What an AI operating partner actually does between entry and exit

The job is not AI. The job is the earnings bridge. Every fund I talk to now underwrites some version of an AI assumption at entry: margin from automation, growth from better commercial coverage, multiple expansion from looking like a modern company at exit. The AI operating partner is the person accountable for making that assumption come true inside real portfolio companies, on the hold-period clock, in numbers the CFO will sign.

That definition sounds obvious. It is not how the seat usually gets staffed. Funds hire a technologist who can build but has never owned a P&L, or a consultant who can size the prize but has never shipped anything, and then wonder why the thesis is still a slide eighteen months in. The seat only works when diagnosis, build, and ownership live in one person. Here is what that person actually does across a hold.

Before the deal: the AI line in the model has to survive diligence

The first place this seat earns its cost is underwriting. An AI assumption in a model is a claim about a specific company's workflows, data, and people, and most of those claims do not survive contact with a data room. I have underwritten 1,000+ acquisition targets with my own capital at risk, and the pattern is consistent: the question is never "can AI help this business" but "which two or three workflows carry enough volume and enough waste that a system would move the margin line, and does the data exist to run it." If the answer is vague, the assumption comes out of the model. An operating partner who cannot do this work personally is negotiating the purchase price with someone else's guess.

Day 0 to 90: find the number, ship one workflow, stand up the scoreboard

The first quarter in a portfolio company is a diligence sprint, not a listening tour. Two or three opportunities sized in margin or revenue terms an investment committee would underwrite. One workflow chosen for confidence, not for spectacle, and shipped to production with an owner, a baseline, and evaluation gates. And a scoreboard the CFO signs, because a number nobody owns is a number that will quietly become a narrative. I have run this spine on nineteen Fortune-100-scale engagements and now on my own companies. It has never needed more than ninety days to produce either a working system or an honest kill decision, and both outcomes are wins. The expensive outcome is the pilot that neither ships nor dies.

The middle of the hold: capability, not heroics

The middle year is where most AI programs rot, because the person who shipped the first win leaves and the muscle leaves with them. The operating partner's real product in this period is an operating model the company runs without them: intake and scoring so the pipeline of opportunities is ranked by value instead of noise, decision rights so a workflow gets killed or scaled by rule instead of by meeting, and governance with evaluation gates so the systems stay honest as models and vendors change. The proof this works is that it outlives you. An operating model I authored for a top-5 U.S. health insurer is still how their automation office runs, years after the last invoice.

Exit prep: make the thesis show up in the bridge

At exit, the AI story is worth exactly what a buyer's diligence team can verify. Realized savings with baselines. Systems in production with owners and uptime, not demos. Governance a buyer can inherit. The operating partner's last job is to make the earnings bridge and the data room tell the same story, so the multiple reflects the work instead of discounting it as vendor risk. Sized and realized are different words, and the discipline of never confusing them is worth real money at exit.

What to ask anyone who wants this seat

Three questions separate operators from narrators. Show me a number you realized, not sized, and walk me to its baseline. Show me a system you built yourself that is still running without you. Show me the P&L you owned when a decision cost you your own money. Candidates who can answer one of the three are common. The seat needs all three, because the job is one unbroken chain: find the full potential, deliver it, and leave it running.

That chain is the whole role. Everything else is commentary.