Building a Trusted Data Foundation for Scalable Enterprise AI Presented by OpenText | Utilizing AI Episode 36

Enterprises want to leverage AI to build trusted systems grounded in their client data to provide context and reliability. This episode of "Utilizing AI" features Waqas Ahmed, VP of AI Development at episode sponsor OpenText, and Futurum VP and Research Director Keith Kirkpatrick discussing trusted data foundations for AI applications. Welcome to "Utilizing AI," the podcast focused on practical applications of artificial intelligence from the Futurum Group.

Every Wednesday, we explore news and use cases of the ways in which AI is transforming enterprise IT and the industries it serves. I'm your host, Stephen Foskett, president of the Tech Field Day business unit here at the Futurum Group. Before we dive into this special episode brought to you by OpenText, let's meet who's on the panel today.

Hi, my name is Waqas Ahmed. I am the Vice President of AI Development and Enablement at OpenText, and our function is to act as a center of excellence and partner with our development business units to lead AI innovation in our products and unlock value for our customers. Hi, I'm Keith Kirkpatrick, Vice President and Research Director at the Futurum Group, and I cover enterprise software and digital workflows.

" Part of that gives me the opportunity to run our AI Field Day events, and this, of course, gets me in touch with the great world of enterprise AI applications and business data. And a very frequent recurring theme I hear from enterprises trying to deploy AI-powered applications is the topic of trust and value. Essentially, all of these people, they want to deploy AI applications.

They see the value in AI, but they're very concerned that, because it's everywhere, everybody hears this in the world, that these systems can be unreliable, that they can give you untrustworthy responses. And of course, that's a catastrophe in an enterprise workload. So let's start off, Waqas, I'm going to start with you.

Tell us a little bit more about the connection between a solid data foundation and a valuable AI application. Certainly, Stephen. And now I am essentially stating the obvious because it's so well understood that the large language models have the world knowledge, but they do not know anything about the information, the context, and processes inside your enterprise.

So they need that data to give you the answers. But that doesn't stop there, because what does having the data access even mean? Can you simply aggregate that in one giant data store and then say, "I'm successful"?

No. That will have both the context management problem, security problem, and the answer reliability and effectiveness problem. So what you need to do is exercise this practice of making the right data available at the right time, so your answers can be grounded in truth, but also ensuring that the information is properly governed, secured.

And during the processing of that information, whether you're giving answers or you're taking actions, you're not violating your own guardrails, your restrictions, your processes, compliance, et cetera. So doing all that in a secure and safe way at scale is what data management for AI is all about. The trust part of it.

If you do a data grounding exercise and are able to get consistent, reliable answers, you still have to ensure that that trust does not break over time. So you have to be able to ensure that you can tune the data, that you can measure the outcomes, whether they are drifting over time, whether they are producing the actual value that was promised to you, and all that foundational elements. Understanding the truthfulness of the answer, understanding whether they are accomplishing the outcomes, understanding if they are changing their behavior over time, and being able to track the provenance of those actions.

All that is what leads to you having trust in your own AI solutions. Because short of that, all we can see is demos that are successful because at the demo time, you put everything together for your demo. But in practice, at scale, you have to have applications and discipline to do that for your AI solutions at large.

Waqas, you bring up some really good points there. I think one of the things that we've been looking at here over at Futurum Group is really thinking about how are organizations measuring the success of AI, and that's ultimately about driving business outcomes. You cannot drive solid business outcomes unless you have all of those elements that you were just mentioning there.

But chief among them is that trust that if you're using an AI system, that the data is, A, valid, it is up to date, and it is relevant for the situation or problem that you're trying to address. I think one of the things that we encounter a lot of times with AI is you might be able to query a data point, it might be factually correct, but it may not be relevant to the scenario that you're trying to solve. So I think that that's one of the sort of big stumbling blocks around this.

We've seen all the data out there in terms of the number of pilots that are stalled around AI. It really does go back to understanding and making sure that the data that is being surfaced through AI is accurate, relevant, and up to date. Exactly.

And we see examples of this in lots of solutions that our customers are trying to put into place. So I'll give you two real-world examples. We have a large financial services customer who want to automate and AI-enable their data for document and loan processing in the financial space.

Well, guess what? That document set is all over the place, and the solution is not to simply put that into a single location, because you lose all your controls, your visibility, your tracking, and everything. So that's where we provide solution in the content management space, where first of all, it ensures that you can properly catalog and organize your data while conforming to your own guardrails and processes and access management, and also define the ontologies in that data.

What data is useful in what context? And that's not a SQL lookup. " No, you need to be able to draw relationships.

