Action Plan: AI-Driven Lead Generation Engine

Establish access to all required systems and data sources. Understand the Smart Scout API schema, available fields, and rate limits. Align on segmentation criteria and campaign logic with Nick and the sales team. Initialize the TypeScript project and trigger.dev environment so the dev setup is correct before any pipeline code is written.

✅ Phase 1 complete. All tasks done as of May 29: SmartScout API explored, trigger.dev configured, credentials confirmed, segmentation spreadsheet reviewed. SPF/DKIM/DMARC verified across all 3 domains — all 6 mailboxes showing 100% warm-up reputation in Smartlead. Warm-up is now complete (ran ~3–4 weeks from May 28) — all 6 mailboxes are ready to send.

Build the automated pipeline that pulls Amazon brand data from the SmartScout API, normalizes it, and fans out to per-brand scoring tasks. Deduplication checks the suppression list in D1 (existing clients + lost/fired, matched by name and Amazon store slug) before processing — no current Navira client enters the pipeline. Snowflake is deferred to Q3; in the interim the full cohort (with raw signals) persists to Cloudflare D1, the pipeline state store, so it can be re-scored without re-pulling SmartScout.

✅ Complete: Full pipeline live — SmartScout ingestion, D1 dedup (100 suppressed: existing clients + lost/fired), D1 state store wired, monthly cron scheduled (first weekend of every month), batchTrigger fan-out to enrichment. End-to-end test passed May 29, 2026: 100 brands fetched and written to D1 in 19 seconds.

Implement the logic that scores and buckets brands based on the agreed-upon criteria, automatically assigning them to prospect segments. Each brand runs as its own trigger.dev task — if one fails, only that brand retries, not the whole pipeline.

✅ Built + deployed — the criteria & bands model is LIVE. Following the June 11 working session, the weight sliders were replaced by the "criteria & bands" model the team aligned on: each segment is defined by must-haves, nice-to-haves, value bands, and deal-breakers, in plain language (e.g. Segment D's locked decision: 4–10 third-party sellers as the sweet spot, eligible to 20). The definitions from the meeting were saved and re-scored on the new engine (now config v43), and the cohort has grown to ~43,981 brands after onboarding ~8,100 previously-missed 1P / Vendor-Central brands (they have no Amazon storefront, so the original ingest skipped them). What's in the Review Queue now is the criteria model, end to end, with ~36,000 non-suppressed brands. Live signals: annual revenue, YoY & MoM growth, 3P sellers per listing (+ the sweet-spot band), catalog size, Prime-coverage gap, Vendor Central dominance (now measured by dominant-seller coverage — the literal "Amazon owns X% of the brand" number), and Brand/page score (covers ~72% of brands while the product backfill finishes). The Scoring page's Step 2 draws every band over a live sample of real brands (drag the band directly on the chart), can back-solve bands from example brands you name, and re-scores real companies in the preview as you adjust — including the now-live Segment F — 1P in Decline, which targets Vendor-Central brands Amazon dominates (~480 brands land in it after calibration). ✅ Calibration complete — Nick and Jeremy signed off on the scoring. The first criteria run was deliberately wide; the calibration passes since then tightened it to the live config v43 — Hot now sits at ~15,600 (down from ~24,000), with ad-spend/TACoS sharpening Segment E's "growth + margin pressure" gate, score thresholds set, and Segment A's bands closed. All six segments (A–F) are now wired and launched in Smartlead.

Enrich segmented brand leads with real contact information for decision-makers at target companies. Apollo.io rate limits are handled automatically via trigger.dev's concurrency limiting and exponential backoff retry — no manual intervention needed if Apollo throttles. Note: LinkedIn API access is enterprise-only and heavily restricted — Phase 4 runs Apollo-only by default. LinkedIn is deprioritized unless access is confirmed before Phase 4 begins.

