Import and score leads
When a customer hands you a list of leads (CSV from a webinar, scraped from an event registration, exported from a marketing tool), you want to land them in the CRM without duplicates, attach them to companies, link them to a sales pipeline, and seed an AI summary that drives later prioritization.
TL;DR
crm_bulk_create_contacts— built-in dedup against email + phone + do-not-contact list.- For the list's companies: one
crm_lookup_companiescall to find the ones you already have,crm_create_companyfor the rest;crm_update_contactto link. - For each lead worth a sales motion:
crm_create_dealagainst the rightpipelineId. crm_log_activityon each new contact so the touchpoint shows up in their timeline.crm_set_ai_summaryon each contact (and deal) so prioritization signals are populated.
Step 1 — bulk import with built-in dedup
{
"name": "crm_bulk_create_contacts",
"arguments": {
"contacts": [
{
"name": "Ola Nordmann",
"email": "ola@acme.com",
"phone": "+4799887766",
"title": "Head of Ops",
"tags": ["webinar-2026-04"],
"customFields": { "leadSource": "spring-webinar" }
},
{ "...up to 500 rows..." : true }
]
}
}
Returns { created: <n>, skipped: <m> }. Skipped rows match an existing contact by email/phone, or match a doNotContact: true contact (compliance). The tool does not tell you which rows were skipped — if you need to know, crm_lookup_contact per row first (slower but explicit).
Hard cap: 500 contacts per call. Chunk larger lists.
Step 2 — companies
Collect the list's unique companies, then check them all in one call. Prefer the business registration (country + registry number, such as a Norwegian orgnr) whenever the list has it — it is the company's legal identity, while a domain can be missing, shared, or changed. Up to 200 registrations and 200 domains per call:
{
"name": "crm_lookup_companies",
"arguments": {
"registrations": [{ "country": "NO", "number": "123 456 785" }],
"domains": ["acme.example", "globex.example"]
}
}
Only the inputs that matched come back, under matches, each with its company. Anything absent is new. invalid lists registrations whose number fails its country's format or check digit — fix or drop those rather than storing them without a registration.
Create each new company with what you know:
{
"name": "crm_create_company",
"arguments": {
"name": "Acme Inc.",
"domain": "acme.example",
"registration": { "country": "NO", "number": "123456785" },
"tags": ["webinar-2026-04"]
}
}
A registration another company already holds fails with crm_company_registration_conflict and names that company — link to it instead of creating a second one. There is no such guard on domain, which is why the lookup comes first.
Then attach each contact to its company:
{
"name": "crm_update_contact",
"arguments": { "id": "<contactId>", "patch": { "companyId": "<companyId>" } }
}
Step 3 — pipeline + deals
{ "name": "crm_list_pipelines", "arguments": {} }
Pick the right pipeline. For each lead worth pursuing:
{
"name": "crm_create_deal",
"arguments": {
"name": "Acme — spring webinar follow-up",
"pipelineId": "<pipelineId>",
"primaryContactId": "<contactId>",
"companyId": "<companyId>",
"amountCents": 500000,
"currency": "USD",
"expectedCloseAt": "2026-07-31T00:00:00Z"
}
}
If stageId is omitted, the deal lands in the pipeline's first stage. Use crm_change_deal_stage later to advance — see skill://crm/progress-deal-through-pipeline.
Step 4 — log the source touchpoint
Per imported contact, log what brought them in:
{
"name": "crm_log_activity",
"arguments": {
"type": "note",
"subject": "Source: Spring 2026 webinar",
"body": "Attended the April 24 product demo. Asked about EU data residency.",
"contactId": "<contactId>",
"metadata": { "leadSource": "spring-webinar" }
}
}
Setting contactId bumps the contact's lastContactedAt automatically — the timeline UI uses this for sorting. Activity types: note | call | email | meeting | task.
Step 5 — AI summary for prioritization
{
"name": "crm_set_ai_summary",
"arguments": {
"entityType": "contact",
"id": "<contactId>",
"summary": "Head of Ops at a 200-person fintech. Engaged on data-residency questions during the webinar — buyer signal. Likely budget cycle Q3.",
"nextAction": "Send EU data-residency one-pager + propose a 30-min discovery call."
}
}
Both summary and nextAction are free-text. null clears the field. Also set on the deal if you created one (entityType: "deal").
What NOT to do
- Don't create companies one lookup at a time.
crm_lookup_companiestakes the whole list; listing every company and comparing names missesAcme ASvsAcme Holding AS. - Don't bulk-create without a tag. Tag the import (
tags: ["webinar-2026-04"]) so it's trivial to find or revert incrm_list_contacts({ tag: "..." })if the source turns out to be junk. - Don't skip the AI summary step. Downstream prioritization (and humans skimming the list) read it. An empty summary is a buried lead.
- Don't paste raw CSV into
customFields. The map is freeform but lives in the contact row forever — only what the team will actually filter on belongs there. - Don't import without checking compliance. GDPR / opt-in: confirm the source had explicit consent before logging the touchpoint as
emailorcall.
Related
skill://crm/onboard-new-customer— slower path for a single named customer.skill://crm/deduplicate-contacts— how to merge a pair when a list turns out to overlap existing contacts.skill://crm/progress-deal-through-pipeline— moving the new deals through stages.skill://playbooks/customer-acquisition— end-to-end CRM + Conv flow that starts from this import.