NODES AI 2026 - Building a Customer-Intelligence Brain: How GraphRAG Turns Data into Decisions

In this talk, Ekar Burma presents the concept of transforming qualitative customer data into a knowledge graph, which addresses the complexities of human behavior in customer journeys. He explains that traditional qualitative data, often relegated to unstructured formats, lacks accessibility and usability. By utilizing graph technology, the speaker demonstrates how to preserve causal structures and relationships within customer data, enabling clearer insights into motivations and decision-making processes. The session outlines a structured method for collecting and converting qualitative interviews into a graph that can be queried for actionable intelligence. Throughout, Ekar also highlights the roles of AI and a living evidence layer in enhancing the usability of customer insights, ultimately advocating for the ongoing evolution of data management in organizations.

[music] This is one customer interview transformed [snorts] into a knowledge graph. It's complex. Human beings are very complex. It covers four journey phases. There are several hundred of nodes and relationships with them preserved in a in a graph. This particular journey was successful. Every emotion was satisfied. Every concern addressed but just one missed anxiety or one missing incentive and the whole journey could have failed. The data

was there. But before we had the graph, nobody could see it. Now with graph technology, qualitative data can be digitized in raw format and that enables a whole new suite of applications. Thanks for joining me for my session on building a customer intelligence brain. My name is Ekar Burma and I'm the creator of the Wheat of Progress and the customer progress design methodology and I'm also co-managing

director of Uniro Solutions. Within the next 30 minutes, I will guide you through our journey to building this customer intelligence knowledge graph. I will first talk about the problem. What are we trying to solve here? Then I will show you the process how we got to building the graph. I have some examples of outputs that we were created that we were able to create and also we'll

take a look at the graph itself. So it's kind of looking as a brain surgeon uh within the brain and see uh kind of the mechanism how it works. So let's jump into the issue. For those of you who are not familiar with qualitative customer data, this this is the kind of data that gets produced by market research departments by stakeholder groups within the company that need

to make strategic decisions like planning the next product or marketing people using qualitative data to produce marketing campaigns or a narrative or any kind of messaging. It's produced to better understand customers, how they think, what they feel, what they want to accomplish. The problem today is qualitative data is not being looked at as real granular kind of itemized data and connected data. It's being looked at as

transcripts that may be chunked up in um in pieces and where they the content is tagged and that's about it today. How qualitative data is used. But when this kind of information stays in this kind of format, it's not really usable. It's not usable to its fullest extent. People usually use AI to summarize qualitative data to get like the gist out of it but they lose a

causal details unfortunately. Uh then still people need to interpret the data. They bring in their own biases. It's not machine readable. So you cannot really easily feed AI to produce outcomes and it's very difficult to trace any decisions any outcomes back to the original source of information coming from that qualitative data piece. So the result is that only a fraction of what was ever captured which is

in transcripts in powerpoints in people's heads in the company really usable. It's very unfortunate. It's a I would say it's a huge waste of precious money and resources today. And the opportunity we see with a knowledge graph with bringing that qualitative data into a knowledge graph is just tremendous. It's we can preserve causal structures. We can query the graph uh traverse it. We can find like the

causes of behavior. Why people bought? Why people didn't buy? What people were looking for? where they got stuck in the customer journey and we can use AI agents to access the data and just to produce outstanding new deliverables. So this is where we think the future lies. It's making it's turning this qualitative data into uh a graph and then making it full accessible. How did we get

here? We did not wake up in the morning and and said, "Okay, let's create a graph out of our data." No, in 2019, we actually started the process of itemizing the qualitative data from inter from customer interviews. So interviews where people talk about customers talk about their experience, buying a product, the whole search process, the experience phase and everything in between. And this is a rich source

of really nice information. And when we structure it, we can determine the value of of those interviews and we can access data directly and respond directly uh to categories of meaning and that's really nice and uh in uh as I said in 2019 we started the whole process of itemizing this data of of creating the rules uh of creating our mental model and later on the data

model which became our ontology. Now this process of interviewing and then evaluating the interview using for example our wheel of progress on on on the right hand side is a lot of work for human beings. It's a great exercise. you roll up your sleeves, um you dive into the data, you internalize it, but it's it's a lot of work, especially when you try later on to um