✅ Logic built May 31 — wired to the live Apollo API & verified June 10. Apollo.io enrichment complete — seniority/title/email ranking selects the best decision-maker contact per brand; brand_contacts table live in D1; idempotent upsert (safe to re-run). On June 10 the master Apollo API key came through: the integration was reworked to Apollo's current API (the original people-search endpoint had been deprecated) and proven with a single live test — one real decision-maker with a verified email landed in brand_contacts. Batch enrichment is now live (ENRICH_MODE=production — its own switch, separate from the outreach gate) — ~620 decision-maker contacts across ~618 brands enriched so far, each flowing into the Review Queue for sign-off before anything touches Smartlead. You control the spend: the Scoring page caps enrichment per tier — the top % or top # of brands by score — and shows the running Apollo-credit estimate live, so a run costs exactly what you choose (e.g. only the top few percent of Hot), in line with the 3–5 partners/quarter target.

✅ Pre-started + confirmed May 29: All 6 mailboxes live and showing 100% warm-up reputation in Smartlead Foundation Pool. Identities: Morgan Capri & Keira Loren (navirabrands.com), Levi Matthews & Leo Strawley (navirapartners.com), Emma Carter & Megan Brooks (navirahq.com). 4F/2M split — recommended 60–70% women. Warm-up is now complete — all 6 mailboxes are ready and campaigns can fire as soon as the cohort is approved.

✅ Live — all six segments (A–F) launched in Smartlead. Each segment, wired in one click from the Scoring page: (1) creates its Smartlead campaign, (2) loads the approved email copy straight from the Sequences page — all 6 steps, with per-brand data merge-fields (revenue, growth, seller count, Prime coverage, page score, etc.), and (3) attaches a reply webhook. At enrollment, approved & enriched contacts go in with those personalization fields filled. Replies, unsubscribes, and bounces flow back automatically so a brand is never contacted twice — and an out-of-office auto-reply correctly does not count as a reply. Positive replies hand off to HubSpot via Smartlead's native CRM integration, and a bounce-rate alert emails the team before Smartlead's 5% auto-pause. Wins are detected automatically — approved brands are matched against HubSpot, so converted partners surface in the dashboard's Partners view with no manual step. All six campaigns are now ACTIVE — Nick and Jeremy approved the scoring and copy, and sends are ramping conservatively (in line with the ~15–20/day plan) under the human review queue.

⚠ Safety model: A PIPELINE_MODE env var gates all Smartlead enrollment — it stayed on dry_run (logs only, no Smartlead API calls) through the entire build, and was flipped to production only after Nick and Jeremy signed off on scoring and copy. The human review queue still gates every brand before it enrolls, and sends ramp conservatively rather than firing the whole cohort at once.

Connect the segmented, enriched lead data to the outreach platform so campaigns fire automatically based on segment assignment. Smartlead.ai handles cold outreach and mailbox rotation, while HubSpot (the CRM) takes over after a reply or booked call. For the first 2–3 weeks post-launch, hot/warm brands land in a human review queue before enrollment — giving Nick a chance to sanity-check output before campaigns fire at scale.

Build an internal-facing dashboard giving the sales team real-time visibility into the pipeline and customizable hit lists. Nick can also be given view-only access to trigger.dev to monitor pipeline health directly.

✅ Dashboard is live — the Dashboard tab (navira.evanslist.tech) is built and in use. Shipped: the brand board with live pipeline counts, a Partners view of converted brands — wins are auto-detected (matching approved brands against HubSpot wins; built in Phase 5), so partners surface here automatically with their score + signals at conversion; pause/resume of Smartlead campaigns without leaving the page; and bounce resolution + deliverability health (open / bounce / unsubscribe with the 5% warning, including resolving a bounced contact in place and re-enriching it). Dashboard & deployment complete. A few optional extras were intentionally left out of v1 as unnecessary: standalone disqualification/snooze filters (per-brand holds & suppression already happen in the Review Queue) and a trigger.dev shortcut button (the Pipeline Map already gives live pipeline visibility). trigger.dev view-only access for Nick is available on request. The dashboard runs on navira.evanslist.tech until NAVIRA's migration of the system onto its own infrastructure completes (agreed at the July 17 handoff) — the final domain lands with NAVIRA then.