Here's what I'm designing and why, before I map it out visually. The core problem is you have five manual handoff points compressed into one chaotic process. The fix is to break it into four clean pipeline stages — each owning its own logic, with explicit routing, validation, and failure paths throughout. **Stage 1 — Ingest:** Typeform webhook fires into n8n. Deduplicate by email before touching anything else. **Stage 2 — Enrich:** Hit Clearbit. Don't block on it — if Clearbit fails or returns thin data, the lead still proceeds with a `low_confidence` flag, not a crash. **Stage 3 — Qualify & Route:** An AI classification node scores the lead (ICP fit, company size, intent signals) and outputs a structured tier: `hot`, `warm`, or `nurture`. Each tier routes to a different rep pool using round-robin or territory logic. **Stage 4 — Activate:** Write to HubSpot (contact + deal + owner assignment) and Notion (onboarding task board) in parallel, then notify the assigned rep via Slack. Throughout: every failure has a dead-letter path that logs to a monitoring record and alerts ops — no silent drops.Every node is clickable — tap any box to drill into how it should be built. --- Here's the implementation breakdown by stage: **Stage 1 — Ingest** Use a Typeform Trigger node (webhook mode, not polling). Immediately run a dedup check against HubSpot by email — if the contact exists and was created within the last 30 days, terminate the execution. Then normalize the payload: lowercase email, trim whitespace, validate required fields (name, email, company). Any missing required field routes to dead-letter, not chaos. **Stage 2 — Enrich** HTTP Request node to Clearbit's Enrichment API. Wrap it in an error-catch branch — if Clearbit returns a 404 or times out, set `enrichment_status: failed` and `confidence: low`, then rejoin the main path. Never let a third-party API failure kill the workflow. Merge Clearbit company data (size, industry, tech stack, funding) into the lead object. **Stage 3 — Qualify & Route** OpenAI node with a tightly scoped system prompt: score ICP fit (0–10), extract intent signals, classify tier (`hot` / `warm` / `nurture`). Output must be structured JSON — validate the schema before routing. Use a Switch node on `tier` field, not a chain of IF nodes. Rep assignment lives in a separate subworkflow (Execute Workflow node) so you can update territory/round-robin logic without touching the main pipeline. **Stage 4 — Activate** HubSpot and Notion writes run in parallel using two branches off the router output — don't chain them serially. Slack notification fires after both writes confirm. Each write node has its own error handler that logs failures without crashing the downstream notify. **Monitoring layer** Every failure path converges into a single logging subworkflow: write to an Airtable or Notion "ops log" table with execution ID, error type, lead email, timestamp, and failed node. Set an n8n error workflow to ping a dedicated Slack `#automation-alerts` channel. --- **Three risks to flag before you build:** 1. Clearbit rate limits — at scale, batch enrichment via their Bulk API is cheaper and more reliable than per-execution HTTP calls. 2. OpenAI token cost — the qualification prompt should receive only the fields it needs (company, title, form answers), not the full enriched payload. Saves ~60% on tokens. 3. HubSpot duplicate contact creation — even with your dedup check, race conditions can happen under load. Use HubSpot's upsert endpoint (create-or-update by email) rather than a raw create. Want me to go deeper on any stage — node-by-node build plan, the AI qualification prompt, or the rep routing subworkflow?