The Control Book for AI-Touched Reporting
Board prep takes a week and the numbers still don't agree
Someone pulls from five systems. Sales pipeline doesn't match finance's view. Two days go into reconciling before anyone writes a single slide.
Then the meeting opens with directors asking how a number was calculated instead of what it means. Boards already spend about a third of their time on financial performance. When that time goes to reconciliation, the strategy conversation never happens.
And when a quarter misses, the explanation gets vague. "Macro headwinds." "Pipeline softness." A board that has sat through enough decks knows the difference between an honest explanation and a managed one.
Now add AI to the stack. Your forecast weighting, pipeline scoring, health scores, and attribution model are probably AI-assisted already. Almost no board deck says which parts.
What AI actually does here
The reporting discipline is human. The assembly, flagging, and screening are AI. That split is the whole design.
Assembles the package. Pulling ARR movement, pipeline coverage, conversion rates, and retention from multiple systems into one consistent format is retrieval work across defined fields. AI is reliable at this, and it's the part that currently eats a week.
Flags every variance above threshold. Instead of someone eyeballing what moved, the AI compares actuals against plan and prior period across every component and surfaces what crosses the line for each section.
Drafts the variance narrative as a starting point. For each flagged variance, AI assembles the supporting evidence: which segments moved, which deals drove it, what the loss reasons say. A human writes the actual explanation and stands behind it.
Checks explanations for false confidence. Before a variance narrative reaches the deck, AI screens it against known markers of unverified causal claims. Single-cause certainty. Mechanism stated without evidence. Narrative coherence standing in for data.
Audits your stack for the disclosure log. AI inventories which tools in your revenue stack use AI, where they touch reporting data, and when they were last calibrated.
The Control Book
Seven sections, same format every cycle, one named owner each.
1. Financial Summary. P&L and cash owned by the CFO. ARR movement with full waterfall owned by the CRO.
2. Pipeline and Forecast. Coverage, forecast summary, stage conversion rates, deal velocity, and lost rates by stage. Reported after hygiene filters, not raw.
3. Marketing Performance. Marketing-sourced and marketing-influenced pipeline, lead volume and quality, contribution to revenue, channel performance. Attribution methodology documented and consistent.
4. Retention and Expansion. GRR and NRR, categorized churn detail, customer health with trend, expansion pipeline, product white space.
5. Unit Economics. CAC, CLTV, the ratio, and payback period. With disclosure of which costs are in CAC.
6. Variance Narrative. Four required parts for every variance above threshold: what happened, why it happened, what's changing, what it means next period.
7. AI Disclosure Log. Which AI tools touched this cycle's data, where, last calibrated, override rate, known limitations.
Section 7 is the one nobody else is running
What changes
Grant Thornton surveyed 950 business leaders in early 2026. Barely one in ten boards met their threshold for strong AI oversight.
Most boards approve AI investments without any mechanism to know whether what they approved is working.
If AI shaped a number in your deck, disclosing it puts you ahead of nearly every other org in the room. Not as a compliance gesture. A board that knows which numbers are AI-assisted asks better questions about them. A board that finds out later asks worse ones.
Get the full framework
The complete control book is in the Consult RevOps GTM Skills repository. All seven sections with component-level detail and ownership, the variance narrative structure, the AI disclosure log format, and the six-step build process. Built to run in Claude Code or any adjacent AI tool.
Board prep goes from a week to a day. Assembly and reconciliation are automated against documented definitions. The human time goes to interpretation.
Reconciliation leaves the meeting. Directors know where each number lives and how it's calculated, so the conversation starts at what it means.
Vague explanations become structurally impossible. The four-part variance structure means you either have the mechanism or you say you don't.
AI involvement is visible before someone finds it. Proactive disclosure is a different conversation than a board member discovering an AI-generated number mid-follow-up.