# Keeping AI Honest Is A Full Time Job : An Interview with Anthony Habayeb, CEO, Monitaur

- **Show:** The Leadership in Insurance Podcast
- **Published:** 4 Dec 2023
- **Source:** https://shows.acast.com/theleadershipininsurancepodcast/episodes/keeping-ai-honest-is-a-full-time-job-an-interview-with-antho
- **Transcript:** https://podcasts.opennash.com/shows/leadership-in-insurance/keeping-ai-honest-is-a-full-time-job-an-interview-with-anthony-habayeb-ceo-monit-7caec6/
- **Provider:** OpenNash Podcasts (machine-generated)

## In brief

Anthony Habayeb argues that responsible AI is a business-management problem, not only a technology or regulatory one. Good governance means setting clear goals, documenting how models are built, testing them and monitoring them. Done well, it makes AI projects more likely to succeed. Insurance, he says, is well placed to lead because it is data-heavy and already built around modelling.

The central tension is between the pressure to adopt AI quickly and the readiness to do it safely. Executives want to signal innovation, but many firms have not 'cleaned their room' on data, goals and collaboration. Model builders and risk teams rarely work together. Habayeb sees governance by design as the way to let both groups move fast without losing trust.

## Key insights

1. **Start AI projects with business goals, not tools** Habayeb says firms often jump to buying AI before asking what success looks like. A simpler statistical model can sometimes solve the problem more accurately and with less risk.
2. **Most AI failures are business failures** He cites a McKinsey statistic that 60 to 80% of AI projects miss their business objectives. He argues governance is a way to protect those objectives, not just to satisfy regulators.
3. **Model builders and risk teams must collaborate** Data scientists build models, while compliance, legal and audit teams ask the hard questions but are not technologists. Habayeb wants risk requirements built into development from day one rather than approved six months later.
4. **Insurance can combine actuarial rigor with data science** Actuaries bring validation discipline and data scientists bring fast experimentation. Habayeb says insurance is the only industry that can merge the two, because actuaries are an insurance profession.
5. **Efficiency, not novelty, is driving AI adoption now** Under margin pressure, carriers favour AI in claims, customer service and marketing, where savings or revenue are clear. Habayeb says underwriting and pricing have less automation.

## Questions this episode answers

### What does AI governance actually mean in insurance?

Habayeb describes it as the documentation, policies, testing and monitoring that show an AI model was built well, performs as intended and follows company rules and regulations. He says it is about delivering business goals, not just meeting regulation.

### Why do so many AI projects fail, according to this episode?

Habayeb cites a McKinsey figure that 60 to 80% of AI projects do not achieve their business objectives. He suggests the cause is often unclear goals and weak structure up front, not just technical flaws.

### Should a company use AI for every problem?

No. Habayeb says some problems can be solved just as accurately, and with less risk, by a traditional statistical or linear regression model. The first step is defining what you are trying to achieve.

### What is the 'black box' worry, and is it fair?

The worry is that people cannot see why an AI system made a decision. Habayeb says most systems can be documented, tested and monitored in ways that build trust, much as a certified car is trusted over one built by a child.

### Why is insurance well placed to lead on AI?

Habayeb says insurers hold large amounts of historical data and already work with predictive models. He also argues that if insurance learns to govern AI, it can underwrite the risk of AI in other industries.