aggregate like 10 different interviews, you're easily working with uh a thousand if not more data points. So now imagine when we have like dozens or even hundreds of interviews or other sources of qualitative data, this becomes really a a huge task. It's almost not doable by human beings. So in 2023, we started using an AI solution that helps us analyze the interviews to categorize each statement that

we heard into like an element, an element that became part of the the whole story of people trying to make progress in their lives or in their work. And this was already a tremendous help. uh because we could process a lot of data very quickly and very reliably. But one thing was missing. So in 2025 we discovered graph technology as the missing piece of our like solution

stack. And the wonderful thing about graph as you all know is that we can connect the dots. We can connect like a a pain that a customer feels to an action. We can connect that to a desired outcome, what people are looking for. So we can create this whole almost like neural network of nodes and edges that that connect in a container a customer interview. It's a

whole different level of intelligence than a let's say a transcript that is tagged maybe or is going to thrown into AI to try to make sense out of it. So we have tested this ontology for many years with client projects and we have very stringent um very stringent definitions of the items. So they are mutually exclusive. We have definitions on our relations that that connect the dots

and we feel very confident that we have a really valid ontology for this customer progress domain if you if you will. So it's been a long journey and uh of course there's some IP in the process. Uh but I'm going to show you some of the things that that we do. Um but I cannot of course I cannot show you all the nodes and all the relations

because as I said that that is part of our of our company asset. So how do we build the graph? Um of course we start with collecting qualitative interviews with customers. And by the way it could also be done with employees or with partners in your ecosystem. It doesn't need to be necessarily customers. I mean, we chose customers because this is what everyone wants to understand. How

do customers tick and uh so we do these qualitative interviews. We have a very structured way um to ask questions. Um we want to understand the whole journey of course and then we use a transcript of that interview and then we have a process where we identify these items. That's step number two. What we call the CPD structuring, customer progress design structuring into 12 elements. But that's

not enough. As we know, we have to create uh connections between the data items and this is going to be done in step number three where the triples are created. It's a very thoughtout process because whatever relation we connect uh we create needs to be valid. It cannot be just connect let's connect the dots anything with anything. Uh it needs to be very very thoughtful because we

are talking about human beings. uh we need to have differentiated relations between the dots and then once we have built all the triples uh we equip uh the data also with properties uh that later on will help us to even better connect the dots uh to do queries more effectively. Um, so it's not just the data and the evidence, it's uh meta data, it's uh timestamps, it's

uh kind of demographic data where the interview came from and and so on. And by the way, uh since this data is so sensitive because it's about human beings and sometimes they share very personal stories, very personal issues, um we give people a sodonym. So the people in our database are not that's not their their real names and we have a translation table uh so we can

if necessary find out who said what. So in the last step, this is to me the most exciting one is using agents uh to access the data via MCP. [clears throat] And this has been really a breakthrough because this enables us to make the graph database available and usable really usable to people who are not experts. People like marketing people, product people, planning people, even sales people

who are not that technical who cannot query uh graph databases uh to use a simple box and ask a question and define their problem to get the output they need. So as you can see here at the bottom part of the screen, we have certain node types. As I said, they are very strictly um defined. Um we have edges, I think almost like 40 different types to

get the right of uh granularity in these and um that's that's what we are working with. So um I already talked enough about the process and why it's useful to put qualitative data in in a graph. Now let's take a look at an example. [snorts] So last year I started to um interview people who are part of the um an experience management global collaborative. This is a

nonprofit organization and uh I have permission to share the data. So um we I interviewed nine people from around the world and how this is going to look like in a graph. I'm going to show you next um and uh can zoom into the causal relationship. So you can see it's uh these are just a few hundred I think 500 nodes here. I'm going to zoom into

an interview with someone I did with uh someone from the US. I hope you can see these different uh types these different dots. Um and you can see how the relationships work. So we have at the b at the bottom right hand side we have a pain. So someone feels a pain. I'm going to click on it so you can see what the pain is. So the

