Read the internal business case for almost any artificial intelligence programme at a private capital firm and you will find the same shape. Hours saved in diligence. Analyst time released. The associate team able to cover more ground without adding two headcount next year. Perhaps a line about reducing reliance on outsourced providers. The number at the bottom is a cost avoided.
That framing is not stupid. It is the framing that gets signed, because it threatens nobody, requires no one to change how they work, and can be approved by a management committee in twenty minutes. It is also, we think, the single best predictor that a programme will quietly stall about eight months in.
The reason is not technical. It is that an efficiency mandate belongs to nobody. When the stated benefit is hours saved across a team, no individual partner's number moves, no one's compensation is affected, and when the inevitable friction arrives, the programme has no defender with anything at stake.
Hours saved is a benefit nobody ever collects
Suppose a deployment genuinely saves an associate team six hours a week each. It is worth following those hours to see where they end up.
In practice they are absorbed. They go into more thorough work on the same deals, into a longer committee paper, into the request that used to get declined, or simply into leaving the office at seven rather than eight, which is a real and worthwhile thing but does not appear anywhere in the firm's results. Nobody is made redundant, because the firm did not intend to make anyone redundant and would damage itself if it did. Nobody is not hired, because hiring decisions are driven by deal volume and fundraising cycles rather than by a spreadsheet of released hours.
So the promised saving never shows up as a saving. A year later the finance partner asks what the programme delivered, and the honest answer is that some people's weeks are somewhat easier. That answer does not survive a budget round. It cannot be defended at renewal, because there is no number to point at, and the person defending it is usually an operations or technology lead who was never the beneficiary in the first place.
This is what we mean by buying to protect headcount. The purchase is genuinely intended to let the firm grow without adding people, or to inoculate the firm against the question of whether it is doing anything about AI. Neither of those is an outcome. They are postures, and a posture cannot be measured, so the programme has no way to prove it deserves to continue.
The pilot economics are worse than they look
To put rough numbers on it, and these are illustrative rather than measured, consider a typical efficiency-framed pilot at a mid-market firm. The direct external cost over nine months might be in the region of two hundred thousand dollars in advisory and implementation. The internal cost is generally larger and rarely counted: a senior operations lead at perhaps a quarter of their time, four or five professionals giving a day a month to workshops and testing, a compliance review, and a procurement cycle. Loaded properly, the internal figure on these illustrative assumptions is comfortably another two to three hundred thousand.
Against that, the stated benefit is around four thousand hours a year of released analyst time, which on paper looks like an easy win. But as above, those hours do not convert. The realised financial benefit, on these illustrative numbers, is close to zero, and the firm has spent roughly half a million dollars to arrive at a capability that works and that nobody is accountable for using.
Now take the same half million against a single outcome instead. Again illustratively, a firm running a fund with two hundred million dollars of commitments that shortens its fundraise by one quarter, or converts two additional re-ups that were genuinely at risk, or avoids one impairment because a lease concentration was caught in diligence, has produced a number that any partner can defend in any room. The point is not that these outcomes are easy to achieve. It is that they are the only kind of benefit that has a natural owner.
The back office is the honest exception
We should be careful here, because there is a category where efficiency genuinely is the right mandate, and treating it as second-class is a mistake we have watched firms make.
Fund accounting, capital call and distribution notice production, subscription document handling, AML and KYC extraction, and the mechanics of investor onboarding are processes with defined inputs, defined outputs and defined deadlines. The work has a floor and a cliff-edge calendar. Efficiency there is not a proxy for something else, it is the thing itself, and it converts into real outcomes that can be counted: fewer restatements, shorter close, fewer errors going out to investors, and less attrition among finance staff who otherwise spend every January in a state the firm should not be asking for.
The difference is that these processes have an owner by construction. The CFO or head of fund operations is accountable for the close calendar and the error rate whether or not any technology is involved. An efficiency mandate handed to that person is not ownerless. It lands on a number that already appears on somebody's objectives.
The failure we are describing is what happens when that model is copied into the investment or investor relations side, where there is no equivalent process owner and no pre-existing metric for the work to attach to.
What an outcome-led mandate actually says
The difference between the two kinds of programme is visible in a single sentence. An efficiency mandate says: reduce the time the team spends preparing investment committee materials. An outcome-led mandate says: the partner responsible for our industrials strategy will take fifteen opportunities to committee this year rather than nine, without lengthening the average time to a decision, and owns the programme that makes that possible.
Three things follow from the second version that do not follow from the first.
The mandate has a named individual attached who is already measured on the outcome, which means the programme has a defender in every budget discussion and a decision-maker when the work requires someone to change their habits. Very few of these programmes fail on capability. Most fail on the fortnight where somebody senior has to alter how they prepare for Tuesday mornings, and only an owner with something at stake pushes through that.
The measurement question is settled before the work begins, and it is a business number rather than a usage statistic. Queries handled and documents processed tell you the tool is switched on. They tell you nothing about whether the firm is better off, and a programme reporting only usage is already in trouble.
The scope gets smaller. An outcome-led mandate almost always produces a narrower initial build than an efficiency one, because the owner has no interest in capability that does not move their number. That narrowness is the reason these programmes tend to finish. Broad efficiency mandates sprawl precisely because nothing in the framing tells you what to leave out.
Naming an owner is a governance decision, not a technology one
None of this requires a different tool. It requires the firm to decide, before anything is procured, whose result is supposed to change and to accept that this person will be asked about it in twelve months.
That is harder than it sounds, because it converts a comfortable infrastructure conversation into an uncomfortable accountability one. A partner asked to put their name against an outcome will quite reasonably ask what happens if the technology does not deliver, and the firm has to answer honestly that the risk is shared. Efficiency mandates persist partly because they let everyone avoid that exchange.
But the exchange is the whole value of the thing. A firm that has a named partner, a named number and a defined date has already done the difficult part of the programme, and the implementation becomes a comparatively ordinary piece of work.
The firms getting value are running fewer, smaller, owned programmes
What we observe in firms that are genuinely getting value is not more ambition. It is less. One partner, one outcome they already care about, a build narrow enough to be finished in a quarter, and a real review at the end where the answer is allowed to be no.
Those programmes tend to produce a second one, because the owner has seen their own number move and wants more of it, and because their partners noticed. Efficiency programmes rarely produce a second, whatever the usage dashboard says.
The encouraging part is that this reframing costs nothing. It is a change in how the mandate is written, made before any money is committed, by people who already know which of their outcomes they most want to change.
