Target Account Scanner: deployment guide

Pack Version: 1.0.0 · Time: 0.5–1 day · Owner role: RevOps / GTM Engineer

Follow in order. Each step ends with a check; don't proceed on a failed check.

Step 1: Agree the ICP config (60–90 min, the meeting matters more than the file)

1. Get RevOps and sales leadership in one call. Fill every field of the config: exact industries, adjacent industries, employee range, revenue range, target/serviceable regions, must-have tech, disqualifying tech, target hiring functions, intent topics, band thresholds.
2. Save as `icp-config.json`, version it `icp-config-1.0.0`, commit to your repo.

Check: the config has no field marked "TBD" and leadership has seen the band thresholds.

Step 2: Create HubSpot properties (20 min)

Follow `hubspot/setup-guide.md`: property group + five properties + dedicated private app.

Check: all five internal names resolve in Settings → Properties search; private app token stored in your secrets manager.

Step 3: Build the Clay workbook (2–3 hours)

Follow `clay/clay-setup-guide.md`: source candidates, normalize and dedupe domains, add the pre-enrichment filter, add enrichment columns, add the HubSpot lookup columns, create the `to_scanner` view.

Check: the view's columns match `clay/clay-template.csv` headers exactly; run 10 rows end to end and confirm fill rates look sane before enriching the full list.

Step 4: Import and configure the n8n workflow (1–2 hours)

1. Import `n8n/workflow.json`.
2. Work through every node whose notes say CONFIGURE: schedule, ICP config (paste your versioned config), Clay URL + credential, HubSpot credential, Slack channel.
3. Leave the schedule trigger disabled for now.

Check: manual execution against `demo-data/sample-input.json` (paste into a temporary Set node in place of the Clay fetch) reproduces `demo-data/sample-output.json`: Acme 88/Target, Petal 10/Exclude, Meridian 30/Exclude, Northwind 65/Monitor + skipped_customer with no CRM write.

Step 5: Shadow mode (2 runs / 2 weeks)

1. Disconnect the HubSpot write node (keep Slack digest and run log connected).
2. Run weekly on the real universe. Review band distribution and the digest each week.
3. Expected tuning: Target band >25% of universe means weights are too generous, usually industry. Update the config, bump its version, rerun.

Check: two consecutive runs where the operator reviewing the digest disagrees with fewer than 1 in 10 band assignments.

Step 6: Go live (30 min)

1. Reconnect the HubSpot write node. Enable the schedule trigger.
2. Create the two HubSpot views from `hubspot/setup-guide.md` and hand "New Targets this week" to whoever owns the review.
3. Complete `qa/qa-checklist.md` with a reviewer who is not the builder. File the signed checklist in the repo.

Check: first live run writes only `icp_*` properties (diff a sample company before/after), digest posts, run log written.

Rollback

Disable the schedule trigger. Agent-owned properties can stay (they're inert) or be bulk-cleared via a HubSpot list + property update; nothing else in the CRM was touched. Re-entering shadow mode is Step 5, not a rebuild.

What not to do

Don't wire the Target band into sequence enrollment, territory assignment, or lifecycle stage. Downstream agents (#003, #004) handle prioritization with their own review gates. This agent proposes a universe; humans approve it.