LISTEN
Listen to this perspective
Written by Oleg Cohen. Companion narration uses the Brian voice.
Every established company has a Susan.
Ask a hard enough question, about a customer who needs special handling, a supplier you can trust when something breaks, or why a design was rejected ten years ago, and eventually someone says: talk to Susan. Or: Mike will know.
That isn’t organizational failure. It’s evidence of success. The company has built expertise worth having. The trouble is where that expertise lives, and what most AI programs are about to do to it.
The question most AI programs start with
Most AI conversations still start with productivity. How many hours can we save, which tasks can we automate, how many people will we need. Those are fair questions, and for some work the honest answer is fewer people. Cost reduction is real, and sometimes it’s the right call.
For an established company, though, it’s rarely the most important question. A $50 million manufacturer, a regional insurance brokerage or a specialty distributor holds something more valuable than a set of tasks to automate. It holds what its people know about how the business actually works. Which customers need special handling. How the best estimator knows a quote is dangerous before the numbers show it. Which production combination works even though the specification says it shouldn’t. When a client’s situation has changed before the client says so.
So the better question isn’t how AI can do more of our employees’ work. It’s how the company can keep, spread and keep improving what its best people know.
The knowledge paradox
Middle-market companies sit in an odd position. They’re large enough to have accumulated serious knowledge, and usually too small to have institutionalized it. It lives in long-tenured employees, founders and second-generation leaders, senior engineers, estimators, plant managers and service veterans.
That works while the company is small enough for proximity to do the job. Growth breaks it. More locations, customers, products and acquisitions mean more decisions than the experts can touch. The organization grows faster than judgment can travel. It ends up with more data than ever and less ability to bring the right knowledge to the right decision.
That is the paradox. The more valuable the company becomes, the more it knows, and the harder that knowledge is to use as a company asset rather than a collection of personal ones.
Companies have fought this for decades with procedures, intranets, shared drives, knowledge bases and training videos. All of them help, and none of them solves it, because what makes Susan valuable isn’t a document. It’s judgment: recognizing that this situation is different, and remembering why a decision was made and what it traded off. Systems of record are good at transactions. They have never been good at the reasoning around them.
What changed
In The Rationed Company I argued that AI moves the constraint companies have been designing around, which is the scarcity of experienced attention. Knowledge follows the same pattern. Making expertise broadly usable used to mean interviews, codified rules and expensive knowledge-management programs, and they rarely paid back. Modern AI can work across documents, records and conversations, keep context, find similar past situations and put relevant experience in front of the person deciding.
That doesn’t mean a model magically knows your business. It means it’s now practical to build institutional memory around what the business already knows.
Consider a specialty manufacturer. Every completed job contains lessons: what was quoted, which machine ran it, what settings worked, where waste occurred, how long setup really took, what an experienced operator changed, and what the margin finally was. Today most of that ends when the job ends. The work is invoiced and the company moves on. Now imagine every completed job making the company better at quoting, scheduling and running the next similar one. That isn’t automation. It’s a company learning from its own operation. I made a version of this point in Your AI Is Getting Smarter. Your Company Isn’t. The tools improve for everyone, but the company only learns if its outcomes flow back into something.
Amplify, don’t replace
If the advantage is decades of accumulated expertise, replacing the people who hold it can destroy the asset you’re trying to grow.
Take a senior estimator with twenty years of judgment. The goal isn’t to eliminate the estimator. It’s to make twenty years of estimating experience available to every estimator in the company. The expert becomes more valuable, because their judgment is no longer limited to the jobs they personally touch. The company becomes less fragile, because the knowledge doesn’t leave when they retire. Newer people get productive faster, because they can draw on experience that used to take years to encounter.
There’s a harder point here. The way a company talks about AI decides whether any of this is possible. If the message employees hear is that AI is about headcount, the people who hold the knowledge have every reason to keep it to themselves. Nobody helps capture what they know for a system they believe is being built to replace them. A labor-reduction program can quietly guarantee that the most valuable knowledge never gets captured. The companies that treat their experts as the source of institutional intelligence, and not as a cost line, are the ones that will get it.
Every acquisition buys knowledge
This matters most in companies that grow by acquisition. A deal brings customers, revenue, contracts, systems, assets and people. It also brings something that never shows up properly on the balance sheet: operating intelligence. The acquired company knows its customers, markets, suppliers and exceptions in ways that took years to develop.
Integration usually focuses on systems, reporting lines and financial consolidation. The acquired knowledge stays in the acquired people. Some of it spreads. Much of it eventually leaves with them.
Now imagine every acquisition making the whole platform more capable, with what the acquired team knows becoming findable and usable across the company. Scale would mean more than more revenue. Each deal would add to a growing base of intelligence. For a sponsor running a buy-and-build strategy, that is a different integration thesis.
Succession raises the same issue. A family company handing off to its second or third generation, or a founder preparing to sell, faces the same risk: the business is worth what it knows, and much of what it knows is about to walk out the door. Employee-owned companies may grasp this most easily, because preserving what generations of owners learned is plainly stewardship of a shared asset. The economics don’t depend on who owns the company.
Where the advantage moves
The underlying technology is becoming broadly available. Your competitors can access similar models, buy the same copilots and hire the same consultants. Generic AI adoption won’t stay a differentiator for long.
The real question is what your company can combine with that intelligence that competitors can’t buy: your customer history, your operating experience, your engineering decisions, your failures and the reasons behind your exceptions.
Anyone can buy access to increasingly capable intelligence. No one can buy what your business has spent thirty years learning. Unless you let it disappear.
A different set of questions
For owners, CEOs and operating partners, this leads to a different agenda than a list of use cases:
- What knowledge would hurt us most to lose?
- Where does important judgment depend on a few people?
- What does the company keep relearning?
- Where could experience from one part of the business improve another, including companies we’ve acquired?
- Which capability could get stronger every time we use it?
Those questions point toward a company built around what it knows.
The companies that do this well may find their greatest AI advantage was never the model. It was the business they had already built.
Where I sit: I advise through Kainora and am building technology aimed at this problem. It’s in production with one company and in pilots with several more.
What would your company lose if Susan retired tomorrow? And who would know?