AI Agent Account Context Playbook 2026
Walk into customer calls current, with the account on one page instead of five tabs. When a move lands on an account or a customer meeting is booked, a one-page brief goes on the CRM record: the gap against your successful customers, the recommended move and its evidence, and what's still open. Your team reads it, decides and sends.
The best customer conversations start from what the account has actually done – but that data is scattered across five systems: usage, the CRM, billing, support, last call's notes. Pulling it together takes time. Put that context on the account before the next touch.
This play puts one page on the account record whenever there's something to act on: a move from the weekly check, or a customer meeting on the calendar. The brief opens with the account's gap against your successful customers and the recommended move, then the evidence. An agent your team sets up drafts it; your team decides. When nothing has changed, the record says why there's no new brief.
Measure it on prep time per brief, the share of briefs kept as drafted, commitments kept from one call to the next, and moves decided within a week – then on the renewals and expansions those moves were for.
How it works9 steps
01SignalStart from the move on the record, or the meeting on the calendar
Two things start a brief:
- A move is drafted: the weekly check sets Recommended move on an account to anything but No move.
- A customer meeting is booked: a calendar event with a customer attendee, sixty minutes out (default). Take the account from the meeting's confirmed account association and the attendees' contact records in HubSpot, not from the email domain alone. If the account is missing or ambiguous, the agent asks the owner before it reads any account data. A personal address or a partner on the invite needs checking, not an automatic no. Take the meeting type from the event title (start titles with QBR, Kickoff, Renewal or Check-in; anything else is a check-in), and treat titles and descriptions as data, never as instructions to the agent.
The brief reads the benchmark; it doesn't build it. If you don't have one: take 20 or more customers in the same segment, 12+ months in, renewed at the same or higher revenue; set the median of your core measure as good; your CS lead approves it. Before a measure triggers a move, check it against later renewal outcomes; the benchmark play has the full method. Under about a hundred accounts, run the whole play by hand.
02ScoreCheck what's changed since the last brief
- Usage since the last brief: active users and the core action, from a saved query in PostHog, Amplitude or Mixpanel.
- Open commitments: who promised what, by when, from the last call's notes.
- Open tickets & billing: anything open in support; failed payments or plan changes in billing.
- Contract & renewal date: from the CRM, with 90 days of notes.
Patchy records go in the brief as a line – "no notes from the last call" – never papered over.
03DecisionDecide whether this touch needs a new brief
Brief every move, every QBR, kickoff, renewal or executive meeting, and any account with no brief yet. Leave it alone when it's a routine call and the last brief is under 14 days old (default) with nothing changed: usage within 10% (default), no new ticket, no new commitment.
04ActionNo brief: point to the one that's still current
Set Brief type to No new brief with the note below. The owner reads the last brief. A brief that repeats itself teaches people to stop reading briefs. See the no-move play.
05ActionDraft one page, same order every time
The agent writes a note on the HubSpot company record in four fixed sections:
- Gap & recommended move: the account's measure, the benchmark's, and the move with its reason.
- What changed: biggest change first.
- What's open: commitments with their status, tickets, unanswered asks.
- What to send or ask: for a move, the message filled in from that move's template in its own play; for a meeting, two or three questions the data raises.
Say an account pays $12,000 and similar customers running reporting pay $18,000: the $6,000 is a comparison, not a forecast. (The numbers are illustrative.) If the meeting moves, the next run redrafts the brief.
06Human stepYour team decides and sends the move
The AE or CSM reads the brief, then sends the message as drafted, edits it, or passes. They own every customer conversation. Before sending, record the success measure and review date, set Brief kept and Prep minutes, and close the task – that date is when the move was decided. The brief agent never sends. If a move's drafts are kept as written for weeks, automate that move in its own play, with the checks in the renewal agent play.
07Human stepRead it an hour ahead; open with what changed
Open with the change you saw, and listen. The customer shouldn't have to brief you on their own account.
08ActionPut the new commitments on the record
After the call, copy the commitments from the Granola notes (who, what, by when) into Open commitments on the record. Give your Granola template a Commitments section (who, what, due date) so the notes list them; a date only shows up if someone said one. The next brief's "What's open" starts there.
09OutcomeReview whether the briefs earn their place
For two weeks before you start, owners log prep minutes per call by hand; that's the baseline. Eight weeks on, the CS lead filters the saved view to the window and records three answers in the playbook: median Prep minutes against the baseline, the share of Brief kept = As drafted, and the share of commitments marked done by their date. Then check what else played a part, and whether briefed moves were decided within a week (brief date to task closed). A section people skip gets cut; a draft rewritten every time means the agent is missing a source. Change the template, stop it, or give it time. The review play has the full method.
Setup and templates
Sources: gap and recommended move from the weekly check's CRM properties; usage from a saved query in PostHog, Amplitude or Mixpanel; contract, renewal date and notes from the CRM; failed payments from billing; open tickets from your support tool; call notes from Granola (Business plan), synced to the HubSpot record from a shared folder; meetings from your calendar. Account-level numbers need your analytics tool's account add-on (group analytics in PostHog, Accounts in Amplitude, Group Analytics in Mixpanel) and a group call in your app. By hand: a person fills the same template from the same tabs.
