Daily Pipeline Review - Agent Prompt

Who this is for

This is for anyone who owns a number and a pipeline: founders doing their own sales, solo sellers, and the first RevOps hire who inherited a CRM nobody reads. The failure mode it attacks is universal - deals do not die from lost negotiations, they die from silence. Somebody went quiet, nobody noticed for three weeks, and by the time you follow up the budget moved on. A daily review is the cheapest fix in sales, and it is also the first ritual people drop, because reading forty records before coffee is miserable work. So delegate the reading. The agent scans; you decide.

Run it at the same time every morning, before email. Email sets other people's agenda; the pipeline review sets yours. Five minutes is the budget: the agent spends thirty seconds pulling data, you spend the rest on the three actions it hands you.

What a daily review actually checks

Strip the ceremony and a good review is five questions. This is the framework - it works in any CRM, a spreadsheet, or on a whiteboard:

Check	The question	Red flag
Staleness	When was each open deal last touched?	No activity in 7+ days
Shape	How do count and value spread across stages?	Everything piled in one stage; nothing entering at the top
Dates	Are close dates still real?	Any close date in the past - that deal is fiction until re-dated
Concentration	How much of the total sits in one deal?	One deal carrying 40%+ of open value
Next actions	What are the three highest-leverage moves today?	You cannot name three - the pipeline is running you

Two thresholds do most of the work. 7 days without a touch = stale. That is calibrated for B2B cycles measured in weeks; tighten to 3-4 days for transactional sales, loosen to 14 for enterprise. 21 days stale, or a close date in the past = zombie. Zombies get one revival attempt and then a close-lost - a pipeline padded with dead deals produces confident, wrong forecasts. And the output is always exactly three actions, not ten. Three named moves get done before lunch; ten become a list you feel bad about.

The prompt

Paste this into Claude or any MCP client connected to Relm. It is read-first by design: the agent writes nothing until step 6, and asks even then. Every tool and field it names is real - value_cents is an integer in cents, activity lists return newest-first, and stages live inside pipelines.

You are my daily pipeline reviewer. My CRM is Relm
(MCP: https://api.relmcrm.com/mcp - use the relm_* tools).
Read-only until step 6, and ask me before writing even then.

1. Call relm_list_pipelines. In each pipeline, treat stages whose
   key or label looks like won / lost / closed as terminal; every
   other stage is open.
2. Collect open deals: relm_list (object "deal", filter
   {"pipeline": <key>, "stage": <key>}) per open stage. Follow
   next_cursor until has_more is false.
3. For each open deal, get its latest touch: relm_list (object
   "activity", filter {"deal_id": <id>}, limit 1) - results are
   newest-first. Last touch = that activity's occurred_at; if the
   deal has no activities, fall back to its updated_at.
4. Flag STALE = no touch in 7+ days.
   Flag ZOMBIE = stale 21+ days, or close_date in the past.
5. Report, in this order:
   a. Snapshot - one table per pipeline: stage, deal count, total
      value (sum value_cents / 100, grouped by currency).
   b. Stale + zombie list, sorted by value descending: title,
      stage, value, days since last touch, last activity type.
   c. Exactly 3 next actions for today. Each names one deal and
      one concrete step (call X about the security review, send
      Y the revised proposal, close-lost Z). Not five. Three.
6. Offer to log the follow-ups. On my yes, create one activity
   per flagged deal via relm_log_activity (type "task", deal_id
   set, body = the recommended next step). Nothing else.

What the agent actually calls

No mystery in the plumbing - a run walks this exact sequence over MCP:

Step	Tool call	What comes back
1	relm_list_pipelines	Every pipeline with its ordered stages - the agent derives "open" from it
2	relm_list deal, per open stage	Pages of deals: title, value_cents, currency, stage, close_date, updated_at
3	relm_list activity, {deal_id}, limit 1	The most recent activity - its occurred_at is the last-touch date
6	relm_log_activity per flagged deal	A task activity holding the next step, only after your approval

Budget the requests honestly: a few fixed calls plus one per open deal. A 20-deal pipeline is ~25 requests a day, ~750 a month - inside the free 1,000-request tier. A 100-deal book lands around 3,200 a month, which is Pro territory at $29. Rehearse the whole loop on a relm_test_ key first; test mode is free, unlimited, and invisible to live data.

Apply it with your agent

Wire the MCP server once, then the prompt is the entire integration (full setup in how to give your AI agent a CRM):

{
  "mcpServers": {
    "relm": {
      "type": "http",
      "url": "https://api.relmcrm.com/mcp",
      "headers": { "Authorization": "Bearer relm_live_..." }
    }
  }
}

Prefer REST, or want the staleness check inside your own cron script? The core of the review is one call per deal:

curl "https://api.relmcrm.com/v1/activities?deal_id=deal_x8f3k2m9q2&limit=1" \
  -H "Authorization: Bearer relm_live_..."

Newest activity first; compare occurred_at to today and you have your stale flag. From there, make the routine compound: feed Friday's numbers into the weekly forecast prompt, keep the data trustworthy with the CRM hygiene checklist, and let the meeting-notes-to-CRM prompt log the touches so fewer deals go stale in the first place.

FAQ

Do I need Relm to use this prompt?

No. The review itself - five checks, a 7-day staleness threshold, exactly three next actions - works in any CRM, on a whiteboard, or in a spreadsheet. The prompt's tool calls are written for Relm's MCP server because that makes the whole routine one paste with zero glue code, but each step maps to any CRM that exposes deals and activities over an API.

Why is 7 days the staleness threshold?

Seven days is an opinionated default for B2B deals with multi-week cycles: long enough to allow a normal reply gap, short enough that a deal cannot quietly die between weekly reviews. Tighten it to 3-4 days for transactional sales and loosen it to 14 for enterprise cycles. Which number you pick matters far less than enforcing one every single day.

Can I run this review automatically every morning?

Yes. Any MCP client that supports scheduled or scripted runs - Claude Code on a cron, a scheduled cloud agent, or your own script calling the REST API - can execute it unattended. Keep step 6 (writing follow-up tasks) behind explicit approval at first; once you trust the output, let it log tasks unattended. Rehearse against a relm_test_ key before pointing it at live data.

How many API requests does a daily review cost?

A few fixed calls plus one per open deal: one pipelines call, one deal-list call per open stage (pages hold 25 deals by default, up to 100), then one latest-activity lookup per deal. A 20-deal pipeline is about 25 requests per day, around 750 per month - inside Relm's free 1,000-request tier. Bigger books fit comfortably in Pro at $29 for 100,000 requests.

Why does the prompt demand exactly three next actions?

Because attention, not information, is the scarce input at 8am. A ten-item list becomes a backlog you skim and ignore; three named moves - this deal, this step, today - actually get done. Anything the cut leaves out resurfaces tomorrow if it still matters, and the stale table keeps the full picture visible anyway.