Tech Transformation with Evan Kirstel: A podcast exploring the latest trends and innovations in the tech industry, and how businesses can leverage them for growth, diving into the world of B2B, discussing strategies, trends, and sharing insights from industry leaders!
With over three decades in telecom and IT, I've mastered the art of transforming social media into a dynamic platform for audience engagement, community building, and establishing thought leadership. My approach isn't about personal brand promotion but about delivering educational and informative content to cultivate a sustainable, long-term business presence. I am the leading content creator in areas like Enterprise AI, UCaaS, CPaaS, CCaaS, Cloud, Telecom, 5G and more!
AI is making decisions that shape real lives, yet most people cannot see how those decisions get made. We sit down with Scott Zoldi, Chief Analytics Officer at FICO, to unpack what “trustworthy AI” actually requires when the stakes include fraud, credit risk, and customer outcomes in heavily regulated financial services. If you have ever wondered why black box models create so much fear and backlash, this conversation puts clear language around the real issues: data provenance, explainable AI, ethical testing, robustness, and the ability to audit a decision after the fact.
We go beyond buzzwords and get specific about AI governance. Scott explains why responsible AI starts with a shared model development standard, so a large organization is not running a hundred different approaches that no one can consistently defend. We talk about why monitoring is often the weakest link in real world machine learning, when to retire models that drift, and why enterprises need to stay in control instead of outsourcing critical decisions to models they did not build.
Then we dig into a practical enforcement mechanism: coupling AI governance with blockchain to create an immutable record of requirements, testing, verification, and release decisions. Think of it as an operating manual that travels with the model and can be inspected years later by regulators, customers, or internal teams. We also look ahead to what changing regulation could mean, including the push toward interpretable models, trust scoring for generative AI, and focused language models or small language models built for narrow tasks with auditable data.
If you care about responsible AI, AI transparency, and building systems people can actually trust, hit play, then subscribe, share this with a friend who works in AI or compliance, and leave a review with the one governance rule you think every model should follow.
Hey everybody, fascinating topic and guest. Today we're talking about FICO and how AI and blockchain and stronger governance could help make critical business decisions more transparent and trustworthy. Scott, how are you? I'm doing great.
SPEAKER_01
How are you?
SPEAKER_00
I'm doing great. Thanks for joining. Uh, really intrigued by the topic. And of course, everyone's heard of FICO here in the U.S. And uh mainly for credit scores, of course. But uh
maybe introduce yourself and what the software side of the business that FICO does.
SPEAKER_01
Yeah, pleased to do so. So FICO is uh is a is well known for the FICO score, but a larger part of our business is actually the fact that we're an analytic uh software company. Um we were one of the early companies to pioneer the use of machine learning and AI in software applications. Uh the very first one of these applications was one called Falcon in the year 1992, which put a profile-based machine learning model at the center of making decisions about whether or not fraud was occurring on a credit card. And since that time, we've continued to advance AI and machine learning associated with our the markets that we operate in, primarily financial services. Um, my role is I'm chief analytics officer at FICO. I'm responsible for our AI direction, the types of algorithms we use, and how we make sure that when we build AI to make decisions on customers, that it's the right AI and things that we can uh explain and uh and and audit.
SPEAKER_00
Brilliant. Uh and beyond uh AI, we'll dive into your approach. What are the sort of business problems that you're helping customers solve?
SPEAKER_01
So we we we specialize in really understanding the customer uh at a very fine-grained level. And this goes back to our pedigree around things like understanding payment card transactions and what would be likely fraud or not fraud. But we are involved in essentially um any sort of decision around a customer lifecycle. Think about things like a banking application. If you're a customer of a bank, you know, you originate, you um you need to be protected from fraud, uh scams, you apply for more credit, you have to manage that credit line. FICO is generally involved in many of those decisions through the application of models uh and then decision logic to help banks um get more intelligence around what is the next um likely outcome for this customer, and how can they best step in and help assist in that customer's life cycle.
