Why 95% sticks out so badly

Here's what IMA Financial Group announced on August 18. The company has moved past AI experimentation to enterprise-wide adoption, over 95% of its 3,000-plus associates use AI daily, and thousands of agentic workflows are running internally.

Put that beside other data and the number leaps off the page.

Days earlier, Linear published its own product data. The function with the highest AI adoption was Product, at 34%. Engineering was 30%. And that's paying users of a developer tool popular with startups — already a biased sample. In the middle of the software industry, 30-something percent. An insurance broker is claiming 95%.

So the right first reaction to this announcement isn't admiration. It's a question: what exactly was counted to produce 95%? Follow that question through and the parts of this case worth learning from separate pretty cleanly from the parts that are just publicity.

Who's who — what kind of company is IMA?

IMA Financial Group is a North American insurance brokerage doing risk management, wholesale brokerage, and investment advisory work, with more than 3,000 associates across multiple offices.

One structural detail matters more than it looks: IMA is majority employee-owned. When employees own the company, resistance to a new tool changes character. Efficiency reads less as "this threatens my job" and more as "this raises the value of my stake." Given that the most common reason enterprise AI rollouts fail is not technology but quiet non-cooperation, that governance structure is a real variable.

It also helps to know what insurance brokering actually is. A broker doesn't sell insurance — a broker stands between a corporate client and carriers, structuring risk and negotiating terms. Which means a large share of the day is document work: reading policy language, comparing quotes from multiple carriers, checking how terms shifted from last year's contract, and producing a summary the client can act on.

That composition matters because it overlaps almost exactly with what AI is currently best at: reading long documents, comparing versions, generating summaries, extracting structured information. On the difficulty scale of AI adoption, insurance brokering is on the easy end.

Which recasts the 95% a little. This company may have executed an exceptional transformation — or the industry may simply have been favorable. Producing that number on a factory floor or in a logistics warehouse is a categorically harder problem. Comparing adoption rates across industries without that context invites misreading.

What the release actually says

Item Detail
Announcement date August 18, 2026
Headcount 3,000+
Daily AI usage Over 95%
Agentic workflows Thousands
Applications Research, analytics, document comparison, workflow automation
Internal org AI Studio
Strategy Platform-agnostic
Governance Majority employee-owned

The key sentence is the company's own framing: it embedded AI into everyday work while preserving the expertise and judgment that drive client outcomes.

The organizational vehicle is AI Studio, described as a place where associates, technologists, and AI specialists turn ideas and pilots into solutions that scale across the enterprise. Not IT buying tools and pushing them down — a structure for propagating what the front line builds.

CEO Rob Cohen's quote summarizes the philosophy.

"For IMA, AI is a people transformation, not a technology transformation; and there is no better example of that than the thousands of agentic AI workflows already put in place, led by the innovation of our associates."

VP and Director of Data and AI Megan Cullen-Meyer is more direct.

"IMA is not outsourcing how AI is applied across our business. Our associates understand our clients, our workflows and where human judgment matters most."

"Not outsourcing" is the real claim in this announcement — that this grew internally rather than arriving as a consultant-designed transformation program. After several years of consultancy-led AI transformations ending as expensive slide decks, leading with that distinction looks like deliberate positioning.

Why the structure matters becomes clear from the failure cases. The typical collapse goes like this: IT or an innovation group picks a tool, runs a few pilots, declares enterprise rollout. The front line doesn't know where the tool fits into their actual work, the old way is still faster, so they don't use it. A few months later the usage dashboard bends downward and the project quietly ends.

AI Studio inverts that order. Instead of choosing a tool first, the front line brings the painful part of its own workflow, and technical staff shape it into something that scales. The question "why should I use this?" never arises, because the person who built it is the person who uses it.

How far should you trust the number?

Let's be honest. This is a self-published company press release with no independent verification.

Several things aren't disclosed.

First, there's no definition of "uses AI daily." Opening an internal chat assistant once counts. Handing an agent a full quote comparison also counts. The depth gap between those is enormous and both fold into the same 95%. In most organizations this metric is assembled from login events or tool-access logs, and counted that way, the number climbs easily.

Second, "thousands of agentic workflows" is equally undefined. Whether that means thousands of complex multi-step automations or thousands of saved prompt templates changes the meaning completely. The release gives neither a specific count nor an example.

Third, and most importantly: there are no outcome numbers. Nothing on cycle-time reduction, win rates, headcount changes, or cost. Every figure disclosed is an input metric. Not one output metric appears.

That's a common pattern. Most press releases announcing enterprise AI adoption report usage and not results — not because results don't exist, but because early measurement is hard and nobody publishes a bad number.

Fourth, the treatment of risk is thin. Insurance brokering is regulated, and a misread policy clause or a missed condition becomes real liability. The release uses the phrase "responsible governance" without specifying what gets verified or at which step a human checks agent output. If thousands of workflows really are running, that verification layer is arguably the most important part of the story.

So here's the precise weight of this announcement. It confirms "we deployed tools and people open them." It is not evidence that "the company got better as a result." The former isn't trivial. But don't confuse it with the latter.