current job role u may be diminished by AI. someone who's just afraid, okay, there's AI coming around the corner and I may lose my job. That creates actually a push [snorts] and the push is a force that moves people to the next step in the customer journey. So there may be lots of pain points building up, building up, building up, and that creates that push. Um but

there are also some other pushes that don't have a relationship to a pain. But we we've heard that um that push as you can see um we have uh lots of uh properties connected even embedding vector because we have to look across interfues and qualitative data um to being able to compare them to collapse them to find similarities. uh so not not only within one interview but

we want of course um finding the same pains in other people's interviews and you can all can also see I'm going to scroll in a little bit more how these pushes pain uh motivate a desired outcome desired outcome is a goal someone wants to achieve yeah sometimes it's we don't know the product we don't know the solution but we know we want to have a better state

in the future And that also could be in reinforced by a pull. I'm going to click on that. So a pull is a is a force that pulls you over. The pushes is just kind of pushing. Pulling is something that's very attractive. So you're looking at a solution and you see, okay, that's interesting and it pulls me over. So this is just one one example out of

this whole kind of system of uh relations and notes uh how an interview could look like. Now what's the value of of all of this? I'm going to switch back uh to the presentation. And there was really one interesting fact that we found out after we put the data into the graph and played around with agents. Uh and there was one one thing that I want to

talk a little bit about that we discovered that um people in this profession experience management profession across the world they want to learn from each other. Before we thought it's like the developing countries learning from the developed countries. So people in let's say countries where experience management is not that popular or not that professional in the time they want to learn from people in the US in

maybe parts of Europe and uh but through the analysis we found that the opposite is true as well. So people in developed countries appreciate that global perspective. So they can exchange yeah experiences with people from around the world. They want to learn also from them. That was a really interesting fact that we found out after we looked at the data and after we analyzed the data and

and that shows us it's it's great to have that kind of data in in a graph because we can invalidate your assumptions. So let's move on to some other examples related to outcome. And I promised a live demo. And I'm going to switch over to um to Claude. And let me switch screens here in a minute. And we'll do a live query. Uh that takes uh usually

a little bit. Need to switch to another window. I hope this is the right one. Is this the right one? Yeah, it's the right one. And I prepared a yeah a query um that should help us to understand why people joined the organization. So it needs to go into the graph figure out what were all the kind of pushes um that people had in their data that

help us understand why they why they joined. So I'm going to let this run. Um it's a live demo so I don't know how well it works. Uh, and um, we'll come back. I'll let it run. Maybe maybe run just a minute or so. And I want to show you some um, examples that we created using Claude Pro. And the first one is a I think you

can you can see it is uh, a prioritization heat map. So something product managers would be naturally interested in. So this is all data from the graph and dynamically uh someone could create this outcome, this output um that's usable. It's customized. It's not it doesn't require a software. They they build so to speak their own solution, their own software. It's even interactive. You can click here on

any of these uh tiles and get uh like the business impact like evidence from the graph uh with member folds with recommended actions and here at the bottom you have a like an overview of of pain profiles by by different people. So it's very usable for people who need to decide what to fix next. The next one is really amazing, an amazing capability uh that I haven't

seen before. It's uh clustering customer data in very innovative ways. It's not the typical thing by let's say by demographics like white, female, 34 years, two kids, single mom, whatever. No, it's about it's based on their needs on their jobs to be done on their with combined maybe with constraints and this is what we created here. It's different categories of customers in a in a new way.

The third one third example is a support plan for an individual. So we had as I said members joining the organizations and we can easily it's like at the click of a button create a support plan in this case for Christina from Switzerland. Uh here's a portfolio of solutions. So of course we have to educate um the u the agent uh with our capabilities. It created a

