Inside the Monthly AI Executive Report: what an AI executive owes you in writing

When I tell an executive team the advisory retainer is $15,000 a month, the fair next question is: what shows up for that? Not “what do you do” — what shows up. Consulting has trained buyers to expect activity instead of artifacts, and a lot of AI advisory is exactly that: meetings, enthusiasm, and a Slack channel.

So I put the answer in writing. Every client, every month, gets the same document: the Monthly AI Executive Report. I’ve published a complete eight-page sample — written for a composite client, a fictional 118-person accounting firm assembled from patterns across real engagements — so you can read the whole thing instead of trusting my description. This post walks through why each section exists, because the structure is the strategy.

Adoption first, always

The report opens with usage numbers: weekly active users from the admin consoles, by department, against a target. Not a survey, not anecdotes — console data.

It opens there because unused tools have zero ROI, and unused tools are the industry’s default outcome. MIT’s research found roughly 5% of enterprise AI pilots produce measurable P&L acceleration; the rest stall. The difference between the two groups is rarely the model. It’s whether anyone was watching the usage curve and treating a flat line as a problem to diagnose.

The department breakdown matters more than the headline number. In the sample report, tax is at 71% weekly active and audit is at 39% — and the report says plainly that some of audit’s gap is correct behavior, because most of their high-stakes work was out of approved scope. A low number isn’t automatically a failure. But it’s always a fact the leadership team should see, with a diagnosis attached.

Shipped work, with the baseline attached

The second section reports what actually went to production and what it measurably changed: in the sample, 212 engagement letters through a new drafting workflow, 3.5 hours down to 45 minutes each, roughly 560 hours returned to client work in seven weeks.

Every number in that sentence has a baseline behind it, captured before the build started. This is the discipline most AI programs skip, and it’s why more than 80% of companies tell McKinsey they see no enterprise-level EBIT impact from gen AI: without a baseline, even a working system produces no evidence it works. If the delta can’t be shown, the report says so — “we believe this is working but can’t prove it yet” is a legitimate status. Pretending is not.

The backlog, including the dead ideas

Section three is the scored opportunity backlog: every candidate workflow ranked on impact, feasibility, data sensitivity, and time-to-value — plus the killed list, kept visible with the reasons attached.

The killed list is the part clients underestimate. Ideas that die in a meeting resurrect by hallway conversation three months later, usually pitched to whichever partner wasn’t in the original meeting. Writing down “we killed the website chatbot in March because the risk exceeded the value, and here’s what would change that” ends the resurrection cycle. Given that S&P Global found 42% of companies abandoning most of their AI initiatives, the choice isn’t whether ideas die — it’s whether they die deliberately, in writing, or expensively, in production.

Blocked decisions: the section that earns the fee

Section four lists the decisions only the CEO or partners can make, each with the context, my written recommendation, an owner, and a date. In the sample: expand a Copilot pilot or consolidate on one platform; approve AI-disclosure language for client deliverables; scope a build before its go-live window closes.

Most stalled AI programs are not stalled on technology. They’re stalled on a decision nobody framed sharply enough to make. A monthly, dated, published queue of “what we’re waiting on you for” keeps decision latency measured in days. It’s also accountability that runs in both directions — when a decision slips, the next report shows what the slip cost.

Risk, vendors, and the next 30 days

The back of the report holds the governance review (near-misses, exceptions, the risk register with owners), vendor and model notes, and dated commitments for the next 30 days.

The vendor section exists because the ground moves. In the sample, a model update quietly shifted the tone of client letters mid-quarter; partner review caught it, and the fix — a house style guide plus a regression check before every model rollout — became process. Your AI vendor will change the model under you. An operating model has to notice.

The next-30-days section is the report’s enforcement mechanism, because July’s report opens by scoring June’s list: done, slipped, or dropped, with reasons. That loop — commit in writing, then grade yourself in writing — is most of what separates an operating model from a vendor relationship.

Why writing is the product

A monthly report can’t be vague the way a status meeting can. Verbal updates absorb ambiguity; documents expose it. If adoption stalled, the chart shows it. If nothing shipped, the section is empty. If my recommendation was wrong, it’s on the record, dated and signed.

That’s the actual answer to “what does $15K a month buy.” Not access, not enthusiasm — a written operating record a board member could pick up cold and understand. Read the sample report and the rest of the sample deliverables, and hold whatever advisor you hire — me included — to that standard.