## Facts

- **Anthony Habayeb** [0:01:55]: "Monitaur is an AI governance software company. So we bring model governance software to the insurance industry to help them really like build and eventually automate responsible and ethical governance"
- **Anthony Habayeb** [0:24:22]: "we started the company in 2019 but there's a there's a lot of learning all of us need to sort of be humble about and listen and learn"
- **Anthony Habayeb** [0:11:02]: "just yesterday, the White House put out a new executive order that talked about like accelerating their mandate and requirements across all major cabinet members and federal agencies"
- **Anthony Habayeb** [0:11:02]: "I think in two weeks, there's a huge summit in the UK around these things."
- **Anthony Habayeb** [0:31:24]: "there's a stat that McKinsey put out not long ago that 60 to 80% of AI projects don't achieve their business objectives"
- **Anthony Habayeb** [0:31:24]: "we have a partnership with PwC in Europe where we sort of like you know approach relationships together they offer their expertise we offer our software"
- **Anthony Habayeb** [0:35:11]: "an article I wrote some months back for a trade magazine called Is Insurance the Hero AI Needs?"
- **Anthony Habayeb** [0:12:00]: "Nobody asked about AI when Netflix was recommending what movie you should watch."
- **Anthony Habayeb** [0:11:32]: "I think GPT really caused this kitchen table moment, is sort of this phrase that I've used. It brought AI to the kitchen table."
- **Anthony Habayeb** [0:13:56]: "there's only three countries in the world that have net gains in workforce productivity."
- **Anthony Habayeb** [0:16:00]: "The CEO of a major life insurance company was touting how COVID helped them to move to paperless in one area of their business."
- **Alex Bond** [0:15:24]: "only 10 years ago I was watching people walk around the Lloyd's Market with suitcases and I was like"
- **Anthony Habayeb** [0:03:10]: "insurance is sitting on the thing that AI needs, which is tremendous amounts of data, historical data that can help to sort of train and inform predictive models going forward."
- **Anthony Habayeb** [0:05:17]: "leverage some new data that's coming in that's maybe not inside the walls but coming from things like telematics or biometrics"
- **Anthony Habayeb** [0:23:43]: "we have, as an example, an entire library of control policies that literally tell the model builder, this is exactly what you need to do to mitigate this risk."
- **Anthony Habayeb** [0:07:53]: "We're in one of those downward margin pressured moments of insurance at the moment, right? And therefore every major carrier is looking for efficiencies."
- **Anthony Habayeb** [0:28:26]: "actuaries have quite a bit of discipline and rigor and structure around their model development journey"
- **Anthony Habayeb** [0:20:47]: "they've historically been a little federated in their governance. Claims manages claims. Personal lines manages personal lines."
- **Alex Bond** [0:27:41]: "They were just actuaries that had done the Coursera course and then decided that they were now data scientists."

## Transcript

**[0:00:00] Alex Bond:** This is the Leadership in Insurance podcast, brought to you by Finpro Search Partners. Insurance companies are businesses and they need to look for the long term and be sustainable.

**[0:00:08] Anthony Habayeb:** We went from zero to one and now it's going from one to a hundred.

**[0:00:13] Alex Bond:** Insurance as a concept, as a kind of service, is brilliant. The execution is what we're looking at now. I think the companies that are going to succeed are the ones that are going to understand and master the art of intent. When we talk about innovation, we lean too heavily to think about technology and we don't think about creating a culture of innovation. I think innovation is essentially continuous improvement of existing processes and platforms and product, right?

**[0:00:41] Anthony Habayeb:** It's got to be easy. It's got to be seamless.

**[0:00:49] Alex Bond:** We'll start customarily saying good morning and welcome to the Leadership Insurance podcast. But obviously it's not morning, it's early afternoon. We're in Vegas. It's day one though, so we're all challenging ourselves with time zones and flights and a bit of jet lag and adjusting to time zones. So, welcome to the podcast. I'm your host, Alex Bond. I'm really lucky today to be joined in person, in situ, by Anthony, who's co-founder and CEO of Monitaur. Anthony, good afternoon, how are you?

**[0:01:18] Alex Bond:** This is great.

**[0:01:19] Anthony Habayeb:** This is great. I feel like I'm looking at you as if you're on a computer screen but you're actually right in front of me like I'm looking in your soul but so if it's

**[0:01:26] Alex Bond:** awkward it's because I'm not used to doing this I know it's it's really strange I think we were just having this conversation that was saying I still don't think I was adjusted to the fact we're in Vegas you know this I mean Vegas is a hell of a place to visit the best of times but being amongst 9 ,000 insure tech execs or interested parties is quite something so um yeah but before that we're going to get distracted by ITC and I'll talk about ITC a bit more.

**[0:01:51] Alex Bond:** Please introduce yourself and obviously the business and give us some context for this

**[0:01:55] Anthony Habayeb:** conversation. Yeah, so I'm Anthony Habayeb, the CEO and co-founder of a company Monitaur. Monitaur is an AI governance software company. So we bring model governance software to the insurance industry to help them really like build and eventually automate responsible and ethical governance across these modeling systems, AI projects that are making really important decisions. What should your price for insurance be? Should I approve this claim? Should I underwrite this individual? And sometimes what ad you're getting.

**[0:02:28] Anthony Habayeb:** So we're excited by the potential of AI. And we built a company to really help engender trust and confidence in AI through something super unsexy, which is governance and auditability.

**[0:02:41] Alex Bond:** But surely the money to be made in insurance is all in the unsexy stuff. It's such a sexy industry. Yeah, yeah. But it is.

**[0:02:47] Anthony Habayeb:** But it is actually. Yeah.

**[0:02:49] Alex Bond:** That's really interesting because I think, you know, you and I have obviously talked previously and I want to try and keep this kind of centered on where you fit into the AI equation because I think, do you think the kind of noise around AI is unhelpful to kind of like building products like your own?

