Handling Duplicate and Merged Accounts

Duplicate accounts are not just messy; they actively break scoring. When one company exists as three records in Salesforce, its signals split across them, so a genuinely hot account never crosses a threshold and reps never get alerted. Subsidiaries, acquisitions, and the gap between a website domain and a legal entity all create duplicates that fragment intent. The same problem hits contacts, where role-based inboxes and re-entered leads scatter engagement. A signal graph riddled with duplicates produces a queue you cannot trust.

The fix begins with treating identity resolution as foundational infrastructure, not periodic cleanup. Before scoring, routing, or enrichment runs, signals must collapse onto one canonical account, so resolution belongs upstream in the pipeline. This is the unglamorous plumbing that makes everything observable: with one record per real entity, you can finally see an account's true heat. Without it, every downstream metric inherits the duplication error.

Detection starts with strong match keys: primary domain, normalized company name, and firmographic fingerprints from Clearbit or Cognism. Domain is the most reliable anchor for B2B, but watch for parent-subsidiary relationships and companies with many domains, which need a corporate-hierarchy view rather than naive matching. Tools like Clay can help dedupe and enrich in one pass, and many CRMs and enrichment providers expose fuzzy matching. The goal is a deterministic-where-possible, fuzzy-where-necessary match that you can audit.

Resolution means picking a canonical record and a survivorship rule for which field wins when records conflict: most recent, most complete, or most trusted source. Decide whether to physically merge or to link duplicates to a master via an identity layer, since hard merges are hard to undo and can lose history. Whichever you choose, preserve the full signal history from every duplicate so the merged account keeps its accumulated intent. Losing that history would defeat the entire purpose of merging.

Merging once does nothing if your intake keeps minting new duplicates. Add dedupe checks at every entry point, so a form fill, an imported list, or an Apollo push checks against existing records before creating a new one. Normalize inputs at intake, lowercasing domains and stripping suffixes, so 'Acme Inc' and 'acme.com' resolve to the same anchor. Prevention at the door is far cheaper than cleanup after the fact, and it keeps the signal graph clean continuously rather than in painful periodic sweeps.

Make dedupe rules versioned, observable code, not a buried admin setting. Log every merge so you can explain why two records became one and roll back a bad match, and monitor duplicate rate as a health metric you watch over time. Tie this to your broader identity and enrichment process so resolution, dedupe, and enrichment reinforce each other. An account graph that stays clean is what lets allbound channels all act on the same trustworthy record instead of three competing copies.