IMPLEMENTATION GUIDE
Bound the fields the agent can propose
Define an allowlist of CRM fields and map each one to a supported enrichment attribute. Use deterministic mapping for numeric or structured fields such as employee count. An agent can explain a proposed change or suggest a category mapping, but it should not invent a value absent from the profile. Keep existing verified values visible to reviewers.
Treat returned text as data
Pass a structured profile and field schema to your assistant, and treat company descriptions or headlines as source text rather than instructions. Require each proposal to identify its source field and validate output against the allowlist and CRM field types. Keep ambiguous matches and conflicting employer associations out of automatic writeback.
Review before writing
Store the record ID, enrichment task IDs, proposed values, and existing-record version with the draft. Re-read the CRM record before applying an approval so a teammate’s intervening edit is not overwritten. Your backend handles API keys, signature verification, and idempotent writeback. EnrichLoops supplies profiles; your assistant and CRM implement the review process.
Connect the workflow
Use the API overview for request and task handling, or the Loop guide for webhook delivery.
Enrichment APIRetrieve completed company and people profiles for your backend tool.LoopsReceive enrichment results through signed webhooks.Common questions
Is the AI CRM assistant provided by EnrichLoops?
This guide describes an assistant you build using EnrichLoops enrichment APIs. Your application implements the agent, review interface, and CRM writeback.
Should an agent replace manually verified CRM fields?
Apply your field policy and show conflicts for review. A returned enrichment field should not silently replace a manually verified value or a newer teammate edit.