Replaced rule-based deduplication with intelligent matching that handles abbreviated company names, different email domains, and name variations.
Every team that goes to trade shows knows this pain. You come back with an attendee list of hundreds of contacts. Some are people you already know, some are companies you already work with, some are net new. Importing the list straight into the CRM creates duplicates. Cleaning it up by hand takes a week.
The usual technical fix is to dedupe on email address or company name. The problem is those rules fail constantly. The same person uses different email addresses. The same company appears as "Bank of America," "BofA," and "BOA" depending on who typed it in. Rule-based dedup either lets duplicates through or rejects legitimate new records.
The engagement: a B2B firm that attends 8-10 industry conferences a year was sitting on multiple un-imported attendee lists because the cleanup work was too painful. The lists were going stale, the leads were going cold, and trade show ROI was suffering.
We built a workflow where the team uploads an attendee list to Claude with a short prompt ("check this against the CRM, flag duplicates, create new records for everyone else"). Claude reads the list, pulls account and contact data from Zoho via MCP, identifies obvious matches, flags fuzzy matches for human review, creates new records for clean new contacts, and adds notes to existing records noting the person was seen at the conference.
How it works: the reasoning happens in Claude. Rule-based systems can't tell that "John Smith" at "Bank of America" and "J. Smith" at "BofA" might be the same person, but a human looking at the full profile can. Claude can do that at scale. The team confirms the ambiguous cases in a quick review session, usually 15-30 minutes for a 500-name list.
What changed: list-to-CRM processing time went from days to hours. The team estimates around 95% match accuracy on the fuzzy cases, with the remaining ambiguity flagged for human decision rather than guessed at. The bigger win was follow-up speed. Lists that previously sat for two or three weeks now get processed and actioned the same week as the event, while the conversations are still fresh.
This pattern fits any team that periodically imports lists of contacts and worries about duplicates. Conferences are the obvious case. Other good fits: webinar attendee lists, content downloads, purchased prospect data, exports from another CRM during migration. The same workflow handles all of them. It works less well when source data is extremely sparse (just an email address with no name or company), because there's not enough signal for the matching logic to work with.
Drawn from real engagements. Details changed to protect client identity.