The Board-Reporting Format for
AI Across a Portfolio

Most AI board updates report activity: pilots launched, tools bought, workshops run. None of that is value. A board and an LP want one thing, whether AI is moving the number, and the report has to answer it on one page per portco and one roll-up for the fund. If your AI reporting cannot separate margin realized from motion, it is theater, and theater is exactly how MIT's 95 percent of pilots quietly died.

Why AI board reporting is broken

The default AI slide is a list of initiatives. It grows every quarter, it never ties to the P&L, and no line item ever gets killed. That is a reporting failure before it is an execution failure: when the format rewards activity, the portfolio produces activity. Fix the format and you change what gets built.

The four things every AI report must show

• Realized P&L impact to date: the margin or cost line each shipped workflow has actually moved, measured against a pre-AI baseline. Not projected. Realized.
• Pilots in flight, with a day-90 kill metric: each active pilot, its target P&L delta, its minimum adoption rate and the date it gets scaled or killed. A pilot with no kill metric does not belong on the slide.
• Adoption: the percentage of the target users actually using each shipped tool. Adoption is where value is won or lost, so it is a first-class metric, not a footnote.
• Spend against value: what has been spent on AI versus the margin realized and underwritten, so the board can see the return, not just the invoice.

The one-page portco format

Each portfolio company gets a single page: realized impact at the top, pilots in flight in the middle with their kill metrics, adoption and spend-vs-value at the bottom. A board member should be able to read it in ninety seconds and know whether AI is working, what is at risk and what decision is needed. Anything longer is hiding the answer.

The portfolio roll-up for LPs

At the fund level, the same four metrics roll up across every portco into one view the GP can put in front of LPs: total realized AI margin, pilots in flight and their status, portfolio-wide adoption and total AI spend against value. This is what turns AI from a talking point in the annual letter into a reported value-creation lever. It only exists if every portco reports in the same format, which is why the format is installed once, centrally, not left to each company.

The anti-theater rule

One rule keeps the report honest: nothing counts as value until it is realized and attributable. Projected savings, pilots without adoption and tools nobody uses are reported as in-flight or at-risk, never as impact. That single discipline is what separates a portfolio that is actually capturing AI value from one that is busy, and it is the reporting equivalent of getting AI past the pilot and into the P&L.

Frequently Asked Questions

1. What should an AI board report include?

Four things per portfolio company on one page: realized P&L impact to date measured against a pre-AI baseline, pilots in flight each with a target metric and a day-90 kill decision, adoption rates for every shipped tool, and spend against realized value. Projected savings and unused tools are reported as in-flight or at-risk, never as impact.

2. How do you report AI ROI to a board?

Report realized margin, not activity. For each shipped AI workflow, show the specific P&L line it moved against a pre-AI baseline, the adoption rate behind that result, and the spend it took to get there. Keep projected and unattributable numbers out of the impact column so the board sees a return it can trust, which is the opposite of the pilot-heavy reporting behind the 95 percent that fail.

3. How do you roll AI reporting up across a portfolio for LPs?

Install one reporting format across every portfolio company so their metrics are comparable, then aggregate: total realized AI margin, pilots in flight and their status, portfolio-wide adoption and total AI spend against value. That single roll-up is what lets a GP present AI to LPs as a measured value-creation lever rather than a talking point.

4. How often should AI be on the board agenda?

Every board meeting, but as a one-page value report, not a demo. The cadence that matters is the day-90 review on each pilot, where it is scaled or killed against its pre-set metric. Standing AI theater on the agenda without a kill discipline is how initiatives accumulate and margin does not.