SPEAKER_00
Brilliant. Of course, every software company is deeply engaged with an AI strategy. Maybe talk about your approach, your strategy, your philosophy when it comes to AI. How
So, you know, at FICO, um, we we very much focus on what is the right AI for our customers. And these customers that we serve are in highly regulated markets. Uh and so, you know, they need to make sure that when they make a decision, that it's explainable, that it's ethical, right, that it's robust, right, and that it's auditable. And that puts very, very strong constraints on the types of AI approaches and machine learning approaches that we may use to solve different types of problems. So, you know, you've you've heard of and industries focused on this concept of responsible AI. Um, you know, I was one of the first to talk about these more than 10 years back. And we've actually built out bespoke types of AI algorithms that will provide that level of robust, uh robustness, explainability, interpretability, um, ethical testing, um, and then auditability around these methods. And so very much our approach is to make sure that we understand the environment that we're operating underneath, understand how our customers take the use of AI very, very seriously because they have a covenant with their customers to make the right decisions in a responsible, measured way. And so we develop our AI around those techniques, which is a big reason why a lot of the things that we do, we we build from scratch. I I myself have more than 100 patents in AI, um, primarily because you know that the generic AI that's out there today is not appropriate for many types of decisions that we make on customers, particularly in the financial services space.
SPEAKER_00
Great insight there. And you know, your your your comment and your your talk about the trust problem in AI over you know the last decade is so timely. What are some of the problems that show up uh most or most often when it comes to trust?
you know, I think one of the biggest sort of challenges with trust is that um, you know, both organizations that use these AI models and also consumers don't have a really good understanding of how that model was built and whether it's appropriate to be used for them, right? We've we've kind of forgotten the basics of data science, being so enamored with a lot of these sort of new technologies that are out there. You know, one of the basics is that you need to be able to understand the data on which the model was built. You have to be able to have a level of control around that, right? Um, and you make very conscious decisions about what algorithms you use, um, how do you monitor those in detail, right? Um, and then, you know, when do you pull that model out when it's no longer working, right? These are all things that are very, very obfuscated when we when organizations choose to use, let's say, a model that they have not constructed for themselves. And so, you know, uh, that is a huge issue. In addition to that, the auditability, right? If a decision gets made with one of these AI models and you don't agree with that decision, right, as a consumer, right, um, you need to be able to audit and ask why. How did it make that decision? And and a lot of that is just uh obfuscated today where you know the data that went into these models originally to build them or the data you used to score may not be readily available. For FICO, what that means is that we have a way that we develop these models, but moreover, right, um it has to be embedded at the very get-go, right? You have to start with this view that every decision that gets made within these models will have to be sort of dissected and audited and demonstrated to be um you know responsible. And and that changes your entire sort of framework around that. And that's where we have this sort of trust issue because you know, without that, right, um, customers are sometimes uh looking at this and saying, no, I'm scared about what I hear in the news, slowing AI development. I'm scared about some of the AI models going off the off the rails, right? Um, and so that that is not something that we can afford to do in the financial services space. And so it's more over of let's choose the right technology that we have confidence in, right? But still provides all of the same level of value as some of these larger models that are are black box and and not auditable.
SPEAKER_00
So there's no seal of approval, I guess, or stamp of uh certifications for trustworthy AI. So is this something you've you've kind of had to develop from the ground up internally? What does it look like?
