Someone above you says "we need to use AI, we're falling behind." No number attached. No deadline. No definition of done.

You'd do what any sensible operator does. Start where the risk is low and the payback is quick. Invoice processing. Customer service triage. Something you can point at by the next board meeting.

That's roughly what happened across private equity over the past year. Then AlixPartners asked both sides of the table whether the money was well spent, and got opposite answers. 72% of the executives running portfolio companies are satisfied with their AI investments. Only 33% of the people who own those companies say the same.

Same companies. Same projects. Opposing verdicts.

Hashi's take: This isn't an AI problem. It's much simpler. The sponsors asked for efficiency, the operators delivered it. They just measured success differently. And it isn't unique to private equity — this is everywhere. Let's fix that.

If you've never asked your sponsor which number they'd pick, forward this their way.

AlixPartners run an annual study of private equity leadership, and this year they put the same questions to both sides of the table — 427 executives: 174 from the funds, 253 running the companies those funds own, surveyed in late 2025.

Both sides agree on quite a few things. At least 80% of executives on both sides rate the relationship positive or very positive on trust and collaboration. Nobody here is fighting.

Where they part company is what the AI money bought. Underneath the 72/33 split, only 2% of operators are dissatisfied with their own results — against 13% of sponsors dissatisfied or very dissatisfied with those same results. And 10% of sponsors say their portfolio companies aren't investing in AI at all.

The odd part is that both sides did exactly what they said they would. About three quarters of PE executives say they encourage their portfolio companies to use AI for operational efficiency, and fewer than half push for top-line growth — which suits the operator anyway, since bolt-ons and the capital for new technology are decided at the fund. So the companies obliged. Year over year, this is where the AI money moved:

Four lines up, one down — and the one that fell is the growth line. AlixPartners side with the owners here, and their test is a good one: if AI were delivering what operators claim, lenders would happily fund deals on the strength of it. They don't. Banks back results that are already in the numbers — or as AlixPartners put it, they "bankroll performance, not promises."

This is where I'd start, and it isn't only an AI thing. Same conversation I've had about ERP, about CRM, about every digital transformation programme of the last fifteen years.

A value creation framework isn't complicated. It just gets overlooked — AI more than anything else I've seen, because everything about this moment tells you to move fast, and writing down what you expect feels like it slows you down.

One rule I don't bend: measure it, or don't claim it. That goes for a vendor's pitch, for my own writing, and for anything I've put my name on.

Seven lines per initiative.

Run one through it. Agents handling tier-one support tickets. Saves around $400k a year in cost to serve. Costs $120k to build and $60k a year to run. Owned by the COO. Main measure: cost per ticket, sitting at $6.10 today. Supporting: first-response time and escalation rate. Reviewed monthly at the ops meeting, and we'd expect the number to move by the end of Q3.

A little effort upfront to avoid big surprises later.

If you can't put a figure on lines two and three, put a range. If you can't put a range on it, you don't understand the project well enough to fund it yet — and that's worth knowing before the money goes out, not after.

And if you can't fill in all seven, it isn't a project. It's an experiment. That's fine — but call it one, and give it an experiment's budget.

No survey measures this part, but I think it sits underneath all of it.

A lot of the people setting AI expectations haven't done the work of understanding AI. They've read the headlines. They've sat through a demo. What they haven't done is find out what these tools are good at, where they fall over, and how long one takes to get running inside a business with real data and real staff. So the expectation gets set from a headline, and the delivery gets handed down.

You can see it in what the funds built. 84% appointed a chief AI officer — the org chart moved. But only 9% run a dedicated AI centre of excellence, and 63% offer their portfolio companies informal guidance only. A title got created. Not much capability did.

They are using AI, to be fair — on their own work. Firm operations up to 48% from 38%, due diligence to 42%, sourcing targets to 34%. Useful, and a different problem entirely. Getting a deal team faster at reading memos teaches you very little about changing how a distribution business runs a warehouse.

Korn Ferry named what that produces: portfolio company leadership managing yet another stakeholder from the fund, and "uncontrolled pressure to adopt AI without business justification." Not a plan. Pressure.

The fix depends on which side of this you're on. If you're the one setting the expectation, you need enough fluency to set a real one. Not a course — use the tools on your own work for a month, and go and watch one working deployment end to end: what it cost, what broke, how long it actually took. Then try filling in lines two and three yourself, even as a range. If you can't, you aren't setting an expectation. You're passing down anxiety.

And if you're on the receiving end, you can't make anyone upstairs do that. What you can do is send the seven lines back up. A vague directive becomes a specific conversation the moment somebody has to put a number next to it.

Last month I wrote that what private equity is short of isn't tools or budget — it's people who can hold a value creation plan in one hand and an operating floor in the other. This is what that job looks like on a Tuesday. Not a strategy. Seven lines, filled in, sent upward.

Whoever sits above you has a picture of where this business is going — a sponsor, a board, a parent company, the family that still owns the shares. You may never have seen it. Only 44% of portfolio company leaders say they always help develop the deal thesis.

AlixPartners' own advice is the way in: start at the endgame and work backwards. In private equity that's the exit. Everywhere else it's whatever the owner is actually driving at. Four questions come out of that.

1. What does this business have to be worth, and to whom? Name the buyer and the number — or if nothing's being sold, what the owner wants this business to be able to do, and by when. Everything else is downstream. A good answer is specific enough to be wrong.

2. What is AI meant to do for that number? Its role in the value creation plan is the thing to settle with whoever sits above you. The seven-line sheets are how — bring them, and ask which ones they'd fund again. "All of them" isn't an answer.

3. What are we deliberately not doing? Write that list down next to the list of what you are doing. Pausing new products for a year is fine, as long as everyone knows it's paused and not dead.

4. What number tells us this worked, and who watches it? One measure, one owner, reported the same way to both sides. Two people looking at different numbers will argue about reality.

Then agree on who owns this if the priorities shift.

This isn't a private equity problem. Private equity just measures it better, because it has to. Everywhere else the same gap sits there unmeasured: you deliver what was asked for, somebody above you was grading something else, and it surfaces a year later as a strategy review, a new hire above you, or a budget that doesn't get renewed. The lesson travels even if the data doesn't. Write the seven lines. Make sure whoever set the expectation has done enough of the work to set a real one. Start at the endgame and work back. None of this is complicated. You don't need a computer science degree or an AI research team to do any of it. What it buys you is everyone grading the same thing — and a far better payoff on the AI money you're already spending.

Keep reading!

- Hashi