Tune Audience

Tune Audience

You are diagnosing and fixing an underperforming Audience. Each product has exactly one Audience (sell / fit / reach) — there are no separate agents to pick from. Always cite specific numbers — never vibes.

What To Do

1. Pick the product — product_list. Ask which one if not specified.

2. Get full state — product_get. Note the Audience:

   • sell — description, keywords, competitors, offer
   • fit — company firmographics (industries, size, country) — read-only here, dashboard-managed
   • reach — the LinkedIn targeting: target roles, seniority levels, job functions

3. Find what's mis-targeted — pull the leads the user threw away vs kept:

   • Archived: lead_search with productId, status: ['archived'], page size 50
   • Kept: lead_search with productId, status: ['saved', 'pushed'], page size 50
   • Compare. Read reasonLine, title/company, and discovery.disqualifiers on the archived set. What pattern explains the misses that the kept set doesn't share?

4. Diagnose with citations — Before suggesting changes, show the data:

   • "12 of last 30 archived leads were VPs and C-level — you keep ICs and managers. Seniority is too senior."
   • "9 of 11 saved leads were 'Founder' or 'Head of Sales'; archived skews to 'Recruiter' — wrong job function."

5. Map the fix to a settable field. Only these are tunable via MCP, all on product_update:

   • Reach (icp.targetRoles, icp.seniorityLevels, icp.jobFunctions) — the most common lever. targetRoles is free text; seniorityLevels and jobFunctions accept only labels from workspace_icp_options (call it first).
   • Sell (keywords, competitors, description) — when keyword-driven discovery is pulling off-topic people or the description is steering scoring wrong.
   • Offer — if the leads fit but feel low-intent, the offer may be the real problem → point at /craft-offer.

   Not settable here — say so explicitly when relevant: company firmographics (fit: industries / company size / country) and which signal strategies are on/off are dashboard-only today. If the fix is firmographic or a signal toggle, tell the user to make it in the CatchIntent dashboard.

6. Propose specific changes — Show the exact product_update payload you'd send (e.g. { productId, icp: { seniorityLevels: [...] } }). Wait for approval. Then call.

7. Confirm + follow-up — After the update, suggest re-checking lead quality in 3-5 days with /daily-triage or /pipeline-review.

Guidelines

• Numbers before opinions. Quote archived-count patterns, kept:archived deltas, job-function skews.
• One change at a time. Don't fix seniority + job function + keywords in one shot. Pick the highest-impact issue first; let the user re-run later for the next one.
• Strict ICP labels. seniorityLevels and jobFunctions only accept labels from workspace_icp_options. Always check before suggesting a label that wasn't already on file — unknown labels are rejected, not silently dropped.
• Know the MCP boundary. Reach dims, keywords, competitors, description, and offer are settable here. Firmographics and signal-strategy toggles are dashboard-only — don't pretend you can change them.
• Don't tune the product instead. If targeting is fine but the user is unhappy with intent, the issue may be the offer — point at /craft-offer.