Which clients and which users have what type of information, how to put this together, and who can see it? And building the tree of related information is where we can enable both feeding the right context to the large language model and ensure that we don't have leakage of security or sensitive information. And this all is connected, of course, to the history of OpenText.

I'm not sure if everybody really understands just how broad the base of OpenText is, but in my career in enterprise storage and data management, I've worked with many of the assets that have now been combined within OpenText. So you have, for example, assets that came from HP, from Documentum, Dell EMC, and all of this, I think, uniquely connects OpenText to the real enterprise data set. And I say the real one because I think a lot of people, when they think data, they think of structured data and databases, but of course, the real value is everywhere else and in everything.

And as you mentioned, Waqas, companies don't necessarily want to have a process of one time going through and organizing everything and collecting everything and building some sort of structured database of everything. What they want is a system that can connect with everything where it is, because where it is is where it's useful and where it's used by the customer and, of course, by their clients. Do I have it right there?

You have it right there, and we actually frame it by saying that you need to bring AI to your data and not data to your AI. Because bringing the data to your AI means that you have lost all meaning, all context, all security, all governance that used to be enforced by the business applications that are producing this data, that are maintaining this data. So in addition to latency concern, you essentially are working against yourself by trying to gather that in such a way.

By bringing AI to your data, it means that you preserve the application and processing context around that structured and unstructured data set. Because when a human knowledge worker works on a business process, whether that's approving a loan application or triaging an incident, you are actually taking into account the information that's in front of you and the related information that is in the past incidents or related documents. " Rather, you want to be able to the AI agent to augment your human knowledge worker and be able to selectively use that information to solve the business problem.

Waqas, you mentioned a really, really great point there about that around the idea of using AI not to just replace people, to really augment workers. Because if we think about the way AI has been positioned in the general press, it's been, oh, AI is going to come for our jobs or come for your jobs. That is not necessarily the best way that it can be utilized by many organizations.

I look at it as being there are certainly processes that will be eventually made essentially extinct over time, but just as we don't have milkmen coming to our door every day delivering milk, there are certain processes that need to go away. But with the help of AI, it can be used to, A, help move people into different roles that will allow them to thrive and help drive that company forward in different ways. And I think we should look at this that every time in human history we have made something more accessible, there was more of it and many more opportunities opened up.

Jobs and profiles changed. They did not get simply eliminated from our day-to-day lives. So in this case, I like to frame it as this way.

AI does not eliminate your job. It elevates your job. You achieve a business outcome, but you are more productive and you are more effective, and you are operating at a higher level then.

One example of that could be in the developers' assistance ecosystem itself. It's not that the action of producing software is going away. Now that writing code is easier, it opens up the door to us to create many more products that were previously not feasible.

So that means the bar of The bar of entry has been lowered, and we'll have many more people employing these AI productivity tools to create many more solutions. So similarly, in the enterprise space, I think what we should think of it as, I'll talk about the OpenText products, our content management, our service management solution. They will elevate the human knowledge worker who are solving those problems right now to be more productive and not be more idle.

So on that note, I think that it's key that one of the abilities that large language models have is to be more flexible in the data that they are able to process. And this has been historically a real challenge when dealing with unstructured data, because essentially it is right there in the name, it's unstructured. And it's been challenging to intelligently extract information from those data sets, whether it is different types of documents, PDFs, images, medical images, all sorts of things.

It's always difficult to pull data out of those, or at least it always has been. But one of the amazing things that's happened in this environment is that these neural networks are able to do things with that sort of data that brings structure to it. So one of the ways, for example, that I use AI, and I think that Keith does the same, is as a research assistant, where you feed it a huge volume of unstructured data, and then you start querying it about the content of that data.

That's something we've never been able to do, and it's been sort of an insurmountable hill for the unstructured data environment. I'm sure that's your entire world, Vikas. And now you have this incredible tool.

It's like having an army of research assistants. Is that right? That is exactly right.

And if you think about it, so I'll build up on your example, that previously if you had to automate some type of process and you're dealing with unstructured data, you would need to use traditional models to extract images, information, put that in the structured data set, and then automate that. And humans, of course, don't work that way. We are able to take a document, look at an image, look at a business document and say, "I need this information.

" That is the high toil work that AI can assist you with, and then you can be more productive in the decision making in the actual business outcome. Vikas, I absolutely agree with what you just said there. The other thing I want to call attention to is one of the things that we're seeing is because you're able to take this unstructured data where there's just a wealth of institutional knowledge, process knowledge, all of the stuff that really helps an organization run, that is really helpful in helping organizations use AI to really target higher value cross-functional use cases.

