If you missed Part 1: our CTO - mid-size carrier, chin-deep in last year’s “transformation” priorities - has been handed a 20-minute slot at the next board meeting to answer the question “What are we doing about AI?” We left her at 11pm, staring at a blank PowerPoint, heading to bed ahead of a 7:30am internal audit meeting about folder permissions in SharePoint.

The Next Night

The audit meeting came and went. (Two findings, one repeat. The remediation plan is due Friday. Will we ever be rid of SharePoint?) The day filled itself the way days do - steering committee, vendor escalation, one-on-ones - and now it is 10pm again, the house is quiet again, and the PowerPoint is still blank.

But something useful happened in the shower this morning, which is where all real strategic planning occurs. She realized she doesn’t have writer’s block. She has two decks in her head, both fully formed, fighting over one set of slides. So tonight she does something she would never admit to in a status meeting: she writes both.

What follows are those two decks, presented as faithfully as I can manage. In the spirit of the choose-your-own-adventure books this series has already invoked, I am going to do something unnatural for me and withhold judgment entirely. No bias, no consequences, no editorializing about which one is “right.” That comes in Part 3. Your only job is to read both and decide, honestly, which deck you would carry into that boardroom.

Path One: The Roadmap

The first deck nearly writes itself. Twenty years of muscle memory sit behind it, and every slide has an ancestor in a deck that once got approved.

It opens with what is already happening - because things are, in fact, already happening. Copilot is deployed, and she has the adoption statistics: licenses issued, monthly active users, a usage curve pointed encouragingly up and to the right. There are pockets of real enthusiasm - a few underwriters, a group in claims - employees using actual AI to do their jobs better today.

Next, an inventory of the AI already embedded in the company. The claims platform, the underwriting workbench, the CRM: every vendor touts AI capabilities, and several of the claims are even true! The message is a sensible one: our existing investments carry AI along with them, and we should harvest what we have already bought.

Then the candid part. There is no approved funding for a significant AI initiative - the current plan was locked in before the ground shifted. So the deck makes the honest, procedural request: additional funding to stand up a focused effort.

That effort takes a form the board will recognize. Identify and evaluate use cases within our existing business processes - the incremental, practical opportunities - and return in three to six months with a prioritized roadmap, each opportunity tied to the current business strategy, with a business case wherever one can reasonably be built. And because the internal team is fully committed to in-flight priorities, an outside firm will be engaged to help prioritize and deliver. They arrive with benchmarks, cross-industry pattern recognition, and arms and legs the current org doesn’t have.

She reads it end to end. It answers the question the board asked. It fits in twenty minutes. It respects the planning cycle, the governance model, and the credibility she has spent years building by not over-promising. Every commitment in it can be met.

She saves the file.

And then she opens a new one - this deck will be written by the version of her standing on the beach, watching the water pull away from the shore with the knowledge of what comes next…

Path Two: The Reset

The second deck is harder to write, because there is no template for it. It has fewer slides and more questions. It admits shortcomings, and opens with the biggest one:

We are not structured, staffed, or funded for what is coming.

It says plainly that the entire technology organization is deployed against last year’s priorities, which are only now limping toward the finish line with the usual pain and friction - and that this focus was rational when it was made, but is rational no longer.

It says that Copilot is not the endgame. Neither is Claude, nor whichever model tops the benchmarks this month. The tools everyone is experimenting with are a floor, not a destination, and adoption dashboards measure activity, not transformation.

It says that the real opportunity is not sprinkling AI onto existing processes, but taking a ground-up look at workflows and staffing across the whole company - underwriting, claims, service, finance - not just IT. This is not a technology program with business sponsors; it is a business redesign with technology at the center.

And it explains what moving fast would truly mean. Unplanned investment in the middle of a budget year. A new - and disruptive - talent strategy. And the potential write-down nobody wants to say out loud: accepting that some of our recent capital investments no longer hold up against the business cases that justified them, however sound those cases were when they were approved.

The last slide says what the whole deck has been building toward:

This is a hard reset.

Cauterize the past.

Start the future on a clean sheet.

She reads this one end to end, too. It also answers the board’s question - just a more dangerous version of it. Not “what are we doing about AI?” but “what would we do if we took our situation seriously?” It is the deck that matches the harsh truth as she sees it. It is also the deck with no precedent, no playbook, and no line in the budget.

The worst thing: it runs contrary to everything she has previously presented to the board to support tens of millions of dollars of prior investment. It reads as if she were the new CTO throwing the previous CTO under the bus.

