HFS & MathCo: AI models are a utility, context is your identity

Top three takeaways

• AI commoditises the ability to produce answers, so the advantage moves to what makes an answer good: the enterprise’s own knowledge of how it works, what decisions it makes, and what data it holds. Anuj’s argument is that organizations are absorbed by the models and under-invested in who they actually are.
• Context has no owner today because it is not a line item. The old semantic layer covered data supply and sat with the data organization; context also has to cover demand — the questions being asked, the decisions being made, and the impact of those decisions, including logging them. He expects a federated model with context stewards by business function, and warns that context is a river rather than a photograph: it needs feedback loops that push down what he calls verification cost as agents multiply.
• Rent the tooling, own the knowledge. Three routes in: a full enterprise context program; a vertical slice such as marketing, connecting its processes, questions, data and metrics into an ontology; or a horizontal slice over an existing corpus like Confluence pages or years of strategy decks. Whichever route, build it to expand. His one wish is that data leadership treats context as a formal program now rather than a decade too late.

Saurabh Gupta — President, Research and Advisory Services, HFS Research[00:21]

Hi everyone, welcome to another episode of HFS Unfiltered. Today, we are going to be talking about how to make this promise of AI real, and to do that, I have none other than Anuj, who is the co-founder of MathCo. So, Anuj, welcome to the show.

Anuj Krishna — Co-founder and President, Technology and Growth, MathCo[00:39]

Thank you so much.

Saurabh Gupta — President, Research and Advisory Services, HFS Research[00:40]

So, Anuj, you know I’ve had so many conversations with enterprise leaders across the Global 2000, and everybody has lofty AI ambitions. You know AI is going to solve for world hunger practically but in reality there’s almost like a death by a thousand POCs. You know very few enterprises if any are able to scale those POCs into production and scale it up and a lot of it comes down to I don’t know where my data is, all of my knowledge is in somebody’s head somewhere, you know my processes are not documented. So do you do you agree with that’s what’s happening?

Anuj Krishna — Co-founder and President, Technology and Growth, MathCo[01:24]

Yeah, I think the promise of AI has been there for some time and really what is missing in enterprises right now is the idea of how do we bring all of this knowledge that already exists within the enterprise to bear to the AI. I mean if you think about AI, AI is really making your ability to provide answers into a much more commoditized fashion. But for those answers to be given, it needs to have the right knowledge behind it, the right context behind it to be able to do that.

And nowadays at least, you know, the word context has become the buzzword. And if you really think about what context means, it is it is what is it that you know within your organization? How do you work? What are the decisions you make? What are the what is the data that you have that can help with answering some of the questions that AI can help answer. So that’s really what context is and a lot of organizations are talking and thinking more about the models themselves. But what they should really be thinking about and what is real business advantage to them is how they operate and who they are and what their knowledge is. And it’s been like that you know for a long time anyway but now it’s become even more prominent with where we are as an ecosystem with respect to AI and modeling and models and so on.

Saurabh Gupta — President, Research and Advisory Services, HFS Research[02:56]

Yeah. So, Anuj, we did this research together, and almost two-thirds of the enterprises said yeah this is a problem. I don’t know where all this resides. But the but the point is context is not a line item in the you know in our enterprise dictionary, so who owns this?

Anuj Krishna — Co-founder and President, Technology and Growth, MathCo[03:14]

Yeah I think that’s a big question and I think the governance of context has become a big question now. Over time typically if you look at if you look at the you know what used to be earlier called the semantic layer with data, it used to sit with the data organization. It is but that’s all about what data you already have. When you think context, you need to start thinking about what are the questions that data is trying to answer. What are the decisions that you as an organization are trying to make? As you said, business process, what is the real business process that you’re following? And more importantly when you make a decision what is the impact of that decision? How do you make sure that decision itself gets logged? These kind of things don’t sit within that semantic layer framework that we have typically seen.

And honestly, we are going to move towards a at least my version of this is that we’re going to move more towards a federated structure where while there is a lot of centrality in some of these some of these supply of knowledge related questions like you know what is the metric, what is the data and so on, the more demand related metrics which is what is a question and what is the what is a good answer to that question and what is the impact of a decision that you’ve made is going to end up being more federated by business function. You will have context stewards who are just like you had data stewards in the past who are going to be necessary to make sure that the context is relevant.

I think that’s the other big thing that organizations need to get right is it’s not just about you know get keeping context static, right? It’s about how do you make sure it’s a living thing that is growing over time. The analogy is that you know you take a picture of a river and it’s great as a picture but the next moment the river has changed. And context is a little bit like that. So how do we how do we make sure that it stays relevant is going to be one of the big questions that needs to be answered and the governance around it hence becomes extremely critical. What is the feedback loop? How do we enable it in the right way? How do we make sure that what I like to call verification cost, which is the amount of times that you need to verify that the context is right, doesn’t keep increasing? Because as you go you’re going to have agents flying all over the place and they’re all going to be making choices and you don’t want you to keep verifying each of those choices. How do you make sure that keeps going down? I think those challenges are be going to become more and more.

The reason I think it’s not been you know not been a line item in the past is because you’ve seen semantic layer be a line item but nobody has dealt with the rest of the rest of the you know what you call knowledge of an organization which is the business process, all of these things, and unless we start treating that as its own thing I don’t think that whole connection is going to happen across all of it.

In some organizations I’ve seen that the central organization takes up the mantle of doing it, of capturing that as well. In some organizations I’ve seen that the business starts doing its version of it and then connects it back to what the what the central organization is doing in terms of data. So I’ve seen both versions of it. But unless you make that start somewhere and create the map you will not move forward in that journey of creating product.

