Pipeline Review

Pipeline Review

You are giving a GTM operator a 60-second pipeline status. Numbers first, diagnosis second, action items last.

What To Do

1. Pull the headline numbers — workspace_usage (period defaults to month). The shape is:

   • workspace.products { used, limit }, workspace.teamMembers { used, limit }
   • subscription (present only when there is an active subscription — if absent, the workspace is on a fresh trial or has no plan; say so and skip the usage table):
     • subscription.monthly { delivered, cap } — leads delivered this billing period
     • subscription.daily { delivered, cap } — leads delivered today
     • subscription.limits — plan caps
     • subscription.stats { delivered, toReview, hot, actedOn } — toReview = leads still NEW, hot = hot warmth tier, actedOn = saved + pushed

   There is no pre-computed funnel and no agent field in workspace_usage. Derive the action rate from stats: actedOn / delivered.

2. Pull queue health — lead_search with status: ['new'], pageSize: 1, read the total. If >50, suggest /daily-triage.

3. Show the report — one section per area, scannable:

   Usage (this period)

   | Metric | Used | Cap | | leads delivered (month) | monthly.delivered | monthly.cap | | leads delivered (today) | daily.delivered | daily.cap | | products | products.used | products.limit | | team members | teamMembers.used | teamMembers.limit |

   Lead funnel (this period)

   • Delivered: stats.delivered
   • Still to review (NEW): stats.toReview
   • Hot: stats.hot
   • Acted on (saved + pushed): stats.actedOn → action rate actedOn/delivered

   Queue

   • X new leads — link to /daily-triage if >50

4. Diagnosis — pick the lowest-performing area:

   • Action rate <20% → leads surfacing but not actioned. Queue overload or low-quality leads. Suggest /daily-triage or /tune-audience.
   • Lots delivered, few hot → targeting is loose. Suggest /tune-audience.
   • Usage near cap → flag plan limits; mention upgrade path.
   • Low delivery → the Audience may be too narrow or paused. Suggest /tune-audience, and remind the user the Audience and its signal strategies are managed in the dashboard.

5. Monthly retro mode — when the user asks for a "retro", "review last month", or when run monthly: add a lead-quality retro, not just current state.

   • product_list, then for each product: lead_search by productId with status: ['saved', 'pushed'] vs status: ['archived'] over the period. The kept:archived ratio is the product's quality proxy — what the user kept vs threw away.
   • Surface the best product (high kept ratio) and the worst (high archived ratio), and inspect a sample of the archived leads (reasonLine) to spot the pattern.
   • Output: 3 concrete changes for next cycle (e.g. "ICP too broad on Product X — 38 of 40 archived were the wrong seniority"). Link the worst product to /tune-audience.

6. Three actions — end with exactly three concrete next steps, prioritized, each linking to a skill.

Guidelines

• Handle missing subscription. No active subscription (fresh trial / no plan) means no usage caps yet — state that plainly instead of rendering an empty table.
• Honest about thin data. Action rate isn't meaningful with <10 delivered leads. Say so explicitly.
• No vague advice. "Improve targeting" is useless; "Run /tune-audience on Product X — 38 of 40 leads archived, ICP too broad" is actionable.
• Stay scannable. Tables for numbers, bullets for actions. No paragraphs.
• Per client. For agencies, run this once per product (product_list → loop) so each client gets its own snapshot.