**[0:03:10] Anthony Habayeb:** That's a tough one, right? I mean, I think that, so from an industry perspective, I think insurance is sitting on the thing that AI needs, which is tremendous amounts of data, historical data that can help to sort of train and inform predictive models going forward. And I think that there's tremendous opportunity around that. I also think that as an industry, sometimes insurance is a little laggard to other industries with technology. and I think that executives currently feel a huge amount of pressure to be innovative and to signal their investments in new technologies and sometimes don't necessarily have the readiness to really achieve the objectives they set out to achieve.

**[0:03:52] Anthony Habayeb:** So I think it's both amazing because it's going to push a little bit harder than some folks might be comfortable but then there's the flip side of that which is I think that there's some sort of clean your room before you go out and play sort of thing that needs to happen in in almost every industry right insurance is not that unique but that's

**[0:04:08] Alex Bond:** definitely the risk that I think we face. I love that phrase clean your room because I think about that in many contexts as you know we talk about as many people that know I'm on a search firm FinPro Search Partners and when we talk to clients about hiring you know the example I always use is that people love psychometric testing for people that are going to join the business what they never do is run that internally so they don't know what they're dealing with internally so they're talking about kind of personalities behaviors and that person is not the right fit for us but they don't know what is making their team work currently sometimes there's a lack of awareness so

**[0:04:45] Alex Bond:** that kind of clean your room comment applies to kind of most things when you're trying to kind of grow it's like the you know don't run before you can walk on that yeah Yeah, absolutely. Yeah, yeah. Yeah, absolutely. What's the, what should people be doing to be ready? And I appreciate you've kind of, you can solve problems in different areas. And you mentioned kind of various ones of them, looking at the models they're using through to kind of what adverts.

**[0:05:09] Alex Bond:** But where do people start in kind of being ready to adopt things like these kind of AI technologies?

**[0:05:17] Anthony Habayeb:** Well, I think first, let's soften the word AI. I mean we're nowhere near sentient or like real artificial intelligence. I think the immediate opportunity in insurance is to use more advanced machine learning models versus maybe some of the linear regression or statistical models have been used historically, the actuarial models that have been built on historical data. That's the now opportunity, right? It's leverage some data, leverage some new data that's coming in that's maybe not inside the walls but coming from things like telematics or biometrics and how to bring those things into ultimately insurance is about predicting risk, mitigating risk.

**[0:05:53] Anthony Habayeb:** There's a lot of interesting data and opportunity to do that I think in a more robust and a unique way. So in that context insurance companies are trying to get their hands around what data do I have, what data is interesting to me, can I bring these things together in a way that it's that accessible for talented resources to build models that align with some business objective. And I think That third piece of a business objective is actually one of the most important parts that it sometimes jumps past.

**[0:06:25] Anthony Habayeb:** People are like, I want to go use this tool. I want to go buy this technology. I want to go use AI. But they haven't maybe necessarily started with just like the fundamentals of almost any project. What am I trying to achieve? What does success look like? Then you sort of walk through, well, is AI the right approach or not? There's many situations where you might be able to solve a problem with a new type of machine learning.

**[0:06:47] Anthony Habayeb:** but you could just as accurately and with less risk solve that problem with some form of statistical or linear regression model. And so I think the first step in, I think, AI success is actually this thing that we do, but it's not the risk side of the word governance. It's the business management side of that word, which is like having good discipline around what am I trying to achieve? What does success look like? let's find the right way to solve that problem if it happens to be AI then great then you can really align around the opportunity so a bit of a long

**[0:07:23] Alex Bond:** answer but no but I think that's it's a prudently well done wrong answer because I was just thinking about those conversations with clients and I was thinking with all the sort of different elements of what you do as a business and where you can integrate and where are the most common conversations that that you're having now, like either sort of opportunities that clients are bringing or challenges. And I was interested to see sort of how that's changed maybe over the last couple of years.

**[0:07:53] Anthony Habayeb:** I mean, efficiencies are really important, right? We're in one of those downward margin pressured moments of insurance at the moment, right? And therefore every major carrier is looking for efficiencies. And even the AI vendors that are seemingly doing best at the moment are the ones that like can really map towards an efficiency or like an obviously new revenue opportunity. So things that can sort of like streamline claims, streamline customer service, right? Reduce the number of heads that you might need to do that kind of work or otherwise reallocate those heads towards higher value work and you can start to automate some of the more obvious tasks and opportunities.