So for us, what we've done is we've we've advocated for a very strong AI model development governance standard, um, as number one, right? So, like the biggest sort of challenge that we see is that you know, people are building AI and you know, observing that phenomenon and getting excited, but um they're not really thinking uh around you know, the what are the right ways to approach this problem? What are the challenges we're gonna have from a regulatory perspective or explainability or audit perspective? And so, first is get everyone aligned on the same AI standard, right? Which for a large organization, right, if you had a hundred data scientists, you'd have a hundred different ways they want to solve problems. That just doesn't work in enterprise. Um, beyond that, you know, having a standard is great, and it's a lot better than these sort of oaths of being responsible and wanting to be safe with AI, but you have to follow it. And one of the things that we did um early on is we uh coupled AI with blockchain and why. Um one is that all the requirements around this um AI uh standard would be persistent to the blockchain. So um we would know up front before anyone started to touch data or to build an AI model, what were the requirements of the model and what we need from a business perspective? But in addition to that, you know, it it records whether it's been developed, tested, and then verified on the blockchain. And so it's an enforcement enforcement mechanism of that standard. So you know you can go to the blockchain, which is this immutable record of what were the requirements for this model, right, at the time that it was specified, who worked on it, who tested it, were all the requirements met to the precision or the success criteria is established. And if so, release the model. If not, right, you failed to meet those requirements. And we don't have this sort of decision later on, like, wow, we just spent six months building this model and it doesn't work. Let's just let's go with it. No, we have a very, very strong sort of uh covenant around the fact that this has to be followed. And so for us, it's it's around that standard that we can evolve and put our best minds against, and it's around the enforcement mechanism around blockchain. And that blockchain can then be inspected for years in the future, right? So as regulation changes, as the business changes, right? We can go look and say, well, what did a scientist do three years ago or or or six months ago? Um, and in great in a fine level of granular detail, um, which is very often required, right? When things go wrong with AI or when we want to understand how AI is operating. Um, so it that's a big part of it. We've actually, you know, I've written four patents on it, two granted and two still pending. And the ones that are pending are all around agent blockchains and agentic AI, which are our our latest sort of challenge with the use of um AI.
SPEAKER_00
Amazing. And sounds like a great use case for blockchain. Congratulations on that. Uh, but why not a traditional database or audit systems or other kinds of tracking that uh is widely used?
SPEAKER_01
So the immutability is is a really big part of this, right? And so there are ways to kind of check some and make sure that things weren't adjusted. But yeah, we we we very much like this concept of the blockchain because we can make the blockchain available for, let's say, uh a regulator that wants to peek uh and understand how we develop this model or a customer, right? And so that immutability of that chain is important. But beyond that, these chains should go with the models. So think of it like, you know, you you know, in the old days at least, we had cars with operating manuals that would be in the glove box, right? And so, you know, this is basically the operating manual for the model. This is how the model was built, this is the data on which it was built on, this is the algorithm was chosen, this is how, you know, these are the types of explanations that come out of it when we make a decision, this is what when it is um ethical or not. And moreover, um, you know, this is how you monitor the model in detail, which is one of the biggest challenges we have in industry today. No one knows how to monitor models appropriately. So the data scientist who built a model would say, this is when it's off the rails, right? It's outside of its sort of precision uh details. And so you could think about it if you deliver a model or deploy a model, you're actually carrying along with it a blockchain, right? And so you're not taking a database, which would be like a great big model card or something like that. You actually have a blockchain you can refer to. Um, and then you know, it's very, very transparent from that perspective. And it can drive, you know, monitoring within an AI decisioning platform or those sort of uh monitoring constraints.
Tell us about putting it into production. Can you walk us through any real world examples or stories of uh where it's being put to good use?
SPEAKER_01
So for the last 10 years, we've used at FICO, which uh, you know, obviously, you know, if you're gonna invent, uh you should you should uh make sure that it is suited for your business. You know, for us it's really important, right? Um one of the benefits that get overlooked is you know, this is one of the ways we enforce responsible AI. We demonstrated the models were built to a standard, we can share that standard with a with a customer or um in a governance team. But the the other side effect, which is not appreciated, is that mistakes go down tremendously, right? Um, as an organization, let's say I have 300 data scientists, they're they're all using the same explainable AI algorithms, they're all using the same robustness tests, they're all using the same ethics tests. And so, you know, what we get is much, much higher quality models. And so, like I look at you know, FICO as a as an analytic software company uh and these models they have produced, right? Um, you know, 15 years ago, there were more issues with models, right? There would be something wrong with them core, or you know, we get more calls from customers today. Things are on much, much uh narrower, narrower sort of rails, right? Um, and from that perspective, the the quality of the software goes way up, the ability to understand how to monitor that model or meet the governance standards become um much more clear. And then the customers that work with FICO, for example, are you know treat many of those things as the standards for how they want to leverage um and understand how they monitor and govern AI. So it provides a lot of efficiencies, and that's really what we need to do more than anything else, is be very, very efficient and effective with developing these technologies in such a way that you know, not you know, every third model gets rejected from a model governance team. We we don't want that. We want all the requirements for a model governance team up front and met as part of the development process, and then the rest of it becomes, you know, the smooth path, a glide path right to production. And I think that's one of the big, big benefits that we've seen in addition to making sure that it follows a standard and we can stand by that model from a responsible AI perspective.