She saves this file too, never having focused so intensely on hitting Ctrl-S.

Two Decks, One Future

It’s past midnight. Two decks sit side by side in a folder. Both are true. Both are incomplete. Both can be delivered with a straight face by a competent executive - which is precisely what makes this a fork in the road rather than a decision with an obvious answer.

So many contradictory emotions spin in her mind. Fear, exhilaration. Confidence, insecurity. A pull toward vision, a deep sense of loyalty to those who have gotten her where she is today.

She closes her laptop. She has to submit her draft to the CEO tomorrow for review. She knows that she will not sleep well tonight.

Your Turn

I promised no judgments in this installment, and I’m keeping that promise - but only until the next one. In Part 3, I’ll walk each path forward: the benefits and the risks, what each one might cost in the short run and the long run, for the company and for the CTO who must stand behind it. Neither path is easy, and the price tags are not printed where you’d expect.

Before I write it, I want to know: which deck would you present? Leave a comment with your views - and more importantly, tell me why. If you have sat in some version of this meeting, on either side of the table, I especially want to hear how it went. The best arguments (and the best dissents) will be incorporated, with credit, into Part 3.

-Darryl

The stock market can sometimes overreact or be fickle. The recent sector wide multiple compression for SaaS stocks is notable considering they were the darlings for quite a while.

I understood the rationale at an intellectual level, but a recent experience made me realize first hand what these companies are dealing with.

Take Hubspot for example. Great company, great product. In the past when I have moved from the company side to an independent advisor or during a job search, it was very easy to spin up HubSpot to use as my personal CRM to manage all the outreach and opportunities in an organized way.

This time, I just used a project in Claude co-work, and within half a day I had a system in place that far exceeded the capability and usefulness of HubSpot for what I want to do. And I’m sure it will continue to evolve more to support my business development activities.

I just started with a simple prompt and took it from there. Now I can add contacts, companies, and rich data about every entity. Claude keep track of my opportunities, gives me a morning brief every day, and interacts with my email and calendar.

All of this is basic stuff, but the difference is that Claude is actively helping me add context and make connections within my network, and I can interact with it in ways that are more natural and satisfying than maintaining something that was designed to meet an entire market’s needs, not my specific needs or preferences.

Granted, this is a corner case, and not an enterprise implementation. But I think the issue extends to enterprise decisions. If you had asked me 3 years ago if a new MGA should build a proprietary stack or just buy something off the shelf I would have confidently advised the latter. Now I think the obvious answer is to build, if your team has the ability to.

These SaaS giants aren’t going to just sit back and let themselves become obsolete. They have a huge customer base and many are deeply integrated into their customers’ overall tech stack and users. They have the resources to respond and evolve.

But the multiple compression reflects the reality that they are now in competition with foundation models which are rapidly evolving and operate on a different plane. Pricing models and costs will have to adapt, and enterprise customers are no longer going to feel that a third-party SaaS is the only viable option.

This is the first installment in a series on AI transformation. Over the next several posts I want to explore the organizational dynamics and the very practical dilemmas of this extraordinary moment - and to do it the way it actually unfolds inside a company: through the eyes of the people in the C-suite who have to decide what to do, and when.

Let’s start at the top of the house.

The View From the Boardroom

Picture a director on the board of a mid-size carrier. Thirty years in the industry, most of it in underwriting and finance, with a stint running a P&C book that taught her more about risk than any model ever could. She is sharp, she is skeptical by training, and she is not a technologist. For most of her career, that last fact was never a problem.

Then ChatGPT arrived, and since that moment AI’s importance to the strategic business agenda has built like a tsunami, in full view of everyone.

Her first instinct was the right one for a risk professional: skepticism. The early conversations centered on hallucinations, data privacy, security, and governance - exactly the concerns you would expect from people whose careers were built on prudence. Those concerns were, and are, legitimate. But if she is honest with herself, they also did some emotional work: they were a way to blunt what looked like a hype cycle, and to manage a quiet fear of the unknown.

What changed her posture was not a vendor pitch. It was velocity. The sheer momentum and rate of improvement of the models - and the tooling around them that makes them genuinely useful to “Normals” and not just engineers - created a different mindset almost overnight. The technology stopped being something she read about and became something she could pick up and try herself.

The outside signals reinforced it. Watching CNBC and reading the Wall Street Journal, she sees immense capital being poured into the sector - a signal that something is afoot. Early headlines of massive staff reductions piqued her interest further. Here, though, the old underwriter in her hesitates: many of those announcements are window dressing, manufacturing the “evidence” of success well ahead of the reality of AI’s actual role in the cost reduction. She has seen enough restructurings dressed up as strategy to know the difference.

