BẢN CHÉP ÂM THANH 🔗
WEBVTT
00:00:00.017 --> 00:00:05.177
Welcome back to the TX Show. I'm John, and with me, as always, is Rakesh.
00:00:05.557 --> 00:00:09.337
Today, we're diving into one of the most exciting shifts in the insurance landscape,
00:00:09.577 --> 00:00:16.257
how agentic AI is redefining parametric insurance and helping carriers better manage climate risk.
00:00:16.757 --> 00:00:20.577
If you're listening to this, you probably don't need convincing that climate
00:00:20.577 --> 00:00:25.677
events are becoming more frequent, more severe, and frankly, more expensive.
00:00:26.297 --> 00:00:32.197
Floods, wildfires, heatwaves, storms, these are no longer once-in-a-century events.
00:00:32.417 --> 00:00:37.457
And that reality is putting insurers under a lot of pressure to do three things at once.
00:00:38.057 --> 00:00:41.157
One, respond faster when events happen.
00:00:41.817 --> 00:00:47.197
Two, price smarter before they happen. And three, deliver payouts with precision
00:00:47.197 --> 00:00:48.857
when customers need it the most.
00:00:49.297 --> 00:00:52.117
That's exactly where agentic AI steps in.
00:00:52.257 --> 00:00:56.417
It's not just tuning risk models a little bit. It's fundamentally changing how
00:00:56.417 --> 00:01:01.237
insurance operates, especially when you combine it with parametric triggers.
00:01:02.057 --> 00:01:05.517
Rakesh, set the stage for us. Why is this such a big deal right now?
00:01:06.237 --> 00:01:08.197
Thanks, John. If you think about
00:01:08.197 --> 00:01:12.837
traditional insurance, it has always been reactive. The event happens.
00:01:13.337 --> 00:01:17.917
The customer reports a claim. Someone assesses the loss, collects documents,
00:01:18.057 --> 00:01:21.177
and then the carrier decides whether and how much to pay.
00:01:21.417 --> 00:01:25.237
That workflow is slow, it's manual, and in a big catastrophe,
00:01:25.537 --> 00:01:27.217
it's often completely overwhelmed.
00:01:27.457 --> 00:01:32.037
With agentic AI and parametric insurance, we're seeing a paradigm shift.
00:01:32.717 --> 00:01:36.457
Instead of waiting for someone to prove a loss, insurers can predict,
00:01:36.757 --> 00:01:40.517
validate, and trigger payouts automatically based on predefined parameters,
00:01:40.877 --> 00:01:44.597
things like rainfall levels, wind speed, temperature thresholds,
00:01:44.697 --> 00:01:47.197
or even river height at a particular location.
00:01:47.997 --> 00:01:53.357
So if we know, for example, that wind speeds above 120 miles per hour in a defined
00:01:53.357 --> 00:01:56.377
radius almost always lead to a certain level of damage,
00:01:56.617 --> 00:02:01.257
you can build a product where once that parameter is met, the event is confirmed
00:02:01.257 --> 00:02:05.817
by AI using trusted data sources, and the payout is triggered automatically.
00:02:05.877 --> 00:02:10.197
No lengthy claims process, no ambiguity about whether a loss is covered,
00:02:10.497 --> 00:02:12.657
just speed, transparency, and trust.
00:02:12.877 --> 00:02:15.017
And that's such an important point, trust.
00:02:15.597 --> 00:02:19.957
When people are dealing with a flood or a storm or a heat wave that disrupts
00:02:19.957 --> 00:02:23.517
their business, what they remember is not your risk model.
00:02:23.697 --> 00:02:27.557
They remember, did my insurer show up for me quickly, fairly,
00:02:27.837 --> 00:02:29.437
and without putting me through pain?
00:02:30.397 --> 00:02:36.577
Parametric insurance plus agentic AI gives carriers a real chance to show up better.
00:02:37.177 --> 00:02:41.777
So let's break this down into parts. We'll start Start with AI-powered risk
00:02:41.777 --> 00:02:45.037
assessment, then move into automated claims of verification,
00:02:45.477 --> 00:02:51.117
basis risk, new opportunities, and finally the future of AI in climate risk.