**[0:08:34] Anthony Habayeb:** So things that are driving efficiencies right now are where I see most of the application happening. so like claims customer service pricing opportunities but that's not pricing underwriting it does not have the same degree some of the automation we would see in claims or in marketing or in customer service I'm obsessed with claims

**[0:08:55] Alex Bond:** people on the podcast know this these the working claims so we're looking at that yeah yeah yeah and the automation aspect I think what excites me about anything automated is we look at people and people getting more rewarding jobs yeah and taking away mundane tasks where do you think we are on that kind of cultural pathway to allowing kind of straight through kind of automation having human in a loop and I suppose I'm really looking at where people are comfortable because I think it's we had a great conversation on the on this prod where we talked about automation before and we're saying ultimate kind of like North Star will be when we're happy to kind of had surgery by full

**[0:09:45] Alex Bond:** automation and and actually it would be crazy to think we're getting a surgeon in we're definitely not there so the insurance kind of piece you know it's

**[0:09:55] Anthony Habayeb:** fun actually healthcare is interesting right because I think that people ask this question how accurate is this system at predicting you know some malady right but doctors are not always right no right but do we ask a doctor like hey what's been your accuracy rate in predicting colds or the flu like we don't but because there's this software system like and because the thing we can't necessarily interact with in the same way we do have certain questions that we feel we can't answer and sometimes that gets captured as the black box but most of these systems actually can be made to be hugely introspective, where you can document them, you can say how they're supposed to perform, they can be

**[0:10:33] Anthony Habayeb:** tested, they can be monitored, and those kinds of things start to create trust in the system. And so I think as we continue to provide visibility to the people impacted by AI systems, visibility regarding was it built well, was it quality controlled, was it tested, is it being monitored, were the people who built it skilled and able to do it, right? Like, I wouldn't get in a car that my 12-year-old son built, but I'll get in a car that is sort of certified and has gone through a whole bunch of things, right?

**[0:11:02] Anthony Habayeb:** I think that's the maturity of a new technology. But I think the good news is the demand for this technology is also forcing that we do need to apply certain controls. You know, like just yesterday, the White House put out a new executive order that talked about like accelerating their mandate and requirements across all major cabinet members and federal agencies to accelerate their auditability and their assurance and their certification processes around AI systems. I think in two weeks, there's a huge summit in the UK around these things.

**[0:11:32] Anthony Habayeb:** So I think we've hit this inflection in, let's say social awareness of AI's potential, that society is now also equally saying, hey, I'm excited about this, but I want to make sure that this thing is okay. And I think GPT really caused this kitchen table moment, is sort of this phrase that I've used. It brought AI to the kitchen table. And because of that, you now have this public awareness of, oh, this is exciting, look at what this thing just did for me.

**[0:12:00] Anthony Habayeb:** How many resumes have you looked at that probably would have been created by a candidate lately? Like optimized for your search interface to find the best person. Those are exciting. And I think the demand is also causing the demand for more transparency and that will accelerate usage. So the higher the risk, the more people want assurances of the quality and efficacy of a system. Nobody asked about AI when Netflix was recommending what movie you should watch.

**[0:12:31] Anthony Habayeb:** But it's been doing it for quite a while. So I think these things track in parallel, and that's a good thing.

**[0:12:39] Alex Bond:** Yeah, we've talked a few times on the podcast about the influence of the consumer mindset. You can't separate the two. We have a consumer personal life, and then we work in businesses. And once you got to what I call the Amazon moment, they are very good at hitting what a customer wants when they have a buying experience. You can talk about the ethics of things and what they have, the electric workers and all that sort of stuff, but actually the kind of delivery of the product is exceptional.

**[0:13:09] Alex Bond:** Now your expectations, you just have to take that into the workplace. So when you look at technology and it's the same for ChatGPT. So if that kitchen table moment was important because it goes, now we know what the possible is on an individual level and how can that apply to us in the business context? So I think that's interesting because that pushes us along the kind of cultural journey towards being comfortable with this stuff. Agreed. Because I get excited about this, and I've been trying to pitch this in several conversations I've had now, so I'd quite like to have your opinion.

**[0:13:40] Alex Bond:** When we first looked at automation in the home, there was a conversation about, oh, we can work less. We can be more productive, and we can work less hours. That is still a possibility. That's one way of looking at using technology. It's like use technology and work less, but we always have to produce more.

**[0:13:56] Anthony Habayeb:** You know what? But I was just at a conference in New York, and the head of software at Microsoft, Microsoft IBM, I'm sure they wouldn't mind that I, I'm not sure which one it was, but it was one of them. One of those. He made some comment about, there's only three countries in the world that have net gains in workforce productivity. Only three. Wow. And I want to say what they are, but I'm not sure, so I won't misstate that, because I'm sure we could find that information.