8%. All of these departments, all of these functions are very, very complex, cross-functional, and require data that is, of course, organized and accessible by many different functions. Yep.

And there are lots of success stories here. We have an offering from our cybersecurity business unit. We call this OpenText Fortify Security Remediation Aviator.

And what that is able to do is to take a look at those mountains of security vulnerability reports that we get from our scans and we're always behind on and identify which are real problems and which ones are false positives, and be able to produce the recommended remediation for the real problems as well to the point where what used to take many days to repair, you essentially can fix them in one hour per issue and with 60% of fewer vulnerabilities going into production. That's a reference data set we are using from where we have deployed it in our clients' environments, and we are seeing real results. And once again, what has changed there?

The data is still the same data. It always existed. But our ability as humans to correlate, to understand, to draw connections, and make judgments on that, that was limited by our scale.

And that's where AI is augmenting us and making our overall solutions more productive. It's really interesting because that's one of the things, Keith, I know that you have talked about. You've talked about the fact that many of the companies in this data management space have been forced to focus on the core challenges of data management, document management- Yep ...

categorization, and yet AI has really opened up the door to allow these companies to move up market. Is that what you see happening here in the market? And I guess it's probably not happening with everybody, but is that what you see happening?

Yeah, I think AI is the great, to use an overused term, unlocker of additional value to the marketplace because it does sort of remove ... that friction of the sort of low-hanging, repetitive toil type work. So it does allow these companies to expand their offerings.

And I think that's one of the things that is most interesting for the end user, which is AI is really making it possible to really look at the entire organization and see how they can actually go out and find these solutions that are going to help them achieve their goals. They're going to have more choice, and ultimately, it's going to result in, hopefully, better outcomes, because you're not going to be limited the way we were before, where it was just a single vendor for one specific function. Yeah.

And another thing that you've said, Keith, and I want to throw this over to Waqas, is that information management is sort of a prerequisite. It's what's required in order to have any kind of effective enterprise AI solutions. So by opening up the door to real information management, not just document management, do you think that OpenText is going to enable your clients to do new things?

Certainly. There are certain business process automations and improvements that were simply inaccessible to our clients. Loan processing is one example.

You need a human knowledge worker to be able to identify the right artifacts from different locations where they might be. And that limits that, okay, if they can only process so many loans, they can only use so many business opportunities. But if we are able to use an AI-assisted automation that brings in the judgment that was lacking there and makes the process more efficient, these type of financial services company now can move upstream.

They can see that, okay, we are now able to process 10 times as many loan applications. We can do better comparisons, and we can do all that while ensuring that this is compliant and secure. So we can now package new type of offerings.

There are new solutions possible because they become more accessible and more feasible because the cost of implementing those solutions has gone down, the cost and complexity of that. So I think we will see a new round of innovation in each of our business domains where certain ideas, innovation that was simply not accessible is more feasible now. Yep, absolutely.

You mentioned some of these new use cases. I think one of the most interesting things that we've seen is the ability of, whether it's your clients or other customers out in the market, is the ability to take these processes and then see them linked throughout an entire workflow chain. So we're not just talking about processing an invoice.

We're talking about being able to use agentic technology to further automate across that entire process, from invoicing to fulfillment, and actually all the way to service. And what that does, that really kind of multiplies the impact of AI across that entire flow, and then ultimately will help that organization achieve whatever its particular outcome that they're seeking, whether it's greater efficiency, greater accuracy, more visibility. And those are certainly important when we look at regulated industries where there's a lot of scrutiny around the use of AI.

And when it comes to making that information available for agentic decision making, of course, context graph comes in. Because when we make those decisions, we don't just need a system of record where we have access to an artifact. We are interested in knowing which client or which other user has acted on it.

What is the related user and activities that have gone on recently? If you are processing an incident, you would want to know what are the changes in the system that have been made recently. All those individual pieces of information do exist.

They have always existed as system of records. What's missing is our ability to build a knowledge graph from this information that allows us to traverse this data intelligently, but without having to statically define it. So this building of knowledge graph that we can be used by our AI agents more effectively allows us to pick just the right information, because as we said earlier, information access does not mean consolidation of information.

It means intelligent selection of information. So these type of techniques allow us to build solutions where you can pick the right data set, which produces repeatable and consistent outcomes that you can trust, and hence it allows you to actually take those solutions to production. And Waqas, what it seems to me, what you're also saying there is this will allow your clients to then actually take meaningful actions, even basically actions that might not have been apparent in the past because they weren't able to sort of segment out the various data, the various insights that were coming from all of this data that they had locked up in the organization.

