AI enrichment
AI enrichment is an AI agent reading public pages row by row inside a workflow tool to answer a specific research question and fill a field, distinct from a static database lookup that only returns whatever a provider already has on file.
A static enrichment lookup checks a provider's existing database and returns what's there or nothing. AI enrichment is different: an agent reads live pages, one row at a time, to answer a question no database field was built to hold. Find the pricing page. Check whether a company uses a specific tool. Pull a founder's most recent public post. It's built for the same lookup repeated identically across hundreds or thousands of rows, which is a different job from open-ended research on a single account.
Where it fits, and where it breaks
The constraint sits in what the agent can actually reach. It's a web agent, not a login: it can't see gated content, can't authenticate into a CRM, can't read a file sitting in a company drive. And on a dynamic page or one with anti-bot protection, it doesn't always fail loudly. Sometimes it returns a blank cell. Sometimes a confidently wrong one, and nobody notices until hundreds of rows later, which is why a well-run enrichment step includes spot-checking a sample of output, not just running it and trusting the column.
Point this kind of agent at a narrow, well-scoped question and it's cheap and predictable, because the same small task repeats down every row. Point the same agent at an open-ended question, tell it to surface everything relevant about a company, and it burns cost on a job it wasn't built for, often producing a thin answer that still costs what a focused one would have.
In practice
AI enrichment earns its place on the row-scale side of the work: qualifying a list of companies against a fixed rubric, checking the same few signals across a target-account list, monitoring a set of accounts for the same trigger. A single account that needs deep, branching research where the next question depends on the last answer is a different job, usually handed to a person or a more interactive research tool instead.
What people get wrong
Teams assume an AI enrichment step either works or errors out clearly, the way a broken API call does. It usually does neither. A page it can't parse comes back blank or wrong without any signal that something failed, so the qualified list underneath can be silently half wrong unless someone checks a sample before trusting the column.