**[0:14:26] Anthony Habayeb:** but sort of fascinating isn't it right and so if you think about that and there's all these worries that like people are gonna lose jobs like we're actually like less productive than we have ever been as a broader society so the idea you could take sort of modern technology and AI and start to like fill a gap of a bunch of the productivity that we've maybe lost like there's gonna be more work to be done and we have the people to do it so if you can really almost like get a foundation of automatable work that can be delivered by certain systems, then we really only have a certain capacity today to do work.

**[0:15:00] Anthony Habayeb:** And so we actually need to offset some of the existing work requirements from the human workforce so that the human workforce can actually give us net gains in productivity by higher value work. It was really interesting. My mind got sort of wrapped around. I'd never thought about that current stat and that actually we have a fundamental human shortage in the amount of demand for productive work that exists.

**[0:15:24] Alex Bond:** I mean it resonates with me because Brits are famously in Europe for the longest hours but for the least production. So something needs to be done there. I think particularly in insurance as well. Look, we're only, well I've been in the insurance industry for 18 years and only 10 years ago I was watching people walk around the Lloyd's Market with suitcases and I was like, oh where are they going? They must be off to wonderful conferences like ITC.

**[0:15:52] Alex Bond:** No, they're carrying around files of paper. And you think, that's within the last decade. Yeah. So we've got a way to go.

**[0:16:00] Anthony Habayeb:** I mean, I was at a different conference not long ago. The CEO of a major life insurance company was touting how COVID helped them to move to paperless in one area of their business. This is in the last three years. Wow. But this was like a celebrating moment. Yeah, yeah, yeah. We moved to paperless because of COVID. So, I mean, I think, but that's back to an earlier comment. Like, insurance might be laggard sometimes in technology, but as it relates to AI, there's some foundation that needs to be set, some clean your room type dynamics that have to happen.

**[0:16:35] Anthony Habayeb:** But this technology in particular, insurance is pretty primed for it because it is so data-centric and already a modeling type industry. The opportunity is tremendous. And I think that's why you see so many CEOs and insure techs and investors really circling this vertical in particular because the efficiencies to help bring underwriters and actuaries and risk executives up to higher level problems because there's a lot of rote work that we could probably take off their plate.

**[0:17:06] Alex Bond:** Sure, sure. I was thinking about your ideal customer profile when you were saying that, and I was thinking, who are they? Because to one extent, presumably they have a certain level of technical competence to be even effective as a partner in a partner business. But then also the people with the biggest problems are probably the least close to the kind of ideal tech setup. So who is that perfect customer profile for your business?

**[0:17:31] Anthony Habayeb:** Actually it's interesting what you just characterized will give those department names. Like that first persona you explained is the model builder. Maybe it's a data scientist, it's a machine learning engineer, It's an IT or software engineer. It could even be an actuary, right? The people who are actually building these modeling systems or applications, right? That other side, let's call them the risk managers, right? You've got your compliance, your regulatory affairs, your legal, your internal audit. They are not technologists, but they're the ones who ask the questions and are doing the work to protect the company, to protect consumers, to protect stakeholders, right?

**[0:18:10] Anthony Habayeb:** Those two groups in particular, for an ideal Monitaur customer, a company has those two groups who want to collaborate but really don't know how. We've got the saying, good AI needs great governance, and great governance really requires the model builders and these risk teams to come together in a way that they don't today. To see the projects, to understand what's happening, to understand if all the work that was required has been done, to be able to even ask questions like, hey, Alex, what if you were a woman, would this model have handled you the same way?

**[0:18:43] Anthony Habayeb:** What if your age was like 55 instead of 35? What would have happened here? How many times did our algorithm make a decision about somebody in California that had the following characteristics? Those kinds of questions are not really accessible to those non-technical partners today. So if we can bring these teams together, that's the opportunity to create transparency and trust in these systems. and our ideal customer really has a technology team that wants to go faster but could use better structure and organization of their work.

**[0:19:14] Anthony Habayeb:** And some risk stakeholders would say, I want my team to go fast, but I need to know if they're doing what we need them to do. It would be really nice if instead of they built a model and now they wait for me six months to approve something, what if they could know what I need from them from day zero all the way through the journey? They could literally embed my risk requirements in their model development.

