LISTEN
Listen to this perspective
Written by Oleg Cohen. Companion narration uses the Brian voice.
Your AI program can improve margins while making the company less able to change.
It can remove the people who understood why a process worked, and it can bury commercial judgment inside automation. It can also make the next operating change dependent on a vendor whose priorities differ from yours. The business gets faster at executing a set of assumptions that nobody is responsible for revisiting.
In my previous essay, EBITDA Up, Value Down, I argued that AI projects can succeed while the investment thesis weakens. That leaves a practical question. Nobody can forecast where this technology or these markets are going, including the people building it. So what should an owner build when the conditions behind the plan will keep changing?
My answer is that the ability to change belongs in the plan itself.
I call this institutional maneuverability. It is a company’s demonstrated ability to recognize a consequential change, decide what it means, and alter how it creates value. It has to do this at an acceptable time, cost and risk, while keeping its commitments and what it has learned. The idea applies to openings as much as to threats.
The idea has a lineage. Teece, Pisano and Shuen developed the dynamic capabilities framework in the 1990s, and Teece later described it as sensing, seizing and reconfiguring. I sometimes call it the capability to be capable. Three things are new. Technology cycles are now shorter than hold periods. Modern intelligence makes reconfiguring a company cheaper than it used to be. And the same technology, deployed for efficiency alone, can lock a company in place.
An efficiency program that closed a door
Consider a hypothetical industrial services company with nine experienced estimators. They know which jobs become delivery problems, which customer requirements deserve another question, and which attractive contracts consume everyone’s attention.
The company introduces AI to speed up estimates and standardize proposals. Turnaround improves, six estimators leave, and the program saves $1.5 million a year, which meets the business case.
Two years later the market moves. Customers start asking for response-time guarantees, and a competitor offers inexpensive remote diagnostics. Management sees a more valuable service built around guaranteed response and prevention, and believes it could add $4 million of EBITDA.
The quoting system can generate the new proposal, but the company still has to work out whether it can deliver the service. Someone has to work out which failures can be diagnosed remotely and where technicians can arrive in time. They need to know which parts must be stocked and what the guarantee should exclude. They also need to know how the promise changes working capital and liability.
The six people who left would have answered those questions. Their judgment is gone, the commercial assumptions sit inside a vendor’s workflows, and the remaining managers are fully committed to this year’s plan. Rebuilding takes twelve months. Customers are signing multi-year agreements within six.
At eight times EBITDA, the savings were worth $12 million of enterprise value. If the new offer had worked, it would have been worth $32 million. Nobody priced the second number when approving the first. This is an illustration, not a case, but the shape will be familiar.
The company automated the estimate. It didn’t preserve the ability to reconsider what should be estimated, promised or sold.
The same program, designed differently
Now run it again. Every override of the quoting system is logged with a reason and reviewed against job outcomes each quarter. Estimates are linked to delivery results. Pricing and scope assumptions are kept where management can see and change them, outside the vendor’s configuration. Three senior estimators stay on to review exceptions and train others. The program saves $1 million a year rather than $1.5 million.
When the market shifts, management can test the new offer against real operating constraints. It picks a defined customer group, allocates delivery capacity, and limits the promises. It agrees in advance what would justify expansion and what would stop the trial. Existing obligations stay intact, and the test produces evidence about whether the opportunity is real.
So there are two questions to ask before approving an AI investment. What does this make us capable of doing next? And what does it make us unable to do?
What to commit to
Flexibility everywhere is expensive. A company that keeps reopening every decision never builds anything distinctive.
I would commit deeply where accumulated experience and complementary investment create value. That means expertise, customer relationships, delivery reliability and the ability to learn from the work. I would protect reversibility where a choice rests on an uncertain assumption. Examples are a particular model, a supplier, a pricing mechanism, a workflow, or an allocation of responsibility.
These categories take judgment. A production process can be the advantage, and a specialized vendor can be worth the dependency. The rule is to ask what the commitment earns and what it would cost to change. Spend on flexibility where change would otherwise be slow, destructive or prohibitively expensive. Accept commitment where the benefit justifies the exposure, and name the evidence that would make you reconsider.
For the services company, trusted technical judgment deserves sustained investment, and a particular AI model doesn’t. Response reliability stays central while the pricing and delivery around it evolve.
Authority and candor
Maneuverability fails in ordinary ways. Sales lifts conversion by promising what service can’t support. Procurement saves money by dropping the supplier whose emergency response protected retention. Each initiative meets its target, and the company gets harder to steer.
Seeing a problem also achieves little if nobody can act on it. People need to know which decisions they own, which constraints apply and when to escalate. They also need to be able to say so when the automated recommendation misses something, or when a successful initiative’s assumptions no longer hold. If bad news threatens someone’s standing, the organization spends its time protecting the plan.
The private equity problem
The standard playbook works against all of this. Leverage, lean teams, fixed plans and a fixed hold period remove slack, and maneuverability needs slack. A team with every hour committed to this year’s number has nothing left to build next year’s business.
Real options are the useful lens here. An initial commitment can preserve a later choice to expand, change course or stop, and options are worth more when uncertainty is high. Aswath Damodaran’s caution applies, though. A possible opportunity isn’t automatically a valuable option. The company needs a credible way to capture the benefit, at a sensible cost and inside a real window.
In practice, reserve capacity for a named possibility or exposure. That might be a defined share of management attention, a specific skill, an alternative supplier or a bounded investment. Know what it preserves, what expansion would require and when the choice expires, and review whether it’s still worth the cost. A board can underwrite that. “We need to be more agile” gives a board much less to work with.
The same applies across a portfolio. Shared technology lowers cost, but it can also concentrate dependencies and leave several management teams queuing for the same people. Ask what stays with each company when ownership changes.
What a buyer should look for
Evidence of maneuverability lives in decisions the business has actually made.
- How long did it take to get from a material signal to a decision, and from the decision to a working change?
- What did the change cost, and what happened to customer commitments, quality and cash while it was made?
- Can a critical provider or component be replaced without losing the knowledge and control the business depends on?
- Which initiatives were stopped or revised because the evidence changed, and what carried into the next attempt?
These questions don’t produce a universal score, because a factory and an advisory firm face different constraints. They do show whether the buyer is getting a company that has practiced changing. The alternative is a company whose flexibility depends on a few people performing rescues.
Where I’d start
I’d start with one consequential possibility. What could this company offer, enter or do materially better if a constraint it has long accepted were removed? Then test the conditions. What has become possible, and what do customers actually value? What gives this company a reason to win? What would have to change together, and what evidence would justify the next commitment? Sometimes the answer is that the opportunity doesn’t deserve investment, and that result is useful too.
Where I sit: I advise through Kainora, and I’m building technology shaped by these views. The Private Equity Executive Exchange is a private, non-commercial setting for investors and operators to examine these questions before any decision about an engagement or a platform.
The question I’d put to an investment committee: what are we committing to because it creates advantage, and what are we keeping changeable because that advantage will have to evolve?
And one for you. When did one of your companies last change something material, and how long did it take from signal to working change?