourabh, welcome to GydeBites. It is great to have you.

Sourabh Mukherjee (1:20)

Thank you, Prasanna. Thanks for the wonderful introduction. Our warm reception, and I'm very happy to be here today. So look forward to an engaging discussion.

Prasanna Vaidya (1:29)

So let's get this rolling. What is data modernization, and why it is so important for enterprise AI?

Sourabh Mukherjee (1:35)

Right. I think that's a great question, and I'll try and answer it by taking you through the layers of an architecture. So essentially, if we start from the bottom, so let's first look at infrastructure. And what we see in a lot of enterprises today is that the critical enterprise data is locked in legacy applications from specific vendors, in apps, in spreadsheets, and even in the hard drives of people. Then there are multiple ingestion parts of data, which means that the source of truth for data is actually not very clear. And there is low trust in the quality of the data, which means that there's a lot of redundant cleansing and remediation of the data, which means that you are beginning to build AI on a foundation that nobody can really vouch for. Now, if I go one layer higher, now if I look at the data platform itself, then what I typically see with a lot of enterprises is that the data pipelines are actually bespoke rather than being reusable and interoperable.

Prasanna Vaidya (2:38)

Right, right.

Sourabh Mukherjee (2:38)

Then the other challenge that we see is that the ability to handle multimodal data, you know, like unstructured data or, you know, semi-structured data, synthetic data, is still very limited in a lot of enterprises, which means that there is a limitation on how broadly and how quickly you would be able to train your AI models. Now, going to the next level, you know, if you really now look at the data sets specifically, what you would see is that, and this is a common observation across multiple organizations that I have interacted with, that there are lots of data sets without quality guarantees or built-in access control. There's no clarity around the lineage of the data, which means that the AI agents, when they are trying to leverage these data sets or the data products, will not have clarity around what data sets or what data products can be trusted or which ones are secure, in order to generate the kind of outcomes that we expect the agents to deliver. And finally, when I come to the consumption part of it, when I look at advanced analytics and when I look at artificial intelligence, what we are seeing is that a lot of organizations do not have a very standardized business taxonomy or a semantic layer. There's very limited use of knowledge graphs across organizations. What that means is that your AI models are now beginning to reason over ambiguous data, right? So data which doesn't have unambiguous meaning, instead of actually having a shared enterprise context, which is absolutely essential for the agents to come up with trustworthy outcomes. And cutting across all of these, you would find that governance policies and standards are not consistently enforced. There is lack of clarity around ownership of the data. And as we go up this whole hierarchy that I sort of talked about, you would realize that the unresolved problems will sort of get compounded as you go higher up across all of these levels. This is where there is a tremendous opportunity for modernization of the data platform and the foundational data capabilities for agent AI to be effective.

Prasanna Vaidya (4:38)

Yeah. And let's talk a little bit about approaches, right? So what sort of modernization approaches are delivering best outcomes when you consider this AI era? Because everyone is trying to build AI agents, AI systems, AI solutions, and so on.