**[0:19:33] Anthony Habayeb:** How amazing would that be? Well, that's really what we're trying to empower. It's almost like governance by design, right? As you're building, it's there. It's not this thing that's hanging off to the side that nobody wants to go down the hall to talk to the governance team, right? Let's make them partners. And they want to be partners. Just don't know how, because they're not technologists.

**[0:19:52] Alex Bond:** And that's making everyone talk to each other, because I think I shared this with you when we first spoke. I did a piece of work, a search work. and we had to build an algorithmic underwriting team. Or it was a project to see how possible that was and if they were going to build it themselves versus buy. And it was an interesting kind of conversation. So they started on, you know, we put this great team in and they were saying, right, what are the component parts we need to make that happen?

**[0:20:19] Alex Bond:** They started talking to all the different stakeholders in that equation, similar to the stakeholders you just mentioned. Every one of those teams was some way on that pathway to doing the bit that they needed to do, but none of them knew it. So there's a huge problem and do you see that thematic with the clients where they just these different individual departments just don't have any clear way of communicating or maybe just don't communicate? That's a corporate problem everywhere.

**[0:20:45] Alex Bond:** Yeah, it's not an insurance unique problem.

**[0:20:47] Anthony Habayeb:** But I think insurance is historically their lines of business operate as separate business units right? Their own P &L and operating division right? So I think that one of the interesting challenges of AI for an insurance company is they've historically been a little federated in their governance. Claims manages claims. Personal lines manages personal lines. Commercial manages commercial specialty. Marketing. These things tend to be a little isolated. But as you talk about AI almost as a horizontal, it's really going across all of these divisions.

**[0:21:21] Anthony Habayeb:** Each of them are excited and believe in the potential to iterate and improve their area by leveraging AI. But the way that you would govern the AI is very similar across all those things. And so, you know, similar software or sort of IT in some ways, there's this opportunity to build a repeatable pattern and create a communication that goes across departments to better manage AI. And so I think these, you know, maybe this can be a catalyst for better collaboration between technology and risk generally, right?

**[0:21:59] Anthony Habayeb:** Like this is a software type that's uniquely requiring more close coupling and partnership between the teams. And less specialization of understanding the business unit, right? And more being an expert like what's the risk of the technology and how do you manage the risk of the technology?

**[0:22:19] Alex Bond:** I keep looking at my screen because there's a really good quote from you and I was trying to fall into a question but I just think it just sits very well. The path from model to proof of ethics isn't a straight line. It's like incrementally let's get better as we go.

**[0:22:34] Anthony Habayeb:** Yeah and I you know there's also this other like turning principles into practice right like is Is there an insurance company that doesn't have some societal or environmental or governance statements and opinions? Is there an insurance company that doesn't have ethical standards and guidelines that they operate with? No way. Right? But how do you actually take those things and translate them down into literally what is my data scientist doing and does this thing ladder up, you know, in?

**[0:23:03] Anthony Habayeb:** And that's hard. That's really difficult. And that's not a thing, frankly, that the data scientists or the risk manager have any expertise at doing. And I think one of the market opportunities that we see is really helping to translate whatever these policies might be up top, whether it's an ethics, or it's a financial, or it's a technology, these higher level governance principles and standards, or even regulations, helping to translate those down to executable guidance and tooling for those two stakeholders, the model builders and the risk teams.

**[0:23:43] Anthony Habayeb:** That's actually a huge piece of our solution is that we have, as an example, an entire library of control policies that literally tell the model builder, this is exactly what you need to do to mitigate this risk. And by the way, if you do it this way, it maps to this corporate policy, it maps to this regulation, and it maps to this standard. right so that's a real big gap and challenge right now is there's really not X I mean a lot of people say they're experts like oh AI risk management AI governance like how you become an expert at that I think sometimes like someone's like oh talk Anthony he's an expert not like you know it's really maybe

**[0:24:22] Anthony Habayeb:** as much as anybody because I've been we started the company in 2019 but there's a there's a lot of learning all of us need to sort of be humble about and listen and learn and absorb and you know test theories and then start to apply real executable practice.

**[0:24:42] Alex Bond:** Yeah there's so much there that I've wanted to pick up but it really resonates with the conversation we had. My colleague Soph and I attended this event which is called Make and Change of Insurance and usually you know it's predominantly a DE &I focused group and were looking at AI and it was kind of an AI ethics but it was it was talking about this challenge of going how do you interrogate it and I always I would say I call it the FD problem so when you're growing a business for the first time and you're not a finance person and you have to go and hire your first finance person now I know what a good CFO looks like in

**[0:25:19] Alex Bond:** insurance I don't know what a good FD looks like in recruitment because I've never had to hire one yet I'm an outsourced one. So I wouldn't know a good one from a bad one, because I don't know how to ask the right questions. I'm not an expert. So that thing of hiring an expert, because they're the expert, how do you find that person? How do you quantify that?

