AI Board Reporting: How to Report AI Results to Investors and the Board

Report AI to a board or investors like any other investment. Show what was spent, what changed in a measured outcome against a baseline, what the risks are and what you will decide next. Keep it to one page a quarter. Leave out activity counts and self-reported time savings, which prove nothing.

What should an AI report to the board include?

An AI report to the board should include total spend, the outcomes AI was meant to change and how they moved against a baseline, adoption across the team, the risks and how they are controlled, and the decisions needed next. Each item fits in a few lines. The board should finish reading it knowing whether to continue.

Keep the format fixed from quarter to quarter. A board that sees the same five sections every time can spot a trend. When the layout changes each quarter, nobody can tell progress from presentation.

Which AI metrics belong in a board report?

Use measures the board already reads: pipeline, cost per lead, cycle times, margin and headcount growth. Then show where AI work changed them. A number the board has never seen before, such as prompts sent or agents built, cannot be judged. A familiar number moving against a known baseline can.

The baseline is what makes the number credible. Record how the work ran before AI touched it, for at least a few weeks, so the change has something to stand against. The guide to measuring AI ROI covers setting one.

Many leaders find this hard. According to Microsoft and LinkedIn's 2024 Work Trend Index, 59% of leaders worry about quantifying AI's productivity gains, and 60% say their company lacks a vision and plan for AI. Boards notice when the plan and the measure are missing.

Few firms can yet show a profit effect, so be honest about the stage. According to McKinsey's 2025 State of AI survey, 39% of organisations report any EBIT impact from AI at the enterprise level. Most early reports show leading measures, such as time to first draft, before the financial ones move.

Why is time saved a weak number for a board?

Time saved is usually estimated by the people saving it, and those estimates run high. It also only becomes value when the hours go somewhere useful, such as more output from the same team or a hire not made. Report what the saved time produced, not the hours themselves.

The best evidence on self-reporting is sobering. In a randomised trial by METR, 16 experienced developers took 19% longer on real tasks when AI tools were allowed. Before starting, they expected to be 24% faster, and afterwards they still believed AI had made them 20% faster. The authors describe it as a snapshot of early-2025 tools in one setting, but the gap between feeling and fact is the point.

So convert time into something the board recognises. Halcyra, a fictional 60-person software company, stopped reporting "hours saved on proposals" and started reporting proposals sent per salesperson per month. The guide to why time saved misleads goes further.

How should you report AI risk and governance?

Report three things: what AI can reach (which systems, and whether it can read or also change them), who approves anything sent, published or spent, and what went wrong this quarter. Add whether the AI policy is current and who owns it. A board cannot weigh the upside without seeing the controls.

Directors are being told to take this seriously. According to the Australian Institute of Company Directors, "the board plays a key role in governing the balance between opportunity and risk" on AI. Its 2026 guide, written with the Human Technology Institute, also covers oversight of AI's human impact on employees and customers.

The gaps are common. IBM's 2025 Cost of a Data Breach report found that 63% of breached organisations had no AI governance policy or were still developing one. Australia's National AI Centre lists deciding who is accountable as the first of its six essential practices, and maintaining human control as the last.

A simple risk table works well:

1. Access. Each system AI is connected to, read only or with write access.
2. Approval. The named person who approves each kind of output.
3. Incidents. Anything that went out wrong, and what changed as a result.

How do investors read AI results differently from a board?

Investors care less about this quarter and more about whether the gains last. They ask whether the system belongs to the company or to one person's accounts, whether results repeat across teams, and whether growth now needs fewer hires. A board asks whether it is working; an investor asks whether it compounds.

That changes what to show. Evidence that knowledge and workflows live in files and accounts the company owns answers the durability question. Results repeated in a second team answer the scaling question. And a falling cost per outcome as volume grows answers the efficiency question.

Be careful with the word "AI" itself. Investors have seen many products relabelled as AI, so specific numbers beat broad claims. "Our AI-powered sales process" says nothing. "Cost per qualified meeting fell from one quarter to the next, with the same team" says something they can check. The guide to an AI strategy for growing businesses shows how to set those targets in the first place.

What does a one-page quarterly AI report look like?

A one-page quarterly AI report has a headline, then five short blocks: spend against budget, two or three outcomes against baseline, adoption by team, risk and incidents, and decisions requested. Each block is a few lines or a small table. Detail sits in an appendix for anyone who wants it.

1. Headline. One sentence: what AI changed this quarter, in a business measure.
2. Spend. This quarter and year to date, against budget.
3. Outcomes. Each with its baseline, current value and the change.
4. Adoption. Weekly users of the shared setup, by team.
5. Risk. Access, approvals and incidents.
6. Decisions. What you will expand or stop, and what you need from the board.

Write the headline last, once the numbers are in. If no outcome moved, say so plainly and explain what will change next quarter. Boards forgive slow results far more readily than vague ones.

Skein measures the outcomes boards and investors recognise, such as pipeline, cost per lead, time to first draft, AI visibility, search demand won and adoption; the investors page shows how a Company OS works across a portfolio. That suits a venture-backed company or a fund that wants AI results reported the same way each quarter. A business with no AI work running yet has nothing to report, and should start with a plan.

Frequently asked questions

How often should AI be reported to the board?

Quarterly suits most growing businesses, in the same fixed format each time so trends are visible. Monthly updates can go to the leadership team instead. Report sooner if something significant happens, such as an incident involving customer data or a decision that needs board approval before the next meeting.

What AI metrics do boards care about?

Boards care about measures they already track: pipeline, cost per lead, cycle times, margin and growth per head. Show how AI work moved those against a baseline. They also care about risk, including what AI can access, who approves its output and any incidents. Activity counts mean little to them.

Should time saved by AI be reported to investors?

Only alongside what the time produced. Self-reported time savings tend to run high, and hours saved are not value until they become more output, faster delivery or a hire not needed. Report the business result, and keep time saved as supporting detail rather than the headline.

Who should own AI reporting to the board?

One named executive, usually whoever owns the AI plan, with each outcome owned by the leader responsible for it. The executive assembles the report, but each number comes from its owner. That keeps the figures honest, because the person reporting a result is the person accountable for it.

What should a board ask about AI?

Ask what was spent, which business measures changed against a baseline, and what the evidence is. Then check which systems AI can reach and whether it can change them, who approves its output, what went wrong, and what management proposes to expand or stop next quarter. Finally, ask who owns the plan.

Related reading