And yet - and this is the part that matters - it doesn’t much matter whether the competitor’s number is real. The questions about governance and risk she started with were quickly amplified by a second, sharper set of questions about obsolescence and competition. Together they did something remarkable: they moved AI from the innovation update near the back of the deck to the very top of the board agenda.

Now flip the camera around. Because every one of those questions, however reasonable in the boardroom, lands on a single desk.

The Email

It begins, as these things do, with an email. “The board wants to know what we are doing about AI.” The next board meeting is three weeks away. The CTO has been assigned a 20-minute slot.

The timing could not be worse. She is chin-deep in the implementation of last year’s “transformation”¹ priorities, under steady pressure to show tangible results. The resource allocation of her technology organization does not match this sudden change of agenda - the team and the org structure she carefully assembled to deliver a portfolio of third-party solutions is simply not built for what is now being asked of it.

It is part of the job to translate technical issues for non-technical people and to plan strategically for change. But she has also been watching the same tsunami, and is blessed - or cursed - with a deeper understanding of just how transformative this change will be. To borrow the metaphor without straining it: she is the one standing on the beach watching the water recede, while everyone else wanders out onto the wet sand to explore, delighted, not realizing what comes next.

This gap in understanding is normal, and in insurance it is almost structural. Boards are composed largely of finance, underwriting, and HR experts. If there is a former technology executive in the room, the odds are good that they learned their hard lessons - and won their victories - in a different era, when “transformation” meant a new policy administration system, and the cycle time was measured in years, not weeks.

The Blank PowerPoint

It’s 10pm, and the board materials are due in a week. This week was like every week: the calendar jammed back-to-back with status meetings, steering committees, risk meetings, compliance meetings, and one-on-ones². The only time she gets to actually think is after the kids are asleep and email and Teams have finally gone quiet.

To make matters worse, the CFO has forwarded an investor deck. A competitor has announced that its AI initiatives have already delivered a 20% efficiency improvement, and that they are reducing headcount. He copies the CEO, who replies: “This is what we should be covering at the board.”

It is, of course, the very same kind of claim the board’s own director privately suspected was window dressing. But a number in a competitor’s deck does not have to be true to be dangerous. It only has to be believed.

Exhausted and distracted, staring at a blank PowerPoint, our CTO is facing a moment of truth, whether she realizes it or not. Like a choose-your-own-adventure book, the next few decisions may permanently alter the company’s path - and her own career.

The maddening part is that two opposite things are true at the same time.

On the one hand, it has never been more thrilling to be a technology leader in insurance. Despite the historical positioning of IT as a back-office service provider - much to the chagrin of insurance CTOs everywhere - this time the revolution is playing out in everyone’s pocket, in everyone’s own late-night experiments with ChatGPT. At long last, the rest of the business is waking up to the reality that technology is foundational to strategy and competition. Or at the very least, that IT is not just a cost center.

On the other hand, technology has never moved this fast, while the industry’s clock speed for absorbing it has remained stubbornly slow - throttled by bureaucracy, governance, resistance to change, and an intense focus on risk. The conversation seems to have changed overnight to “What have you done with AI?” - when the entire focus of the last planning cycle was finishing the core system implementation (currently a year late) and, of course, reducing cost in IT.

It’s 11pm. Her head is spinning and the PowerPoint³ is still blank. She closes the laptop and decides to sleep on it for one more night. She needs the rest, because internal audit has scheduled a 7:30am meeting to review their findings on folder permissions in SharePoint.

To Be Continued

In the next installment, I’ll follow our beleaguered CTO into the fork in the road - the genuinely different paths she can take from this critical juncture, and what each one costs her in the short run and the long run. It would be comforting to believe there is a single “right path forward.” Unfortunately, the velocity of the change, its implications for the business, and the insurance industry’s peculiar history with technology innovation together mean that the right path forward is anything but clear.

-Darryl

Coming Soon in “Hot Takes”

A few things I couldn’t resist needling in this post. Each is getting its own short, opinionated installment. If one of them made you nod (or wince), tell me which - I’ll move it up the queue.

1. “Transformation” has lost all meaning. Businesses are always evolving and transforming, it is not a project with an end date. “Transformation” programs are often not transformative and delegate the natural process of continuous change, either to an internal “transformation office” or, even worse, external consultants.

