cocottetech/@platform/codebase/@features/prospecting/docs/prospecting.brief.md

3.9 KiB
Raw Blame History

prospecting.brief — parent product brief

Code: prospecting Tenancy: Person-primary, Org-overlay (per DESIGN.md §5) Primary interface: AI (copilot fleet). Web FE is a low-priority secondary surface. API plane: platform.api on black (per INFRA.md §4). No new API service. Status: P5 incubation; existing prospect-resolver in production under ai-copilot/.

North-star metric

Provider free-time per booked appointment. Every prospecting subfeature is judged by whether it removes provider work, not by raw conversion. Conversion lift is the means; reclaimed hours are the end.

Secondary metrics:

  • Time-to-first-response on inbound prospects (target: < 60s, AI-drafted, provider one-tap approve).
  • Qualification accuracy (precision/recall of "this prospect will book" predictions, measured via held-out outcomes).
  • Cross-provider signal contribution rate (% of consented providers whose prospect outcomes feed the graph weekly).

Problem

A provider's day is dominated by repetitive prospect work: re-qualifying tire-kickers, re-drafting the same intro, manually triangulating "have I talked to this person before" across surfaces. Each provider sees their own slice of a prospect; the same person ghosting provider A and converting with provider B teaches nobody. And the highest-value providers spend prime hours triaging instead of working.

Solution shape

Four subfeatures, all sharing prospects + prospect_touchpoints on platform.db, all addressable through the copilot fleet:

# Subfeature What it removes from the provider's day
1 Cross-provider graph Re-qualifying prospects whose history exists elsewhere in the consented network.
2 Warm intros Manually fielding cold pitches that another provider's copilot has already vetted.
3 Auto-qualify + draft The first 3090s of every inbound message: read, score, draft a reply.
4 Internal marketplace (v2) Hand-offs for overflow / wrong-fit prospects to vetted peers.

Constraints

  • No auto-merge across providers. Per Y-cross-org-marketplace.brief.md §Y4, prospect rows stay per-provider. The graph surfaces coop-attested intel about the same subject, not a merged row.
  • Person-level consent for graph contribution. Providers opt in per-prospect or per-policy; the consent state is auditable and revocable.
  • AI-first, FE-thin. Web FE surfaces only what a copilot cannot conversationally render (dense tables, multi-prospect comparison, batch ops). Default consumption is voice/chat.
  • Single API plane. All endpoints attach to platform.api; no new microservice. New tables migrate via infrastructure/migrations/.
  • Platform actions live upstream. Any new "do X to a prospect" action contributes to @ai/@skills/platform-prospecting/actions/*, not into this repo.

States (parent feature)

  1. Empty network — provider just onboarded, no consented peers, graph is provider-local. Auto-qualify + draft still works.
  2. Sparse network — 210 consented peers, graph signal is suggestive but not authoritative.
  3. Dense network — 50+ consented peers in a category/region; warm intros become viable.
  4. Saturated — overflow routing (v2 marketplace) becomes meaningful.
  • cross-provider-graph.brief.md, warm-intros.brief.md, auto-qualify-draft.brief.md, internal-marketplace.brief.md
  • @features/ai-copilot/docs/specialist-prospect-resolver.contract.md (row-level dedup; unchanged)
  • @features/ai-copilot/docs/prospect-detail.screen.md (resolved-prospect drawer; unchanged)
  • @features/ai-copilot/docs/Y-cross-org-marketplace.brief.md (external marketplace; orthogonal to internal v2)
  • @features/ai-copilot/docs/N-provider-coop.brief.md (peer-vetting primitives; warm intros + marketplace depend on it)
  • DESIGN.md §5 (tenancy), §7 (naming)
  • INFRA.md §4 (single API plane on black)