Meeting-Prep Agent Playbook 2026

The trigger is any calendar event with an external attendee from a customer domain, sixty minutes out. It arrives with three fields: the account (matched by domain), the attendees and their roles, and the meeting type — a QBR, a kickoff and a "quick sync" need different briefs. T-minus 60 is deliberate: early enough to read, late enough that nothing goes stale. Briefs generated the night before are outdated by a morning's worth of tickets.

The agent assembles what the CSM would have tab-hopped for, with Accoil as the product-data source of truth: the engagement score and its 90-day trend, seat changes, the features that went quiet, renewal distance. Around that spine it pulls open support tickets and — the piece most prep skips — the commitments made on the last call, straight from the Granola notes. One rule the agent follows religiously: it flags what it doesn't know. "No notes from the last call" is a line in the brief, not a gap papered over with plausible filler. A brief you can't trust about its own blind spots trains the CSM to stop reading it.

Everything gathered gets ranked into the same three sections, every time:

• What changed — score trend, seat movements, quiet features, anything new in the ticket queue. Biggest change first.
• What's open — commitments from the last call with their status, open tickets, unanswered asks.
• What to ask — two or three questions the data raises, like "Exports usage doubled — new use case or workaround?"

The fixed format is what makes the play work. CSMs stop reading variable-format briefs within a week because skimming requires knowing where things live. One page is a hard limit — if the ranker can't cut to a page, it's ranking, not summarizing.

The brief posts to the CSM in Slack at T-minus 60 — a DM or the team's prep channel, not an email that files itself. Same headline structure every time ("Acme · QBR · score 71 ▼9 · renewal in 84d") so the 90-second skim starts at the notification. If the meeting moves, the brief regenerates; a brief for a meeting that got rescheduled yesterday is worse than none.

The CSM reads for ninety seconds and walks in knowing what changed, what's open, and what to ask. The meeting changes shape: instead of the customer re-briefing the CSM on their own account, the CSM opens with the observed change and listens. The "last commitments" section is the trust engine — dropping a promise the customer remembers is how relationships die quietly, and it's the single most common blind-side in post-sale work.

Granola captures the call and the new commitments come out of it — who promised what, by when. Those land back on the account so the next brief's "what's open" section starts current. This is the step that turns a prep tool into a memory: every call inherits every previous call, and commitments stop depending on whether anyone updated the CRM on a Friday afternoon.

The steady state: prep is a 90-second read, blind-side moments go to zero because the usage drop is the first line of the brief, and commitments are tracked call-over-call by default. Watch the brief read-rate — if it slips, the brief got long or the format drifted. Fix the brief, not the CSM.