2. The compliance, risk, and audit tax. Each meeting is individually justifiable and collectively suffocating. We have professionalized the control function so thoroughly that the people we most need thinking about the future spend their days accounting for the past. A defense of compliance, paired with a suggested tweak, is coming.

3. The slide is not the strategy. Somewhere along the way we built an industry that judges the quality of the thinking by the quality and quantity of the PowerPoint deck, where “let me put something together” quietly eats many hours that should have gone to actual work. There are better ways to communicate.

Sam Lessin, of Slow Ventures and one of my favorite podcasts, “More or Less”, vibe coded a simple website that shows the prices of tokens from various models.

When I first saw the wide variation in token prices I was surprised. But I am curious as to what to conclude.

Open AI and Anthropic are the premium priced tokens, although competition between the two leaders may drive that down. (WSJ/Paywall)

In comparison, Xai is dirt cheap. Is it a bargain alternative or is it (as I suspect) the market signaling its lack of value. I’ve never used Grok so I can’t offer a first hand opinion.

Google is interesting. Gemini is a powerful LLM with a lot of usage (although it’s hard to compare apples to apples with Google’s embedded AI in search and Google Apps.). Google has deep pockets so their mid-range pricing might be a very intentional positioning to gain share with developers.

In any case, the cost of tokens and the future ramifications for Insurance Enterprises is something I think is not broadly understood in the C-Suite. I’ll be writing more on that topic in the future.

Douglas Ver Mulm notes “Corgi is bad for all the great tech-forward and Insurtech MGAs/wholesalers/carriers out there”

Alexander Niehenke notes “If momentum investing and stupidity had a bastard child, it would be vibe investing”

Corgi’s job posting for Chief of Staff, Trucking Insurance, says:

“The $40B+ trucking insurance market is still being underwritten by people with paper files and 1990s rating engines. We're going to fix that.”

Well, Progressive has 50% market share in trucking and you can get a binding quote online here. Also check out Cover Whale, Nirvana, even GEICO

Also from the job posting:

“In year one, expect to help build a division from scratch to $1B in premium”

To be fair, if you underprice the product in a highly price sensitive, non-loyal, high churn segment, you can definitely grow fast!

The losses will likely be spectacular, and while RRGs enable high capital leverage, that leverage works against you when the losses are high, wiping out the surplus capital. And with no state backstop, it can easily end in tears.

Never before have I seen the pace of technology move so fast, and with such large implications for our industry across all parts of the value chain. I believe that our industry is transforming before our eyes and the opportunity for accelerating value creation is huge.

As an operating executive, that pace of change can feel exasperating. After a full day of meetings, you try to catch up on the latest developments and more has changed in a single day than you can possibly hope to absorb, much less integrate into your strategy or concrete plans.

If you are like my mom, you might place your palm on your forehead and say “Aye, yai yai!” I don’t blame you.

Todays insurance executives are stuck between the realities of their operating model, the legacy of technology disappointments, and the realization that if they don’t move fast to keep up with the brutal realities of change, they may be left behind.

That’s what this Substack is about.

As a long time insider, I hope to use my industry knowledge, my tech & ops experience at companies ranging from startups to large scale enterprises, and my understanding of the realities of how hard it is to make real change happen. I hope to create a place for discussions that my readers can actually use to navigate these challenging times and come out the other end with real insights that they can put into action.

I also hope to foster a community of thought leaders that don’t just play by the traditional rules of insurance and technology. Leaders that operate beyond the shackles of an industry that has viewed technology as a subservient service provider and not as the key to unlocking the next wave of value creation.

People outside our industry tend to think we are a fat, dumb target for easy disruption. My readers know that insurance is complicated and change is hard. However, every tech thought leader in insurance realizes that the rules have changed, and that the clock is ticking.

The time has come for us to embrace the reality that technology (and it’s symbiotic partner in operations) will do more to unlock value in our industry than any other discipline because the pace of change in technology is unprecedented and will inevitably change everything about how insurance works.

A lot of people are once again convinced that the insurance broker is about to be obsolete. A Sequoia article on AI’s impact on brokerage moved markets earlier this year. Citrini Research painted a picture of a future without intermediaries. Boards and investors have taken note of this, and it has generated a lot of discussion.

“The Industry is History” - Again?

I keep coming back to a single observation: this all reminds me of the late nineties, when there was a similarly confident conviction among the Technorati that the internet would destroy independent agents. As time went by, insurance distribution did evolve — primarily in personal auto — but things did not play out the way people predicted during the internet boom.