00:02:52.017 --> 00:02:56.357
Historically, risk models were pretty static. You'd build them on years of historical
00:02:56.357 --> 00:03:00.497
loss data, maybe some vendor models, and you'd revisit them occasionally.
00:03:01.893 --> 00:03:04.853
Climate doesn't behave in nicely stationary patterns anymore.
00:03:05.653 --> 00:03:07.913
The past doesn't perfectly represent the future.
00:03:08.913 --> 00:03:13.093
Eugenic AI changes that because it brings dynamic intelligence to the table.
00:03:13.573 --> 00:03:17.253
It can pull in real-time data from satellites,
00:03:17.453 --> 00:03:22.213
IoT devices, and weather systems, combine that with historical data sets,
00:03:22.313 --> 00:03:26.513
and continuously refine climate risk profiles for different regions,
00:03:26.833 --> 00:03:28.573
segments, or portfolios.
00:03:29.573 --> 00:03:35.413
Rakesh, can you make this real with a use case? Sure. Let's take a coastal parametric
00:03:35.413 --> 00:03:36.853
cyclone product as an example.
00:03:37.673 --> 00:03:41.833
Traditionally, you might use historical storm tracks, historical wind speeds,
00:03:42.233 --> 00:03:46.153
historical loss experience, and then you design a product with a fixed trigger.
00:03:46.353 --> 00:03:50.853
Say, if a storm of category 3 or higher crosses this zone, we pay X.
00:03:51.353 --> 00:03:54.733
But with agentic AI, you can go much deeper.
00:03:54.733 --> 00:03:59.933
It can track live storm formation using satellite feeds, monitor wind speeds
00:03:59.933 --> 00:04:02.353
in near real time from multiple data providers,
00:04:02.693 --> 00:04:06.933
blend that with local vulnerability data, building density, construction type,
00:04:07.173 --> 00:04:11.033
elevation, and so on, and even factor in climate projections for that region.
00:04:12.057 --> 00:04:16.597
Now, the AI is not just saying Category 3 equals X. Instead,
00:04:16.857 --> 00:04:19.577
it helps you design hyper-precise parametric triggers.
00:04:20.357 --> 00:04:24.697
Different payouts for different bands of wind speed adjusted for local vulnerability
00:04:24.697 --> 00:04:28.817
and
Load full transcript
recalibrated over time as more events and claims data come in.
00:04:29.337 --> 00:04:35.097
That means better product accuracy, fairer pricing, and more resilient portfolios
00:04:35.097 --> 00:04:37.137
that adapt as the environment changes.
00:04:37.737 --> 00:04:43.317
Right. And that adaptability is huge. because one of the reasons parametric
00:04:43.317 --> 00:04:48.257
insurance sometimes gets a bad reputation is when the trigger doesn't line up
00:04:48.257 --> 00:04:50.617
with the actual experience on the ground.
00:04:51.397 --> 00:04:54.857
We'll come back to that in the basis risk section.
00:04:55.417 --> 00:05:02.697
But the key idea here in part one is agentic A, I turns risk assessment from a static,
00:05:03.097 --> 00:05:09.217
backward-looking exercise into a living learning system that continuously refines
00:05:09.217 --> 00:05:11.037
how you understand climate risk.
00:05:11.737 --> 00:05:14.517
Anything you'd add before we move to claims, Rakesh?
00:05:15.677 --> 00:05:19.877
Just one quick point. This also allows more frequent product updates.
00:05:20.737 --> 00:05:23.697
Historically, launching or tweaking a product could take months.
00:05:24.057 --> 00:05:29.877
With a modern AI-enabled parametric setup, you can iterate, pilot in one geography,
00:05:30.337 --> 00:05:34.997
learn from behavior and outcomes, and then refine your triggers or pricing for the next launch.
00:05:35.477 --> 00:05:37.717
That agility is critical in
00:05:37.717 --> 00:05:41.677
a climate environment that's changing faster than our old models assume.
00:05:42.717 --> 00:05:47.517
Let's talk about claims, arguably the most emotional part of the insurance journey.
00:05:48.657 --> 00:05:52.197
Traditionally, claims processing has been one of the biggest pain points.
00:05:52.557 --> 00:05:57.957
It's slow, it's heavily manual, it's costly, and in a big event,
00:05:58.237 --> 00:06:02.217
it can feel overwhelming for both the insurer and the insured.