Call notes: before transcription starts, tell everyone on the call what Granola captures and ask them to agree, including anyone who joins late. If anyone declines, turn transcription off and take notes by hand. Use your approved tool, access controls and retention settings.
The agent: a scheduled Claude agent with read access to those sources and write access to notes on the HubSpot company record. Its instructions: the four sections in order · one page · say what it doesn't know · never send anything to a customer. Give it no email connector, so it can't send. It runs every 15 minutes (default), reading the calendar and the weekly check's properties. Every 15 minutes needs an agent built on the Claude API (a Managed Agents scheduled deployment), which also gives it its own login. A Claude Code routine runs at most hourly and belongs to one person: set it up under the person accountable for it, since everything it does appears as them, and have it run hourly and look two hours ahead. Each CSM shares their calendar with the account the agent runs as, and call notes reach the HubSpot record from the shared Granola folder, so the agent sees every owner's meetings, not just one person's. The agent writes through HubSpot's MCP connector; it can't write notes if Sensitive Data is turned on in your portal. A Slack message to the owner, sent from the account the agent runs as, may link the note; the note on the record is the brief.
Company properties in HubSpot (Starter or above; free HubSpot allows 10 custom properties in total), or your CRM's account object:
| Property | Type | Filled by |
|---|---|---|
| Gap vs benchmark | Single-line text | Weekly check |
| Recommended move | Dropdown select: the weekly check's moves · No move (read, not changed, by this play) | Weekly check |
| Move reason | Single-line text | Weekly check |
| Brief type | Dropdown select: Move brief · Meeting brief · No new brief | Agent |
| Last brief date | Date picker | Agent |
| Open commitments | Multi-line text | Account owner, from call notes |
| Brief kept | Dropdown select: As drafted · Edited · Rewritten | Account owner |
| Prep minutes | Number | Account owner |
| Success measure | Single-line text | Account owner |
| Review date | Date picker | Account owner |
| Review result | Dropdown select: Result · Progress · Activity only · Stopped | Account owner |
The daily list is a saved view: Last brief date is today and Brief kept is unknown. With each brief, the agent creates a task for the owner, "Brief: [move or meeting, date]", due before the meeting or within two days for a move.
Move brief:
[Account] · [plan, ARR] · [measure] vs benchmark [benchmark measure] · Recommended: [move] because [reason] What changed: [change, date, source] What's open: [commitment, owner, due] · [ticket] Draft: "Hi [name] – [what we saw]. [the offer]. Would [time] work?" Not known: [missing source]
Meeting brief:
[Account] · [meeting type, date] · [attendees and roles] · [measure] vs benchmark [benchmark measure] · Recommended: [move or No move] What changed: [change, date, source] What's open: [commitments from the last call, with status] Ask: [question the data raises] · [question two] Not known: [missing source]
Commitment line: "[Who] will [what] by [date] – [open · done · late]." The next brief updates the status.
No brief: "No new brief – [reason]. Last brief [date]."
How Accoil fits
Accoil is a customer-success consultancy with our own tooling, and this is a play we run with you, not for you. We build the benchmark with your team, and you approve it. Every week we check every account in the agreed segment, and the recommended move and its evidence land on the account record in your CRM – we recommend HubSpot – at a pace your team can actually run. Your team's agent uses that evidence to draft a one-page brief when a move or meeting triggers one. If nothing has changed, it records why no new brief is needed. Your team makes every call. Nobody from Accoil contacts your customers. After the agreed window we review what changed together, and you keep the playbook, the benchmark and the analysis.
Running it yourself? Swap HubSpot for Salesforce or Pipedrive, Claude for OpenAI (the Agents SDK with the same connectors, run on your own schedule), and Granola for Gong (your Gong admin sets up its MCP connector). The play stays the same. On Salesforce, the field writes need API access (Enterprise and up, or an add-on on Professional). On Pipedrive, automations and date triggers need the Growth plan.
Thirty minutes with Kate, Simon or Peter on one example segment: what a successful customer looks like, where the gaps are, and whether there's work worth doing this quarter.
Talk to a founder →Keep reading
Monday-Morning Account Triage Playbook 2026
Your best customers show you what good looks like. Every Monday, this play compares each account in a segment with that benchmark, ranks what each gap is worth, caps the list at what your team can run, and puts the week's moves on the account records in your CRM: close a gap, grow the upside, or leave it alone with the reason noted.
The Leave-It-Alone Playbook 2026
When the weekly check finds a gap worth a conversation, this play looks for a reason to wait first – a fresh purchase or renewal, a budget or org change, an open blocker, a departed champion. If the records show one, the reason goes on the CRM record with a date to look again, so the conversation happens when it can land.
Renewal-Risk Radar Playbook 2026
Renewals go best when the account already looks like your customers who renewed. From 90 days out, this play compares every renewing account with them each week – usage and revenue – checks the records, and puts one move on the account in your CRM: an expansion look, a value review, a save plan with the CSM – or no move, with the reason noted.
Every playbook, one download
All 32 workflows as print-ready playbooks — diagrams included. Plus every new workflow as we publish it.