Exactly. And those additional insights come from the ontologies, the relationships between the data. They also come from the extraction of useful information from unstructured content.

There are images. There is text buried all over. And if you don't know what you're looking for, you're never going to be able to derive those insights.

But by building this AI-assisted context graph and knowledge graph, you actually have a storyboard that you can navigate to derive those insights that weren't possible before. This was one of the topics that you focused on at OpenText World in the fall. I remember seeing some customers on stage and mentions of customers there talking about the ways that they were able to transform their environments.

Do you have any of that that you can share to really talk about what the value that they're getting from AI? Yeah, of course. We have numerous examples, one of which is a large industrial manufacturing customer who have transitioned to our service management solution, and they have employed our AI-assisted ticket processing, and we are seeing 97% accuracy in the results on auto-classification and assignment of tickets to the right teams that need to support those incidents or those service desk requests.

And those are 97% first time across their global estate. And that's a big success story because that is a solution that is working at scale in production, is producing measurable, meaningful outcomes, and that allows them to elevate their problem space higher and now improve other aspects of their business. Keith, if an enterprise buyer was listening to this and they were thinking, "This is the sort of thing that I wish that I could do," what kind of advice would you give them about approaching unstructured data, about unlocking data from the incredible data sets that they have?

How would you help these clients to adjust to this new AI world? Well, I think the main thing to keep in mind with this is you have to look at data as being this resource, this almost untapped resource. You need to have the software, and really also, I would say the expertise on how to manage all this, how to go in through, and again, as Waqas was saying, make sure that you can understand not only what data you have, how it should be classified, how it should be set up to have proper guardrails so the right information at the right time is made available.

And of course, most importantly, making sure that the unstructured data that you're capturing really does... You're casting a wide net to capture all of that institutional knowledge process. And it could be individual assessments that are captured in things like emails, things that are not necessarily working their way into client records.

Making sure you're able to capture all of that data, make sure it's classified properly, and essentially, made available for use where it is most appropriate across all of the entire organization. I think that's really key to making sure that you're able to get the most out of AI, because ultimately, AI is never going to really deliver on its promises unless it has that strong foundation of being able to get that right data at the right time for the right scenario. And I would like to double-click on the right data at the right time in the right scenario, because too often our first instinct is to say information management for AI could mean that I'm going to consolidate all my information.

" And we advise our clients and our partners, by doing so, you're actually losing all the meaning in the content. Is this information safe to be viewed by everybody? It could be sensitive HR information.

It could be sensitive policy information. What are the related data sets that this business process actually relies on? And those pieces of information, that metadata that is both executed and produced when you're using the application, that doesn't exist as a system of record that you can simply export into a giant data lake.

So that's why it's important to understand that you need to build and employ the context-aware solutions that are acting on the business processes where they are executing, as opposed to exporting data someplace else where you have a giant data set but no meaning behind it. I think that one of the things you said earlier in the conversation, Waqas, really brings it home for me, bring the AI to the data instead of bringing the data to the AI. Right.

I think that's absolutely right because ultimately, these applications are supposed to be helpful. They're supposed to be useful. It's not supposed to transform everything under the hood.

It's supposed to transform value. It's supposed to transform everything on the client side, on the employee side. And I think that in order to do that, you have to approach it the way that you've discussed.

And I think that that insight comes from a long history of experience with how enterprises really collect data, how they really organize it, and how they really wish that they could use it. So it's really been an interesting conversation. I know that some folks would love to learn more.

I do recommend checking out OpenText World. I think that's in October. They'll be having that again.

But also, where else can they connect with you and learn more about OpenText and maybe continue the conversation with you personally, Waqas? Of course. com, and we have areas to highlight our AI-related innovations and areas where we provide data management and secure information management solutions.

And myself, I like to interact with people on LinkedIn. Once again, it's Waqas Ahmed at OpenText at LinkedIn. Great, and we will provide your name, of course, in the show notes here.

Keith as well. Sure. " Would love to see you in future episodes.

But where can people find you right now? Sure. Thanks, Steven.

com, and just search for my name, Keith Kirkpatrick. I'm also active on LinkedIn. So again, just look for my name and you'll find all of my content there as well.

Excellent. Well, thank you everyone for listening to "Utilizing AI" today. If you enjoyed this discussion, please do subscribe on YouTube or in your favorite podcast application and give us a rating or a nice review.

This podcast was brought to you by OpenText, as well as the analysts and experts from the Futurum Group, where insights meet AI. ai, the "Utilizing AI" YouTube channel, or the Techstrong TV app. Thanks for listening, and we'll see you next Wednesday.