00:06:03.136 --> 00:06:08.456
With agentic AI and parametric triggers, that picture changes dramatically.
00:06:08.996 --> 00:06:12.536
Imagine this. A cyclone hits a coastal city.
00:06:13.516 --> 00:06:17.616
Wind speeds recorded by trusted data sources exceed the agreed threshold.
00:06:17.956 --> 00:06:22.456
The system automatically cross-checks those readings against the policy parameters.
00:06:22.836 --> 00:06:27.476
And as soon as everything lines up, payouts are triggered, sometimes within
00:06:27.476 --> 00:06:30.396
hours, even without a traditional claim being filed.
00:06:31.376 --> 00:06:36.516
Rakesh. Walk us through what agentic AI is doing behind the scenes there.
00:06:37.156 --> 00:06:40.276
There are a few layers. Event detection.
00:06:40.636 --> 00:06:44.916
The AI agent monitors multiple data streams, meteorological feeds,
00:06:45.336 --> 00:06:48.296
satellite data, maybe even local IoT sensors.
00:06:48.656 --> 00:06:52.116
It identifies that a climate event matching the policy's domain,
00:06:52.416 --> 00:06:55.256
say cyclone or extreme rainfall, is occurring.
00:06:55.836 --> 00:07:00.076
Trigger validation. It checks whether the predefined parameters have been met.
00:07:00.076 --> 00:07:02.556
For example, wind speeds above
00:07:02.556 --> 00:07:06.376
X miles per hour sustained for Y minutes within this geo-fenced zone.
00:07:06.936 --> 00:07:13.136
Policy matching. It maps impacted locations back to covered policies and evaluates
00:07:13.136 --> 00:07:16.136
which policies, what limits, what payout structure apply.
00:07:17.036 --> 00:07:22.276
Decision and payout once everything checks out. It initiates payouts automatically.
00:07:22.856 --> 00:07:28.176
No manual review for standard cases and only escalates the exceptions.
00:07:28.996 --> 00:07:32.036
So the AI is not just detecting the event.
00:07:32.356 --> 00:07:37.176
It's validating it against policy terms, checking data authenticity,
00:07:37.436 --> 00:07:39.076
and initiating the payout.
00:07:39.516 --> 00:07:44.036
And that's such a radical improvement over call center, plus forms,
00:07:44.276 --> 00:07:46.416
plus assessor visit, plus waiting.
00:07:46.416 --> 00:07:51.056
For customers, particularly those who've just experienced catastrophic events,
00:07:51.416 --> 00:07:56.616
getting money quickly can be the difference between rebuilding versus shutting
00:07:56.616 --> 00:08:01.856
down, staying afloat versus defaulting, or even basic safety and shelter.
00:08:02.821 --> 00:08:07.661
So for them, it's not just efficiency, it's peace of mind and dignity.
00:08:08.441 --> 00:08:13.801
Exactly. And from the insurer's side, you also gain consistency in how similar
00:08:13.801 --> 00:08:18.421
events are treated, lower operational costs per claim, and richer data on how
00:08:18.421 --> 00:08:20.541
events unfold and how policies respond.
00:08:20.901 --> 00:08:24.121
That feeds back into better product design and better risk management.
00:08:24.621 --> 00:08:30.381
Let's move to verification, because this is where trust and regulatory comfort really come into play.
00:08:31.187 --> 00:08:35.387
One concern people often have with automation is, how do you know the data is
00:08:35.387 --> 00:08:38.127
right? And what if someone tries to game the system?
00:08:38.927 --> 00:08:43.387
Agentic A, I can add tremendous value here by combining computer vision,
00:08:43.667 --> 00:08:46.727
natural language processing, and cross-data validation.
00:08:47.507 --> 00:08:50.587
Rakesh, give us a sense of what that looks like in practice.
00:08:51.147 --> 00:08:56.067
Think of a layered defense. At the data level, the AI cross-checks multiple
00:08:56.067 --> 00:09:01.567
independent sources. satellite imagery, ground sensors, meteorological data.
00:09:01.947 --> 00:09:06.287
At the policy level, it validates that the event matches what's actually covered.