**[0:25:38] Anthony Habayeb:** I mean, that's hard. It's not an easy thing, right? I think every hiring manager struggles a little bit with hiring a thing that they don't have intimate experience or knowledge for. There were moments in my career where I stressed personally about a decision I made early on to be a pretty intentional generalist. You know, like I've run every division. You know, my journey to Monitaur came through intrapreneurship at big companies. You know, sold, managed sales, built products, managed products.

**[0:26:11] Anthony Habayeb:** You know, all of the departments, biz debt, mergers, acquisitions, right, global responsibilities. But this reason I decided to go down that route is because I wanted to have like good general understanding of all of these different pieces to try and de-risk exactly what you just said, that I would be in some situation that I had like zero context for, or like no foundational awareness to reference to help me think through a problem. But even as the founder of a company that's like focused on AI governance, I continue to find myself like learning and absorbing and and consuming more.

**[0:26:47] Anthony Habayeb:** And so I think I would almost run as a hiring manager. I'm just thinking about this in real time. I always run away from someone who is overly espousing to be an expert. Because this is a pretty nascent space. And if someone doesn't come to you with humility and curiosity, they know something, but they also are hugely aware of what they don't know. and they showed eagerness to stay plugged in and continually learn and engage, that might be one of the most important characteristics you're looking for as you're building, let's say you're looking for a head of AI, right?

**[0:27:26] Anthony Habayeb:** Or someone to take on a senior role. Like, you need, I think you need that person to be hugely curious and aware that they are not truly an expert so much as they might try to propose they are. It's interesting. I've never thought about this before.

**[0:27:41] Alex Bond:** No, I mean the thing I was able to reflect on that is that when we had this change in insurance where we went from all data was kind of managed by actuaries in an actuarial mindset and then we started to build data science teams. One, it was about who reports to who but all of a sudden because the demand for data science was so large we had loads of data scientists suddenly emerge. They were just actuaries that had done the Coursera course and then decided that they were now data scientists.

**[0:28:06] Alex Bond:** I said my concern is that with because it's such a nice kind of area. We're getting that sort of development. So kind of trying to spot the true experts from the not. And to your point, it's like, you can only be an expert to a point anyway. And then it's about being curious. And it is a challenging place to particularly hire. But I wanted to... Actually, if you don't mind,

**[0:28:26] Anthony Habayeb:** something you just said that I think is actually a really interesting observation I have about insurance. I believe that the silo of data science versus actuaries is an opportunity that insurance uniquely has versus other industries to cause the best possible AI systems. So actuaries have quite a bit of discipline and rigor and structure around their model development journey, right? Their methodology, their validations, the work that they do to sort of get to a place where like, okay, this thing is now ready to be used, right?

**[0:29:00] Anthony Habayeb:** Data science, which has come up for more like the software and agile world, is a little bit more like quick in experimentation and getting the results and then maybe making optimizations as you go or course correct. I think that insurance, because they have this history of actuaries, bringing some of those sort of actuarial standards of practice and methodologies into the more modern sort of, modern is not necessarily the right word, but the alternative approach that data science takes to solving problems, better synergies between those is a tremendous opportunity that insurance is like the only industry that could sort of cause that because actuaries are an insurance thing effectively.

**[0:29:39] Anthony Habayeb:** Yeah. So I'm excited by that. It's something that we realize potential, like our software helps to actually cause better alignment of actuarial practices and data science practices. So if I'm a carrier, I'm not having to reinvent the wheel for how do I manage my data science and AI projects versus my actuary. So a lot of opportunity to bring this together. I think as we see insurance move forward, you're going to see probably more organizational alignment happening between actuaries and data science resources.

**[0:30:08] Alex Bond:** Yeah, I mean, from a, I think we might have touched on this before, but my complete belief is that ultimately we're going to have that data will be the top-down department and actuary will be part of that and data science may be part of that. And it's like overall we'll have a custodian of how we use data inside the business and strategically what do we want to do with it. Whereas when it started, the evolution of it, I was going out and hiring a head of data science who was going to look at data across this entire carrier in the most exciting experimental ways but then they were reporting into a chief actuary and it seemed at odds because

**[0:30:43] Alex Bond:** an actuary thinks in a certain way about data in an insurance environment. They don't have that broad a context. Yeah, we see that often. Something I also want to go back to, and I think this is important with your business, is that we talk about ethics and then we talk about ethics in governance and then we talk about sort of regulation. I think that's an important point to kind of hammer home because I think when people think about governance, they think about regulation.

