The seat has a name now
Three years ago the AI work in a portfolio sat wherever it landed. The CTO of the largest platform company. A data lead borrowed from the fund. A partner who had read enough to be dangerous. None of them owned it, so none of them were measured on it.
That has changed. Korn Ferry now writes about "the AI operating partner" as a defined role. Heidrick reports a sharp rise in hiring for it. Large funds have built the capability in-house under their own labels: Vista, Hg, Insight and KKR Capstone all run internal operating teams with a data and AI function inside them. The fractional form of the seat is recognized too, because most funds do not have fifteen portfolio companies to keep a full-time hire busy.
Here is the working definition. An AI operating partner owns the AI line of the value-creation plan across a portfolio and is accountable for moving it into the P&L inside the hold period. Not for advising on it. For moving it.
That single word, owns, is what separates the seat from everything adjacent to it. A strategy firm produces a roadmap and leaves. A development shop produces software and leaves. A training vendor produces certificates and leaves. The operating partner is still in the room at the next quarterly review, holding the number.
What the role actually owns
Seven things. If a candidate or a firm cannot map their week to these, they are selling something else.
1. The AI line of the value-creation plan
Every portfolio company gets one AI line in the VCP, with a dollar figure, an owner and a date. Not a workstream list. One line the CFO can defend. The operating partner writes it, defends it at the investment committee and revises it when the evidence changes.
2. Per-function 100-day plans
AI does not get deployed to a company. It gets deployed to a function. Finance close. Field service scheduling. Underwriting. Claims. Inside sales. Each one gets a 100-day plan with a named owner, a baseline number and a system that reaches production. One function at a time, in sequence, is faster than five at once.
3. Adoption design
The hardest part of the job. Who is expected to use the system, how their day changes, what stops being their responsibility, what shows up in their goals and how their manager sees the usage number. This is org design work. It is where most programs die.
4. Governance and the do-not-automate list
Data handling, model access, retention, vendor terms, audit trail. Just as important, the written list of workflows the company will not automate this year: anything touching protected classes in employment decisions, anything the regulator reads, anything where a wrong answer is expensive and unrecoverable. A do-not-automate list makes the rest of the program faster because it ends the argument.
5. The build backlog
A ranked list of systems to ship, with the estimate, the owner and the maintainer named before work starts. The operating partner decides build versus buy each time and is accountable for what happens when the vendor renewal comes up.
6. Board reporting
One slide, monthly. Baseline, current, target, adoption rate, what shipped, what is next, what is blocked. In the CFO's units. The board should never see a slide of tool logos.
7. The cross-portfolio playbook
What worked at portfolio company one gets packaged so portfolio company two can run it in half the time. This is where the fund-level return sits. A single portfolio company deployment is a project. A repeatable one is an asset.
Why the seat exists
Four numbers explain the market.
First, capacity. Two-thirds of operating partners cover five or more portfolio companies. An operating partner running five companies through commercial diligence, pricing work, ERP replacements and a CFO search does not have a spare quarter to run an AI rollout, however much they want to.
Second, the constraint is people. FTI's 2026 Private Equity AI Radar finds that talent and capability is the primary constraint on portfolio AI at 35%. Not budget. Not model access. Capability.
Third, the portfolio companies are stuck at the starting line. Accordion found that 98% of sponsors have told portfolio CFOs to prioritize AI, fewer than one in three has meaningfully implemented it and 68% do not know where to begin. That last figure is the whole market in one number. The mandate arrived. The method did not.
Fourth, the results are thin. Only 36% of PE-backed portfolio companies use AI in day-to-day operations and 7% call it fully integrated across the portfolio. At the exit end, only 9% of operating partners have seen a demonstrable AI premium in a completed transaction.
Put those together. Boards are asking. Portfolio companies do not know how to start. The people who would normally fix that are already covering five companies. The seat exists to absorb that work and to be measured on it.
The timing argument
The average holding period is at a record 6.6 years. That is long enough for AI work to compound into the exit story and short enough that a company which starts in year five has nothing to show a buyer. The seat is a hold-period decision, not a technology decision.
Full-time versus fractional: the arithmetic
A full-time AI operating partner costs roughly $400K to $700K all-in once you count base, carry participation, bonus and the fund overhead that follows the seat. Against a portfolio of fifteen companies that is $27K to $47K per company per year, which is defensible. Against a portfolio of six it is $67K to $117K per company, for a person who will spend part of the year waiting for the next platform deal to close.
The economics of a full-time hire only work above roughly fifteen portfolio companies. That is the honest threshold, and it is why the fractional form of the seat exists at all.