00:09:06.667 --> 00:09:09.287
Correct location, correct dates, correct peril.
00:09:09.807 --> 00:09:14.307
At the customer level, if needed, it can analyze photos, documents,
00:09:14.407 --> 00:09:16.027
or descriptions for consistency.
00:09:16.347 --> 00:09:21.167
For example, if a farmer in a parametric drought program uploads photos or descriptions,
00:09:21.167 --> 00:09:25.587
a Gentic AI can use computer vision to understand crop condition,
00:09:25.827 --> 00:09:28.247
use NLP to interpret text descriptions,
00:09:28.507 --> 00:09:31.967
and cross-validate with rainfall or soil moisture data for that location.
00:09:32.487 --> 00:09:37.247
If something looks off or inconsistent, they can flag it for human review.
00:09:38.127 --> 00:09:42.707
So you're essentially getting speed for straightforward cases and extra scrutiny
00:09:42.707 --> 00:09:46.367
for edge cases without making everyone go through friction.
00:09:46.707 --> 00:09:48.947
And that's critical for building trust.
00:09:49.587 --> 00:09:53.927
Fewer disputes, lower fraud, faster settlements, and better relationships with
00:09:53.927 --> 00:09:57.867
regulators because you can show your logic and your audit trail.
00:09:58.627 --> 00:10:03.387
Yes. And something I really like here is it frees up human adjusters.
00:10:03.847 --> 00:10:07.687
Instead of spending their time on small straightforward claims that could be
00:10:07.687 --> 00:10:12.467
automated, they can focus on complex cases, large commercial accounts,
00:10:12.607 --> 00:10:15.507
or situations where empathy and judgment really matter.
00:10:15.827 --> 00:10:20.967
So AI doesn't replace them, it amplifies them All right, time to talk about
00:10:20.967 --> 00:10:24.987
one of the toughest topics in parametric insurance, basis risk.
00:10:25.894 --> 00:10:30.474
For anyone new to that term, basis risk is what happens when the parametric
00:10:30.474 --> 00:10:35.114
trigger doesn't perfectly align with the actual loss experienced by the customer.
00:10:36.067 --> 00:10:40.767
So you can end up with a payout even when the customer didn't suffer much damage,
00:10:40.767 --> 00:10:44.847
or worse, no payout even though the customer clearly experienced a loss.
00:10:45.327 --> 00:10:48.527
That's where parametric covers have historically been criticized.
00:10:49.427 --> 00:10:55.307
Rakesh, how does Agenic AI help with this? First, it helps by reducing the gap
00:10:55.307 --> 00:10:57.647
between the trigger and the actual experience.
00:10:58.367 --> 00:11:03.607
Agentic AI can continuously learn from live event data, compare that against
00:11:03.607 --> 00:11:07.407
claims, outcomes, and customer feedback, and then recalibrate triggers over
00:11:07.407 --> 00:11:09.647
time. Let's take a simple example.
00:11:10.087 --> 00:11:14.547
Suppose you have a rainfall-based parametric product. The trigger is less than
00:11:14.547 --> 00:11:19.347
X millimeters of rain over Y days, but you observe that farmers are still reporting
00:11:19.347 --> 00:11:24.427
losses when rainfall is slightly above that threshold due to local factors—soil type,
00:11:24.587 --> 00:11:27.087
irrigation issues, microclimate, and so on.
00:11:27.647 --> 00:11:31.727
Agentic A, I can discover patterns in, where mismatches occur,
00:11:31.987 --> 00:11:37.067
what local variables seem to matter, and how often payouts didn't match the actual experience.
00:11:37.407 --> 00:11:41.707
It can then suggest adjusting thresholds for specific microregions,
00:11:41.907 --> 00:11:45.387
layering in additional parameters like soil moisture or temperature,
00:11:45.607 --> 00:11:48.287
or revising payout curves to better match reality.
00:11:48.867 --> 00:11:52.507
Over time, the models evolve and the basis risk shrinks.
00:11:52.507 --> 00:11:57.687
So we move from a one-size-fits-all threshold to something a lot more nuanced
00:11:57.687 --> 00:12:03.067
and data-driven, and that leads to fairer and more reliable experiences for
00:12:03.067 --> 00:12:04.807
both insurers and policyholders.