Saurabh Gupta — President, Research and Advisory Services, HFS Research[06:52]

So Anuj to some of the enterprise leaders who are listening to this, how do they get started because this feels like a scary hairy problem.

Anuj Krishna — Co-founder and President, Technology and Growth, MathCo[07:03]

I think the first thing is that I think there is already a lot of document documentation, most likely structuring of those documents that already exists within organizations. The question first is how do we leverage that effectively? The second thing is that as you as we all know we live in the world of AI. How can we use AI to speed up the process of context creation and collection is the other question that we really need to be asking. And it’s not just about putting a framework for it, but is there really something that you have that is going to help speed that process up is the other question that we need to be asking.

We’ve seen it multiple ways. One, some organizations take up a whole program around context and say I’m going to build context out, right, and you know take that up as a challenge and a mantle knowing that it’s going to be the business advantage of the future but take that mantle. Some organizations have said I’ll take a slice of the business and a set of decisions, could be a function like say marketing as a function, and say all the business processes within marketing, the questions within marketing, the data related to marketing, the metrics related to marketing. I’m going to connect all of them into an ontology that makes sense and that can be used as context. Some folks have said that I’m going to look at it as a horizontal slice which is more about take a corpus of knowledge. Let’s say all your Confluence pages or all of your decks or presentations that you’ve made about strategy and brand strategy over the years. How can I take that and create a corpus of knowledge context around it?

So I think depending on who is the person going down that route, you can go by either of these ways. The one call out I would make though is it’s important to in any of these ways is to think of the system overall and create it in such a way that it is expandable as you keep going above the scope of what is there at any given scope. You have to make sure that the system is robust. So it’s not just about you know, just the P right.

Saurabh Gupta — President, Research and Advisory Services, HFS Research[09:23]

So Anuj, you mentioned context has become the buzzword right now and I’ve seen you know companies like Databricks saying we are we are going to be the context kings. You on the other hand you have a lot of IT service providers and GSI saying look we’ll be the context kings. Yeah. And you have companies like yourself right MathCo and you know I would call them almost like AI native companies. Yeah. Who are coming in and saying look you need to look at look at us. Yeah. So tell us how is your approach different than what’s being shared in the market.

Anuj Krishna — Co-founder and President, Technology and Growth, MathCo[09:59]

That’s a great question. So firstly I think the platform owner, platform owners like a Databricks or any of these — and we are great partners with Databricks as well — they have the features that you need to be able to pull context off. But what and I keep saying this and our tagline as a company is own your intelligence. You can’t own your intelligence unless you own your knowledge. So the first thing I want to call out is that irrespective of who’s building it, I think the important thing is for an enterprise to make sure that the context belongs to them, and for that to happen, it’s important to use the tech ecosystem around you as a renting model but the owning model has to be yours.

So we come at it from the perspective of hey I’m going to bring those accelerators whether it’s in Databricks or any other tech platform, I’m going to bring my knowledge of what I already have in a domain, I’m going to bring my ability to as agents to pull information out, but I want to create this in such a way so that it’s yours. I look at it almost like a it’s like a like each of us have our own personalities and the context and the knowledge of an organization is its personality. So how can we create that personality for the organization needs to be custom to that organization. You can use all the tools you want, you can use all the agents you want and you can do anything you want but the end goal has to be how can we create that for that organization. So that’s at least our approach towards doing. I don’t know if who will win that game. But we feel strongly that unless you’re approaching it with the mindset of giving them the knowledge, it’ll fail.

Saurabh Gupta — President, Research and Advisory Services, HFS Research[11:55]

Yeah. Know that makes a that makes a lot of sense. It’s basically context is your identity. Yeah. That has to be your source of competitive differentiation. Yeah. And you can’t have that reside somewhere else. Yes. You have to own it. So that makes a lot of sense. So, Anuj, before I let you go, look, this entire market is in a state of flux right now. So as you look at this space, you know AI services, data, context ambitions, what’s your one wish? You know, if I could grant you one wish that could come true, what would that be?

Anuj Krishna — Co-founder and President, Technology and Growth, MathCo[12:34]

One wish that could come true. I mean, I have so many of them, I’ll just stick to one. Yeah, but I think the one wish that I want to come true is for organizations to start thinking about clean program for context. There has to be a context program. I mean, given related to this conversation. I mean, I can talk about other wishes which were not related but for this conversation I think central like a data co right a central data or a CEO. has to think seriously about, okay can I create a program now for making sure that my next decade is set up right, and unless they start that journey now it I think will become a problem later for them. So I think my one wish is at least let that realization land and let that let that become a program.

Saurabh Gupta — President, Research and Advisory Services, HFS Research[13:32]

Yeah, no, it’s been a fantastic conversation. I think my big takeaway is look I think the LLMs, then the model layers will become a utility. There’s not going to be a competitive differentiation there. Building the agents is also going to be easy. You know, you can’t differentiate on that. But context, as Anuj rightly mentioned, is your identity. You’ve got to own it, and that’s going to be the source of sustainable competitive differentiation going forward. So Anuj, thanks a lot. I enjoyed a lot.

Anuj Krishna — Co-founder and President, Technology and Growth, MathCo[14:00]

Thank you so much.

Saurabh Gupta — President, Research and Advisory Services, HFS Research[14:02]

Thank you for taking some time out.

Anuj Krishna — Co-founder and President, Technology and Growth, MathCo[14:04]

Yeah. Thank you. Yeah.