toflow-enrich-contacts

toflow — Enrich Contacts

Get verified contact data before reaching out. Enrichment is asynchronous for phone/email lookups — expect to poll for results.

Single person — email/phone

1. If you only have a LinkedIn URL and no CRM record yet, call enrich_person_by_linkedin first — it gets-or-creates the person and returns name, title, company, location, and any available contact data. Results are cached, so a recently-enriched person returns immediately.
2. For a verified email: call enrich_person_email with the person's CRM ID. If the result is pending, poll with get_person_enrichment_status.
3. For a phone number: call enrich_person_phone the same way, polling with get_person_enrichment_status if pending.
4. Always enrich email before enrolling someone in a sequence that includes email steps — don't enroll on a guess.

Bulk (a whole list)

1. Call estimate_bulk_enrich_list first and show the user the credit cost — always confirm before spending credits on a large batch.
2. Call bulk_enrich_list with the list ID and enrichment type (email or phone).
3. To enrich specific people rather than a whole list, use bulk_enrich_emails / bulk_enrich_phones with a set of CRM person IDs instead of looping enrich_person_email one at a time.
4. Poll completion with get_bulk_enrichment_status, passing the task IDs returned by the bulk call.

AI enrichment (custom, derived fields)

Use this when the user wants something beyond email/phone — e.g. deriving a company's tech stack, a person's likely pain point, or any field the standard enrichment tools don't cover.

1. Call get_ai_enrichment_guide before create_ai_enrichment. Present the model tiers it returns (they trade cost per run for capability — cheapest tier for simple extraction like job title, mid tiers for scoring/classification, a higher tier for ICP-fit/persona research, top tier for complex reasoning) and get the user's explicit choice before drafting any prompts.
2. Call list_attributes for the resource type to find the field the output should write into (target_attribute_id). If no suitable attribute exists, ask the user whether to create one (and what type — text/select/number/boolean) via create_attribute before proceeding.
3. Call list_ai_enrichments to check whether a similar enrichment already exists before creating a new one.
4. Show the full draft (model, cost, prompt, every field mapping with its target attribute) and get explicit approval before calling create_ai_enrichment.
5. Call run_ai_enrichment to execute it, get_ai_enrichment to inspect a specific enrichment's config/results, and update_ai_enrichment to adjust its prompt or target field.