00:12:05.327 --> 00:12:09.207
Because frankly, if customers keep feeling like the trigger doesn't represent
00:12:09.207 --> 00:12:12.007
their reality, they won't renew. Exactly.
00:12:12.547 --> 00:12:16.607
And it also helps regulators feel more comfortable because you can demonstrate
00:12:16.607 --> 00:12:20.607
how you designed the triggers, how you're monitoring performance,
00:12:20.807 --> 00:12:23.087
and how you're improving them based on evidence.
00:12:23.507 --> 00:12:28.867
That's a much more compelling story than we set the threshold once and never revisited it.
00:12:29.702 --> 00:12:34.282
Let's talk about the upside, new business opportunities. Parametric insurance
00:12:34.282 --> 00:12:40.042
powered by Agentic A, I is not just about doing the same thing a little bit better.
00:12:40.242 --> 00:12:43.102
It's expanding the boundaries of what's insurable.
00:12:44.583 --> 00:12:48.423
We're seeing opportunities in segments like smallholder farmers,
00:12:49.003 --> 00:12:54.143
renewable energy operators, tourism and hospitality supply chains vulnerable
00:12:54.143 --> 00:12:59.943
to climate volatility, and even public sector risk transfer and resilience programs.
00:13:00.863 --> 00:13:04.223
Rakesh, pick one or two examples and walk us through.
00:13:04.743 --> 00:13:09.683
Let's start with smallholder farmers. In many markets, they're highly exposed
00:13:09.683 --> 00:13:14.363
to climate risk, often underinsured or entirely uninsured and difficult to serve
00:13:14.363 --> 00:13:19.503
through traditional indemnity products because admin costs are high and data is limited.
00:13:19.963 --> 00:13:25.963
Parametric products combined with agentic AI allow you to use satellite imagery
00:13:25.963 --> 00:13:30.543
and weather data to assess drought or flood events, set simple transparent triggers,
00:13:30.763 --> 00:13:34.843
and push payouts directly to farmers' mobile wallets when conditions are met.
00:13:35.303 --> 00:13:41.023
AI helps by
scaling this across millions of small policies, continuously learning
00:13:41.023 --> 00:13:42.443
which triggers work best.
00:13:42.543 --> 00:13:45.523
In each region, and optimizing portfolio performance.
00:13:45.943 --> 00:13:51.843
That's financial inclusion plus sustainable business. Another example, renewable energy.
00:13:52.243 --> 00:13:57.343
Wind farms, solar parks, hydro plants. They all depend heavily on weather.
00:13:57.883 --> 00:14:02.283
Parametric covers can protect revenue when wind speeds are too low or too high,
00:14:02.483 --> 00:14:07.343
output when solar radiation is below expected levels, or water levels for hydro plants.
00:14:07.923 --> 00:14:12.963
Agentic AI ingests real-time performance data, correlates it with weather conditions,
00:14:13.363 --> 00:14:14.983
helps design covers that align
00:14:14.983 --> 00:14:19.043
with actual revenue risks, and automates payouts when deviations occur.
00:14:19.423 --> 00:14:23.703
You get products that are not only innovative, but also scalable and manageable
00:14:23.703 --> 00:14:24.923
from a risk perspective.
00:14:25.503 --> 00:14:30.003
That's such a powerful narrative. It's not just innovation for efficiency sake,
00:14:30.143 --> 00:14:31.903
so it's innovation with impact.
00:14:32.853 --> 00:14:36.553
You're enabling coverage for segments that were traditionally seen as too risky
00:14:36.553 --> 00:14:39.913
or too expensive to serve with traditional indemnity insurance.
00:14:40.733 --> 00:14:45.293
And for carriers, that's also a way to differentiate in the market and build
00:14:45.293 --> 00:14:50.633
new partnerships and align with sustainability and ESG goals. Let's look ahead.
00:14:51.473 --> 00:14:55.493
Agentic A. I won't just influence underwriting or claims in isolation.
00:14:55.593 --> 00:15:00.113
It will reshape how insurers think about capital allocation,
00:15:00.593 --> 00:15:03.393
reinsurance structures, and overall portfolio resilience.
00:15:03.993 --> 00:15:07.093
Rakesh, how do you see this playing out over the next few years?
