Skip to main content
Sameet Gupte

Sameet Gupte

Nov 18, 2026

0:11
But what about transparency and communication when it comes to governance and what should insurers consider there?
0:16
So, you know, pilots are designed to succeed, right? I mean, they use clean data, enthusiastic teams and, you know, a very narrow scope.
0:28
Production in the real world is messy.
0:31
You know, over the last few weeks, you know, sorry, not last few weeks, over the last few years at Evolut IQ, we have seen a lot of organizations stuck in what I personally call the pilot purgatory, right? and the blockers are rarely the model they are usually ownership process not point solutions drift and change of control fair outcomes third party risk and kill switches and fall back right I think those are some of the you know things that we should use so let me let me just elaborate on that so if you look at ownership right you know in production a business owner must be accountable for outcomes not the innovation team or the IT The ownership should be with the business.
1:18
If you look at process, you should look at the whole process, not point solution.
1:23
See, dozens of disconnected AI tools with each its controls are ungovernable, right? You need to have consistent policies, permissions and audits across the whole process.
0:11
But what about transparency and communication when it comes to governance and what should insurers consider there?
0:16
So, you know, pilots are designed to succeed, right? I mean, they use clean data, enthusiastic teams and, you know, a very narrow scope.
0:28
Production in the real world is messy.
0:31
You know, over the last few weeks, you know, sorry, not last few weeks, over the last few years at Evolut IQ, we have seen a lot of organizations stuck in what I personally call the pilot purgatory, right? and the blockers are rarely the model they are usually ownership process not point solutions drift and change of control fair outcomes third party risk and kill switches and fall back right I think those are some of the you know things that we should use so let me let me just elaborate on that so if you look at ownership right you know in production a business owner must be accountable for outcomes not the innovation team or the IT The ownership should be with the business.
1:18
If you look at process, you should look at the whole process, not point solution.
1:23
See, dozens of disconnected AI tools with each its controls are ungovernable, right? You need to have consistent policies, permissions and audits across the whole process.
0:11
But what about transparency and communication when it comes to governance and what should insurers consider there?
0:16
So, you know, pilots are designed to succeed, right? I mean, they use clean data, enthusiastic teams and, you know, a very narrow scope.
0:28
Production in the real world is messy.
0:31
You know, over the last few weeks, you know, sorry, not last few weeks, over the last few years at Evolut IQ, we have seen a lot of organizations stuck in what I personally call the pilot purgatory, right? and the blockers are rarely the model they are usually ownership process not point solutions drift and change of control fair outcomes third party risk and kill switches and fall back right I think those are some of the you know things that we should use so let me let me just elaborate on that so if you look at ownership right you know in production a business owner must be accountable for outcomes not the innovation team or the IT The ownership should be with the business.
1:18
If you look at process, you should look at the whole process, not point solution.
1:23
See, dozens of disconnected AI tools with each its controls are ungovernable, right? You need to have consistent policies, permissions and audits across the whole process.
3:31
So I'm curious when an AI-assisted underwriting or claims decision is sort of challenged or disagreed with after the fact, what does a truly traceable decision record kind of need to show or how do you achieve that transparency?
3:43
You've got to think of it as a flight recorder, not a log file.
3:49
When a decision is challenged, possibly months later, you need to reconstruct it exactly as it was made, not as the system would make it today.
4:01
It's like the plane, right? I mean, when I say like a flight recorder, when crashes happen or challenges happen, you then simulate it as what it was when the thing was occurring, not in today's environment or in a standstill environment, right? So to enable any claims or any such decision by which the AI is making, you need to capture the following, right? One, the inputs.
4:30
which data and documents were used from which sources and their versions to the model and the rules right which model version prompt policies thresholds were in force at that point in time when the decision was made then the reasoning why the system recommended what it did in a language a customer or you know a businessman or whatever it was right a regulator can follow right the reasoning of that The human touch points.
5:01
Who reviewed it? What did they see? What changed and why? And then the last part that you need to capture is the alternatives.
5:10
What else was considered and with what level of confidence? So more organizations can tell you what was decided.
9:15
You mentioned transparency when keeping the humans as part of the process, but what about transparency and communication when it comes to governance and what should insurers consider there?
3:31
So I'm curious when an AI-assisted underwriting or claims decision is sort of challenged or disagreed with after the fact, what does a truly traceable decision record kind of need to show or how do you achieve that transparency?
3:43
You've got to think of it as a flight recorder, not a log file.
3:49
When a decision is challenged, possibly months later, you need to reconstruct it exactly as it was made, not as the system would make it today.
4:01
It's like the plane, right? I mean, when I say like a flight recorder, when crashes happen or challenges happen, you then simulate it as what it was when the thing was occurring, not in today's environment or in a standstill environment, right? So to enable any claims or any such decision by which the AI is making, you need to capture the following, right? One, the inputs.
4:30
which data and documents were used from which sources and their versions to the model and the rules right which model version prompt policies thresholds were in force at that point in time when the decision was made then the reasoning why the system recommended what it did in a language a customer or you know a businessman or whatever it was right a regulator can follow right the reasoning of that The human touch points.
5:01
Who reviewed it? What did they see? What changed and why? And then the last part that you need to capture is the alternatives.
5:10
What else was considered and with what level of confidence? So more organizations can tell you what was decided.
9:15
You mentioned transparency when keeping the humans as part of the process, but what about transparency and communication when it comes to governance and what should insurers consider there?
3:31
So I'm curious when an AI-assisted underwriting or claims decision is sort of challenged or disagreed with after the fact, what does a truly traceable decision record kind of need to show or how do you achieve that transparency?
3:43
You've got to think of it as a flight recorder, not a log file.
3:49
When a decision is challenged, possibly months later, you need to reconstruct it exactly as it was made, not as the system would make it today.
4:01
It's like the plane, right? I mean, when I say like a flight recorder, when crashes happen or challenges happen, you then simulate it as what it was when the thing was occurring, not in today's environment or in a standstill environment, right? So to enable any claims or any such decision by which the AI is making, you need to capture the following, right? One, the inputs.
4:30
which data and documents were used from which sources and their versions to the model and the rules right which model version prompt policies thresholds were in force at that point in time when the decision was made then the reasoning why the system recommended what it did in a language a customer or you know a businessman or whatever it was right a regulator can follow right the reasoning of that The human touch points.
5:01
Who reviewed it? What did they see? What changed and why? And then the last part that you need to capture is the alternatives.
5:10
What else was considered and with what level of confidence? So more organizations can tell you what was decided.
9:15
You mentioned transparency when keeping the humans as part of the process, but what about transparency and communication when it comes to governance and what should insurers consider there?

We value your privacy

We use cookies to understand how you use our platform and to improve your experience. Click “Accept All” to consent, or “Decline non-essential” to opt out of non-essential cookies. Read our Privacy Policy.