Dismissing the case entirely would be lazy, though. Even if 95% represents shallow usage, getting tool access and baseline training that broadly across 3,000 people is not nothing. Most organizations stall exactly there — licenses purchased, half of them never opened. Depth comes later, and depth doesn't happen without breadth.

Who gains what

IMA gains position in hiring and sales. Brokerage talent moves frequently, and a reputation for being technically ahead genuinely helps recruit good brokers. For corporate clients, "our broker compares a hundred quotes in a day" is a sellable line.

Associates gain relief from repetitive work. Per the release, less time gathering information and more helping clients navigate complex decisions. Given that brokerage value is created in the second activity, the direction is right.

Other non-tech companies evaluating AI arguably gain the most. What this case demonstrates isn't a technology choice — it's organizational design: a front-line-driven propagation structure like AI Studio, role-specific education, and a platform-agnostic strategy. That combination is a usable template.

Platform-agnostic deserves a note. It means not binding workflows to a single model provider — a reasonable defense when model performance and pricing invert every few months. But it has a cost: you maintain the abstraction layer yourself, and you give up optimizations tuned to each model's strengths.

The history of the phrase "enterprise-wide adoption"

Companies have been declaring enterprise-wide adoption of new technology for a long time, with split results.

The early 2010s had an "enterprise social" boom. Internal social networks got deployed and every employee signed up. Signup rates were high; actual usage collapsed within months. The problem wasn't the tool — it was that the behavior the tool meant to replace was still easier.

Cloud migration is the counter-case. Early resistance was heavy, but organizations that crossed over never went back. The difference was clear: cloud created a structure where not doing it cost you.

Which category AI lands in varies by industry. And brokerage is likely closer to cloud. In a business whose skeleton is document comparison and summarization, once a competitor does it in a day, not doing it stops being an option.

One more thing. Over the past year the industry has repeated a finding that most enterprise AI pilots fail at the scale-up stage, with absent front-line participation named most often as the cause — IT builds it, the business doesn't use it. IMA's AI Studio looks designed squarely at that failure mode. Whether it worked requires outcome numbers we don't have. But the diagnosis is correct.

How competitors respond

The large brokerages — Marsh, Aon, Willis, Gallagher — are far bigger than IMA and run substantial technology organizations. If they make the same announcement, the numbers will be larger. But scale makes enterprise-wide adoption harder, not easier. Getting to 95% across 3,000 people and across 50,000 are different problems.

So the frame IMA is playing for isn't scale, it's speed — mid-size meant it could move. Companies in this size band often are the most advantaged in enterprise transformation: small enough to persuade, large enough to resource.

There's a mid-size weakness too. Fine-tuning proprietary models and building large data infrastructure favor the giants. IMA's platform-agnostic strategy looks connected to that constraint — if you can't build it, get good at choosing and switching.

Insurtech startups face different pressure. Many were built on the premise that incumbent brokers are slow. When an incumbent runs agents at 3,000-person scale, that premise wobbles. The side that already owns distribution and client relationships usually catches up on technology faster than the side with technology builds distribution.

For carriers, a different picture emerges. When brokers start comparing quotes at volume automatically, terms competition becomes more transparent and more brutal. Broker AI adoption applies pressure to carrier margins.

So what actually changes

If you own AI adoption at a non-tech company — copy the structure, not the tools. A front-line propagation org, role-specific training, platform independence. That trio is the actual content of this case.

If you're in insurance or finance — this reconfirms that document comparison, policy analysis, and quote reconciliation are the top automation targets. If people are still comparing in spreadsheets, competitors are likely doing it differently. And that gap surfaces directly as quote turnaround time, which is hard to hide from clients.

If you want to use this as a benchmark — be careful. The 95% is unverified self-reporting with no published definition of "use." Setting your own target against it makes opening a tool into a KPI.

If you handle risk or compliance — how verification is designed in an organization running thousands of agentic workflows is the thing to watch. If that layer breaks first in a regulated industry, it can reverse the adoption curve wholesale.

If you watch the AI industry — note that the center of gravity is shifting. An insurance broker just occupied a slot that used to be filled only by tech companies. We may be entering a stretch where document-heavy industries post higher real adoption than software companies do.

🥄 Three Things You're Probably Wondering

— Is 95% real? It's unverified company self-reporting, with no published definition of "daily use." Counted from tool-access records, a number like that is achievable. There's no particular reason to disbelieve it, but don't read it as a measure of depth.

— Did headcount go down? The release says nothing about it. The tone actually runs the other way — associates spend less time gathering information and more advising clients. No mention of workforce size or hiring plans. Omitting staffing is standard in these announcements, so no conclusion is available.

— Could my company do this? Depends on your work mix. Brokerage is built on reading and comparing long documents, which suits AI unusually well. An industry heavy on physical work or in-person interaction won't produce the same number. Better to start from which tasks are genuinely substitutable than to target an adoption rate.

Sources

Numbers and criteria are as of announcement and may change.