LISTEN
Listen to this perspective
Written by Oleg Cohen. Companion narration uses the Brian voice.
At a recent ACG leadership event, one CEO described how often his company was interviewing AI consultants. Another replied, “Let me know when you find a good one.”
These were people building substantial businesses. They were already experimenting, and some had found that employees moved faster than company policy. They weren’t short on activity. They were short on a way to decide what deserved attention.
I think that’s because most AI conversation answers the wrong question. “Implement AI,” “drive adoption across the portfolio,” and “find someone to lead AI” all address how to do today’s work better. An owner has a different question:
What will make this business worth owning as intelligence becomes widely available?
A company can get better at today’s work while the economics of that work change underneath it. Two examples show why that matters.
1. Earnings can rise while value falls
A company earns $20 million of EBITDA and is valued at ten times. That’s $200 million, split between $100 million of net debt and $100 million of equity. Over the hold, AI-enabled efficiency lifts EBITDA 10 percent, to $22 million.
But part of what the company sells now looks easier to reproduce. The next buyer doubts how durable those earnings are and pays eight times, so enterprise value is $176 million. Hold net debt constant and ignore interim distributions, and equity is now worth $76 million. Earnings rose 10 percent while equity value fell 24 percent. At eight times, EBITDA would have to reach $25 million just to get back to even.
The efficiency initiative succeeded, but the value-creation plan did not.
This is an illustration, not a forecast. But Bain’s midyear report found software valuations in buyout portfolios down about 8 percent over a single quarter. Marks aren’t transaction prices. They do show owners reassessing durability in the sector where substitution is easiest to see. For any other business, the useful question is which mechanism applies and which doesn’t. Substitution, customers doing the work themselves, and price pressure from lower-cost competitors are the obvious candidates.
2. Productivity doesn’t tell you who keeps the gain
Most AI business cases count hours saved. Fewer ask who captures the benefit.
Take a services firm that bills by the hour. AI lets its people finish an analysis in 30 percent less time. On that engagement, at the same rate, revenue falls 30 percent. Whether the firm ends up better off depends on things the productivity number doesn’t show. Can it sell the freed hours, and at what price? Can it move pricing toward outcomes before a competitor does? Firms with real specialization, relationships and reliability can hold their price. Firms selling hours of generic analysis probably can’t.
Customers have options too. In McKinsey’s 2026 survey, 32 percent of respondents said they decided against buying software because they could now build it in-house. That finding concerns software, not services, but the mechanism is the same. A customer decides that something it used to buy is now something it can do.
So the forecast has to say who captures the gain: the company, its customers, or its competitors. Saved hours also have to become cash through a named route, such as added revenue, less overtime, avoided hires or fewer errors. Each hour can be counted once.
This has a portfolio consequence. Holdings diversified by industry can share one exposure, such as the same pricing model under pressure, the same work moving in-house, or the same model provider. Industry diversification won’t reveal that.
What the evidence says
The popular story is that AI is failing, but the data is more complicated. In McKinsey’s survey, 80 percent of respondents say AI improved their productivity, yet only 37 percent of organizations report any positive EBIT contribution, essentially flat from last year. About 6 percent attribute at least 5 percent of EBIT to AI and also describe its impact as significant.
FTI’s survey of fund and operating leaders looks more encouraging. Ninety-five percent of funds say their AI initiatives met or exceeded the original business case, though only 17 percent significantly exceeded it. The same FTI report identifies siloed ownership and unclear accountability among the barriers to scaling AI.
These are different surveys measuring different things, so I’ll draw only a narrow conclusion. Meeting a project’s business case doesn’t settle the strategic question. AI projects can succeed while the investment thesis weakens.
The other side of the ledger
If this were only a warning about multiples, it would be half an argument. The same shift that erodes some advantages makes others more valuable.
Established companies hold experience that competitors can’t download. That includes the circumstances behind a customer exception, the reason an estimate went wrong, and the supplier who performs under pressure. Most of it travels through particular people. Consider a specialist services company whose best estimators can tell which attractive-looking contracts will become delivery problems. Helping someone draft a proposal faster is useful. Helping the whole company recognize those patterns, ask better questions, price the work properly and learn from the outcome changes the business.
It should also prompt leaders to question limits they’ve come to accept. The company might serve a segment it could never afford to serve before. It might grow without coordination costs growing at the same rate. Expertise that supported one location might support several, while local managers keep their authority. I’m interested in whether a company can become reliably better at something customers value, and strengthen that ability through use. That can reduce dependence on a few individuals and strengthen the investment case. It won’t automatically earn a premium.
One financial distinction follows from this. Some AI spending creates new earning capacity, and some preserves the position you already have. Both can be justified, but they should be underwritten differently.
Why the hire doesn’t settle it
A Head of AI can be a good appointment. But improving the firm’s internal research, changing a portfolio company’s operating model and assessing a target’s exposure are three different jobs. Each needs different authority. A strong systems builder may have no say over pricing, staffing or customer commitments.
Before hiring, write the mandate in business terms. Define which outcomes the person owns, which decisions they can make, what resources are committed, and what evidence the board will review. The role can help develop those answers. It can’t take them off the CEO’s or the sponsor’s desk.
What the next buyer will ask
- Do the data rights and licenses stay with the business?
- Does the improvement survive a change in leadership?
- Can the buyer run it without the seller’s portfolio team or one adviser?
- What does it cost each year to sustain?
- When an automated decision is wrong, who notices, and how fast can they recover?
The buyer is underwriting its own hold period. Your exit date doesn’t end the competitive change.
Where I’d start
I wouldn’t start with dozens of experiments, because a lean mid-market team can’t supervise them and still run the company. I’d start with one consequential question: what does this company need to become dramatically better at, and what could change outside it that would make the answer more urgent or less valuable? Then name a business owner, set a baseline, and agree on what counts as progress. Agree on what would invalidate the approach and a date to review it. Sometimes the first fix is clearer responsibility or a different commercial model, with no AI involved.
I’m not an investor. I’ve spent thirty years building technology and running the organizations that depend on it. I’m describing what I see from that side. The underwriting is your job, and I’d like to know whether the concern survives it.
Where I sit: I advise through Kainora, and I’m building technology shaped by these views. I also think this conversation is worth having before anyone chooses a platform or an engagement. That’s the purpose of the Private Equity Executive Exchange, a private, non-commercial setting for investors and operators to compare notes.
If you run a fund or sit in an operating role, I’d value your pushback on two things. Which of your holdings would pass the $22-million-at-eight-times test? And where is this argument wrong for your kind of business?