00:15:07.533 --> 00:15:11.813
I see three big shifts from static to dynamic portfolios.
00:15:12.033 --> 00:15:16.193
Instead of setting risk appetite once a year, carriers will use AI to continuously
00:15:16.193 --> 00:15:20.153
monitor climate exposure and adjust portfolios in near real-time,
00:15:20.333 --> 00:15:22.373
by region, by product, by segment.
00:15:22.973 --> 00:15:27.553
Smarter reinsurance and risk transfer. At Gentic A, I will help carriers structure
00:15:27.553 --> 00:15:33.013
parametric covers, not only for end customers, but also for their own reinsurance and ILW programs.
00:15:33.233 --> 00:15:35.033
They'll be able to simulate scenarios
00:15:35.033 --> 00:15:39.113
more accurately and negotiate better terms backed by robust data.
00:15:39.873 --> 00:15:44.733
Integrated resilience strategies. Insurers won't just be payers of claims.
00:15:44.933 --> 00:15:48.393
They'll be partners in
resilience, advising governments, businesses,
00:15:48.393 --> 00:15:50.793
and communities on how to mitigate risk.
00:15:51.433 --> 00:15:54.833
Parametric Plus A becomes a tool for planning, not just for repairing.
00:15:55.453 --> 00:15:58.033
There's also a capabilities and governance angle here.
00:15:59.005 --> 00:16:02.765
If you want to be an AI-first insurer, especially in climate risk,
00:16:03.165 --> 00:16:07.985
you need the right data foundations, clear model governance at the right talent
00:16:07.985 --> 00:16:11.925
mix, actuaries, climatologists, data scientists, engineers, business leaders,
00:16:12.345 --> 00:16:14.565
and strong ethics and transparency frameworks.
00:16:15.125 --> 00:16:19.225
Because these are powerful tools, you want to make sure they're used in ways
00:16:19.225 --> 00:16:22.985
that are, one, fair, two, explainable,
00:16:23.105 --> 00:16:27.625
and three, aligned with your brand promise. I completely agree.
00:16:27.965 --> 00:16:32.625
The key message for insurers is climate risk is permanent, but your ability
00:16:32.625 --> 00:16:34.265
to manage it doesn't have to be static.
00:16:34.785 --> 00:16:39.465
With agentic AI, you can move from reactive, manual fragmented approaches to
00:16:39.465 --> 00:16:42.185
something predictive, adaptive, and responsible.
00:16:42.665 --> 00:16:46.965
It won't happen overnight, but the ones who start now will be in a much stronger
00:16:46.965 --> 00:16:52.585
position a few years down the line. That brings us to the end of today's episode of the TX Show.
00:16:53.005 --> 00:16:57.165
We've explored how agentic AI is transforming parametric insurance,
00:16:57.405 --> 00:17:01.705
making payouts faster, making them fairer, and making them more predictable
00:17:01.705 --> 00:17:03.085
in the face of climate change.
00:17:03.225 --> 00:17:08.145
We talked about AI-powered risk assessment, automated claims and verification
00:17:08.145 --> 00:17:11.625
of mitigating basis risk, unlocking new business opportunities,
00:17:11.805 --> 00:17:15.485
and what the future of AI and climate risk might look like.
00:17:15.485 --> 00:17:23.045
If you're an insurer, an MGA, a broker, or even a regulator listening to this, the takeaway is simple.
00:17:23.685 --> 00:17:27.485
Parametric insurance plus agentic AI isn't science fiction anymore.
00:17:27.625 --> 00:17:30.925
It's a practical toolkit you can start experimenting with today.
00:17:31.605 --> 00:17:36.405
Join us next time as we discuss what it truly means to become an AI-first insurer
00:17:36.405 --> 00:17:40.705
and how forward-thinking carriers are already getting there in the real world.
00:17:41.405 --> 00:17:45.445
Rakesh, thanks for another great conversation. Thanks, John.
00:17:45.625 --> 00:17:47.165
And thanks to everyone listening.
00:17:47.745 --> 00:17:51.345
If you found this helpful, share it with a colleague who's thinking about climate
00:17:51.345 --> 00:17:54.845
risk or parametric products. We'll see you in the next episode.