And taking a bigger picture industry standpoint beyond FICO and your clients, um, how do you expect responsible AI practices to kind of evolve over the next year or two? Change, there's lots of regulation being introduced around the world. Uh, how do you see yourself navigating through all that?
SPEAKER_01
So, you know, I think one of the great things about FICO and my job, and I've been at this at FICO now for like 27 years, and so I've seen a lot of evolution. Um, the global reach is important, right? Um, we have parts of the world where there's very strong AI regulations, we have parts of the world where there's very little AI regulation. And so, you know, that provides, you know, myself a really interesting view of like um, you know, the the spectrum of of where AI regulation is and trying to make sure that we develop AI that that meets all, right? So uh that that I think is going to be important. I I think right now what we're seeing is is a lot of sort of challenges with you know the trust issues not getting better, right? We recently had this, you know, thousand scientists calling for a slowdown in AI development, right? Whether that is, you know, bringing more attention to these models or not, right? That the the real truth is that there is a big concern about the use of AI. And I think where where regulation is going to go is that we are going to stop, you know, treating these things as science experiments. We're gonna get much more sort of serious about um imposing sort of governance around these models, explainability, auditability um requirements. And and I'm hoping uh as a consequence that uh organizations will choose models that are interpretable. We'll choose things like you know generative AI trust scores to monitor the outputs, right? Um they will enforce things like uh what we call here at FICO focused language models, small language models that are built from scratch, bespoke for a very specific purpose with a with the auditability around the data that was used, um, and um you know iterative cycles to do so. Uh, I think it'll become a situation where organizations that want to be, you know, uh really focused on responsible AI will have to be in control of the development of these models. And one of the things I try to talk to people about is that it's it's not something that you have to rely on three major vendors to get a model from, right? It's something that organizations can learn to do themselves, right? Um, and they probably are going to have to be in control, right, of their business, which means they're going to have to be in control of their AI models that they use.
SPEAKER_00
Yeah, so many implications. Uh no longer just three models either, with all the open weight, open source models coming from China and elsewhere and uh other industries too. I imagine so much applicability to healthcare where uh you know similar issues are in play, even more life and death type of issues. So you have your work ahead of you.
SPEAKER_01
Yeah, I I think this is gonna be an amazing space, right? And you know, sometimes people treat you know regulation as uh innovation inhibitor. I I actually think it's an innovation um you know driver, right? And I think we can have the power of these models. I mean, I've been in this industry for almost 30 years, right? And I believe in it strongly. I also believe in the fact that you know we we need to have the right sort of constraints around it. So yeah, uh to be determined and to be monitored, but it's a tremendously exciting space. But I do think we need to start to see things like blockchain um and governance be built in from scratch. I mean, that that's a that's gonna have to be a requirement, and there'll be new innovations around how we meet you know that covenant around a responsible use, along with that powerful sort of return investment that AI provides.
SPEAKER_00
Well, that's quite a mic drop moment. So, on that, thank you so much for sharing all the insight and good luck. Uh, we're all counting on you.
Thank you, and thanks everyone for listening, watching, sharing this episode. Also, check out our TV show, techimpact.tv on Bloomberg TV and Fox Business. Thanks, everyone. Thanks, Scott. Thank you.