A fractional AI operating partner is the same mandate bought by the day. PortLev prices it at $19,650 per month on a six-month minimum, about two days a week, concentrated on the one to three portfolio companies where the AI line is real this year. That is $236K a year for a seat that shows up in the operating reviews, ships the systems and reports the number.
The trade is straightforward. Full-time buys availability. Fractional buys senior time on the companies that matter, without paying for the quarters where there is nothing to own.
The decision table
Four options are usually on the table at once. They are not interchangeable and the failure mode of each is different.
| Option | Choose it when | Typical cost | Fails when |
|---|---|---|---|
| Full-time AI operating partner | 15+ portfolio companies, a fund-level data platform in flight, a mandate with two or more years of runway | $400K to $700K all-in per year | The portfolio is smaller than the seat. The person spends half the year in search of a problem. |
| Fractional AI operating partner | 4 to 15 portfolio companies, one to three with a live AI line, an exit inside 24 months, no internal owner | $19,650 per month, 6-month minimum | Nobody inside the portfolio company is assigned as counterpart. A fractional seat needs a named adoption owner to work with. |
| Strategy firm | The fund needs a defensible portfolio-wide thesis, a market map or a diligence opinion on an AI-native target | $150K to $1M+ per engagement | You expected implementation. You received a roadmap and a handover meeting. |
| Development shop | The workflow is already specified, the owner exists and you need hands to build a known thing | $50K to $500K per build | Nobody designed adoption. The software works and the function keeps using the spreadsheet. |
Costs are market ranges for scoping conversations. PortLev pricing is on the operating partner page.
The common mistake is buying the third and fourth options in sequence and calling it a program. A strategy firm writes the plan. Eighteen months later a development shop builds one piece of it. No one owned the gap between them, which is where adoption lives.
Ten questions to ask a candidate
Whether the candidate is a full-time hire, a fractional seat or a firm, these ten separate operators from presenters.
- Show me a system you shipped that is still in production and tell me who maintains it.
- Walk me through one adoption plan you wrote. Who was the owner and what changed in their goals?
- What was the baseline number before you started and how did you measure it?
- Which workflow did you refuse to automate, and why?
- Describe a rollout that failed. What was the first signal and how late did you catch it?
- How do you decide build versus buy, and when did you last choose buy?
- What does your monthly board slide look like? Show me a redacted one.
- How do you handle a function head who does not want the system?
- What did the second portfolio company deployment cost compared with the first?
- Who on our side needs to exist for you to succeed, and what happens if that person is not appointed?
Question one does most of the work. A demo is not a system. A pilot is not a system. A system has users, a maintainer and a next release. If the answer is a screenshot, keep interviewing.
The CHRO angle: adoption is workforce design
Most funds scope this seat as a technology hire. That is the expensive error. The binding constraint in a portfolio company is almost never model quality. It is that forty people in a function have been asked to change how they work, and nothing about their week, their goals or their manager's dashboard has changed to make that rational.
Adoption design is four decisions.
- Owner. One named person inside the function, with the mandate written down and time protected. Not a steering committee. Not the CIO. Not a champion with a side project.
- Incentives. The usage number appears in the owner's goals and in the function head's operating review. What is not measured in the review does not survive a busy quarter.
- Roles. The job descriptions change. If the system removes four hours a week of work, say what those four hours now go to. Ambiguity here produces quiet non-adoption.
- Compliance. Employment decisions, protected data, retention and audit trail get written rules before the rollout, not after the first incident.
This is why a practising CHRO with legal training is a different profile than a data science lead in the same seat. The systems are the easy half. The org change is the half that determines whether the P&L moves.
How to measure the seat
Four numbers, reported monthly, in the same format every time.
- Basis points of EBITDA. The dollars attributed to shipped AI work, expressed against EBITDA, with the attribution method written down and agreed with the CFO before the first system ships. Argue about the method once, at the start.
- Adoption rate by function. Weekly active users over eligible users, per function. Licenses purchased is not an adoption number and should never appear on the slide.
- Systems in production with a named maintainer. Count them. A system with no maintainer is a liability with a countdown on it.
- Playbook reuse. How many portfolio companies redeployed a workstream, and how many days the second deployment took against the first. This is the number that turns a project into portfolio value.
Two quarters is the fair test. If none of the four has moved in two quarters, the problem is the seat, the scope or the counterpart inside the portfolio company. Find out which and fix it. Do not renew and hope.
What the seat is not
It is not a chief AI officer badge on an existing CTO. It is not a vendor selection committee. It is not an innovation function reporting to no one. And it is not a training program. Certificates are not adoption, and a workshop that does not change what appears in the operating review changes nothing.
The seat is one person with a number, a deadline and the authority to say no to work that will not reach the P&L. If you are writing the job description, write that sentence first and build the rest around it.