12 months journey timeline uh 9point action plan. And the next day when I sent this to Christina, she said, "Where can I sign up? This is so great. This is so customized. It's exactly responding to my needs." And uh that was that was really amazing to to hear that from her. All right, let's uh go back to our uh to our cloud application and see what came

out of and we can see this is what's got uh created. Uh so what pushes are people um pushing to become members? And we can even click on that chart and you can see the evidence coming up here and we have an explanation. We have a summary why this why this happened. So I wanted to show you this is real. This is not something uh that that

cannot that's like uh invented or so. It's it's real data. uh an unreal output and I I guess you don't need many other solutions that you have used before because you you can create them on the fly. So let's jump and um look at kind of the bigger picture a little bit. So you saw these three examples and this is our vision is helping organizations build these

customer intelligence brains how we call them. It's a living evidence layer. It's all the data they take in and saving it as a as a graph so anyone can access it. We we have done it with interviews but we are sure it can be done with other contextbridge qualitative data as well maybe with focus groups maybe with support and CRM uh records uh with review and feedback

um important is to us it's contextri and uh we store it in a in a neoforj graph database which becomes the brain and then it can be accessed by different stakeholder groups to create whatever outcome they want. You can see that at at the bottom. The more interviews you add, the bigger your brain grows. And uh the I I think the value proposition for this one is

also uh doing something and not just analyzing the data. So three things we are coming towards the end of the presentation. I want you to to take away. One is uh yeah qualitative data is one of the most underutilized assets in any organization today. That's after we have done um this kind of work we discovered there is we can do we can create much more value out

of qualitative data. The graph is to us the ideal data structure because it preserves the causality because we we cannot only store data but we can store connected data. It becomes like a bakedin intelligence. And the third one is is is a doing. We cannot only use the data to analyze like in a project you do market research project and then you think about it, you discuss

it and then you file it and then it's gone. No, this becomes a living asset for any company and uh can be updated uh can be extended and and so on. So um with that we have a website unimprossolutions.com where we have a simulation so you can simulate different uh queries. Um it's a simulation but the outcomes are real and um and don't be afraid to get

in contact uh we always appreciate feedback and also discussing possible your issues. And uh now we have a little bit of time for Q&A. So I'm looking uh might it give more weight to nine interviews than it should? The data is not extensive. No, the data is not extensive. Uh it's the more of course the more interviews we do the more the the richer the data becomes.

Uh but it's evidence it's how people perceive their reality. So I did this nine interviews because that was all the time I had but we are already working on a system how to systematically take in more interviews and automate the the whole process. Yeah. Any other questions? No. Is it advisable to use an LLM to create relationships? Uh that's how we do it. That's how we do

it. We have a we have a very dedicated process how to create the relationships. We call this the triple building. That was I guess uh step number three in our process and uh this is something we have done as human beings before building these triples on white boards the little white boards. Um but yes this is this is how we do it. Stephen has another question. No

that's it. Amazing presentation. Thank you very much. Thank you. Um, how did you go from interview to graph or is that the I cannot read this question unfortunately. Conrad, I cannot uh it's somehow cut off but thanks for the feedback. I'm happy to stay on and discuss it a little bit later. Would this work? Have you tried this with survey data with open-ended responses? Yes, we have

done it with uh not not really sur um survey data but with um an uh an an uh survey tool where people speak their responses because we like people to speak and not to think too much and write it down and overthink it. We want spontaneous responses and uh we have described um or the engine describe transcribes uh the verbal comments and then we take we we

took this in in in in our graph. Yeah, that's that's possible. Do you use human interfer? Yes, we do. We like to do still like to do that. Um if so that's because you get better Well, I cannot say that with certainty, but certainly we we like to interview people uh right now. Um yeah, also testing AI solutions uh to to do this more dynamically. But um

if people spend um doing this uh our customers, we we want to appreciate that and also spend time with them. And I think it's it's kind of a human to human thing which doesn't mean in future uh we'll have uh AI uh more AI based solutions. So we are overrunning unfortunately. Um how do you consolidate customer journeys across different organizations? of course they would need to be

part of the data and as I said we are using uh vector embeddings um to find similarities across uh all these interfues. I hope that that answers the question. Uh what else is needed to go from text to graph aside from an ontology? Well, that that's basically our process uh our five-step process. uh it's there are many many steps I can tell you with human in the

loop uh to to make sure the data is accurate and valid um but it's it's it's quite quite some work uh involved in in doing that but we are we are working on a fully automated solution all right thanks so much um Hana do you have any closing words No. All right. Then I want to saying say thank you to everyone. I hope this was inspiring and

uh would love to talk if you are interested. Thanks very much.