Indeed, I think AI will be massively transformational for insurance. I also think we are very early in the hype cycle. You can be in a hype cycle with a massive bubble and still have longer-term transformational change underneath it. The internet build-out in the late 90s is the obvious example. Both things can be true.

So with that in mind - what’s actually different, and what isn’t, when we layer AI onto this old debate?

The Difference Between E*Trade and eCoverage

I think it’s always important to focus on consumer behavior: the primary reason consumers, particularly at the lower end of the market (including commercial, where the small business buyer behaves more like a consumer) did not meaningfully shift to direct online over the last decades is simply because

a) they have absolutely no desire to deal with the insurance purchase at all and

b) the price of the broker is invisible to them (and online products turned out not to be cheaper)

Securities brokerage, by contrast, was massively disrupted. Consumers moved en masse to online trading, and the value captured by the middleman for simple trades evaporated. The distinction is that consumers love doing their own trading. It’s fun, addictive, and can make you rich.

Insurance is none of those things.

Ironically, this aligns with Sequoia’s Roelof Botha’s “seven vices” philosophy of investing in the post-internet era — bet on the durable human desire to indulge.

E*Trade thrived. eCoverage didn’t. The reasons weren’t technological.

“This Time it’s Different” — Maybe

“This time it’s different” may indeed apply, because the internet only provided direct access to consumers. AI can potentially do something much harder: handle significantly complex transactions, reason, advise, and negotiate.

If AI can manage all of that, will consumers at the low end of the market care?

I think there’s a good chance most of them won’t. There will be agentic-AI-powered brokers, and some segment of the market will migrate to them — just as some segment migrated to online direct. But I suspect that channel will cannibalize direct online market share more than it cannibalizes the traditional broker. The psychographic profile is the same. The people who wanted to “do it themselves” already did. The people who didn’t, likely still don’t.

Stubborn Consumers vs. Professional Buyers

That got me thinking that the real AI disruption to insurance brokerage may not be at the lower end of the market at all. It may be in the middle market.

The key difference is that middle market buyers are professional buyers. They have an intrinsic motivation to manage their insurance costs and risk. The products at that level are still not impossibly complex for an educated buyer. And there may be significant cost advantages to bypassing the broker channel — advantages that an AI agent can actually surface and act on.

The internet-era promise of “cutting out the middleman” never came true for small commercial because the cost of the agent’s commission was simply substituted by Google Search and other customer acquisition costs. Those costs were so high that most insurtechs that started out direct-to-small-business eventually pivoted to selling through agents.

AI changes that math. Not by replacing the broker’s relationship with the small business owner — that relationship is, again, mostly a relief from having to think about insurance at all. But by potentially making it economically viable for a professional middle-market buyer to get high-quality, multi-carrier placement without a traditional retail broker in the loop. The higher end of the market will still be dominated by big brokers.

A Double-Edged sword

I believe retail brokers — especially those weighted toward small commercial — have an enormous opportunity to expand EBITDA margins using AI. To the extent that disruption to the channel manifests as margin compression across the industry, the brokers who drive substantial internal efficiency can offset some or all of that pressure.

Predictably, the margin pressure will most likely take the form of lower carrier commissions over time. The significant differential between broker margins and carrier margins is not sustainable in a logical market. Unfortunately, even brokers who win the efficiency race will face topline headwinds and a lower overall TAM.

That dynamic is not great news for PE-backed roll-ups carrying heavy debt loads. The brokers in that position who don’t move quickly on internal AI-driven efficiency may find themselves squeezed from both sides — slower topline growth, with debt facilities that assumed a different, more predictable world. That risk has been making the rounds among industry peers for good reason.

So, Will AI Eat the Broker?

My one-line answer: no, AI will not obsolete the retail brokerage model. But it will radically transform the operating model — for the better, in the hands of operators who actually use the next few years to rebuild the workflows underneath. I believe that larger brokers can use AI and automation to reduce service costs by 50-70%.

Negative market sentiment may persist ahead of the actual outcome, and that’s a real risk to public-market valuations on a one-to-three year view. The fundamentals, on a longer view, favor operators who lean heavily into small commercial, where buyer behavior is most stubborn and the efficiency gains are most dramatic.

Of course, I might just be wrong. The brokerage industry could be eaten by AI in a way I am underestimating. Reasonable people can land in a different place on this, and I’d like to hear from those of you who do.

Much more to discuss. Everything is moving fast enough that I expect this won’t be the last time I write on the topic.

—Darryl

Subscribe now