**[0:31:13] Alex Bond:** They don't think about, oh, we've got values as an organization and we want to stay true to them. Is that something that gets missed in the conversation about what you do?

**[0:31:24] Anthony Habayeb:** there's a stat that McKinsey put out not long ago that 60 to 80% of AI projects don't achieve their business objectives wow and we have a partnership with PwC in Europe where we sort of like you know approach relationships together they offer their expertise we offer our software and those things come together pretty nicely. Governance, in the sort of GRC definition, it sort of controls to help protect achievement of business objectives. It's not just about regulation at all.

**[0:32:00] Anthony Habayeb:** How do I know if I'm going to give you $3 million, what's happening with this money? What structure do you have in place for this project to help ensure that the project you said you're going to deliver gets delivered on time with good quality and performs the way that you want it to? What's all the work that you're doing to help cause that? I think that sort of the ethics is a subset of that what does good look like definition set that sometimes is missing in some of these AI projects and journeys, right?

**[0:32:26] Anthony Habayeb:** So there's that upfront sort of grooming that needs to happen around a project that understands like what are my goals? What are my objectives? What standards do I need to align against, right? What sort of corporate policies might this thing infringe upon or need to reinforce, right? What data am I using? Do I have the permission to use? Like all of these things are really important pieces of causing a successful project. And bonus, demonstrating your regulatory and internal corporate alignment against risks.

**[0:32:53] Anthony Habayeb:** So that's a big piece of what we really do in a lot of our relationships is actually like almost neutralize the word governance to not be viewed as a singular risk requirement. but as more of like a creating better structure and likelihood of success in your AI projects. And that's done by better documentation, clear policies, integration within your systems and infrastructure, creating alignment between people. But I think actually most companies, they have an expectation of what they think is good or bad, what they would be okay with or not okay with.

**[0:33:31] Anthony Habayeb:** They don't have a way to really like affect that stuff onto projects, right? Or sniff it out afterwards, like, are we staying true to what we thought we would? Your question earlier about how you would introspect or test the system. It happens throughout a journey. And I do believe that companies absolutely can apply their values and their ethics policies into their systems. Because ultimately, this is software built by people. There are decisions made by individuals as these applications are built that can cause very specific outcomes.

**[0:34:05] Anthony Habayeb:** You can build a model to do anything. And if you do not understand sometimes, that's a gap as well. Do the people building the models, do they have visibility into what's expected of them? Sure. So that they can ingrain those in what they're doing.

**[0:34:20] Alex Bond:** Yeah, that's that good governance as a pathway to growth as a business rather than kind of just looking at, Ryan, a copy for you. Yeah, totally. Let's keep going. What else do you have? But I thought that's really important because we both did ourselves a disservice at the start of this conversation because we said, oh yeah, we were sort of joking about saying governance is a non-sexy piece. And obviously there's a joke in that. In the same way that insurance, everyone in insurance here drolly says, oh, it's not the most exciting sector.

**[0:34:47] Alex Bond:** But you're like, without insurance, business isn't allowed to happen. Period. Without governance, insurance is fail or succeed. But when we say governance, I think it gets tied to regulation so quickly. But regulation is just like table stakes. It has to do that. We're in insurance. Yeah, it has to do that. They have to be regulated. So the sexy side of governance, and I'm going to go that far, is the delivery of these business objectives, which obviously your software helps you do.

**[0:35:11] Anthony Habayeb:** Maybe somewhere with this post we could link to or share an article I wrote some months back for a trade magazine called Is Insurance the Hero AI Needs? And the whole premise of this was insurance, if insurance solves how to govern AI, it then is able to backstop AI for all the other places that we want to be using AI, right? To exactly your point, companies don't get built. People don't take risks, right? Like insurance is literally critical for anything that happens.

**[0:35:47] Anthony Habayeb:** When you really look at it, it's almost overwhelming like holy crap. Like insurance is everything, right? And I think by this industry solving how to govern itself around this technology it becomes more prepared to support the technology across all of the use cases where it can impact our lives and that's really inspiring I think that's actually a really exciting idea yeah I want to leave it there

**[0:36:09] Alex Bond:** because that's the best a really nice place to end it and we're championing the bit for kind of insurance as being a you know catalyst for great business which I really believe it to be so Anthony thank you so much for taking time out of your very busy schedule at ITC really enjoyed this conversation

**[0:36:24] Anthony Habayeb:** This is great. Thanks, man.
