Your CRM has 47 versions of Oracle and zero confidence about which contacts actually work there. Fixing one without the other is a recipe for expensive half-measures.
The entity problem is real
Enterprise CRMs average 20-30% duplicate rates. A 500K-record Salesforce instance? That's 100K-150K records that should be merged or linked.
The duplicates aren't just obvious copies. They're "Meta" and "Facebook" and "Meta Platforms, Inc." and "Instagram" living as four separate accounts with four separate pipelines. "Oracle" the database company coexisting with "Oracle Cloud" and "Oracle NetSuite" — three accounts, one company.
This isn't a data hygiene problem. It's a structural problem that corrupts every downstream metric: pipeline attribution, territory assignments, account-based campaigns, win rates, and revenue forecasting.
Entity resolution is necessary but not sufficient
A new wave of tools is emerging that uses AI reasoning to resolve entities — examining domains, email patterns, company registries, and LinkedIn profiles to determine which records refer to the same company. This is genuinely valuable work. The best of these tools provide human-readable justifications for every merge decision, confidence scores per resolution, and audit trails that let ops teams trust the output.
But entity resolution alone solves half the problem.
The other half: who actually works there?
Once you know that "Meta", "Facebook", and "Instagram" are the same parent company, the next question is immediate: which of the 47 contacts spread across those three accounts are still there?
Sarah Chen's record says "Facebook" — but is she still at Meta? Did she move to a startup six months ago? Is her email still valid? Is her title accurate?
Entity-only tools can't answer these questions. They resolve the company but leave the contacts untouched. You end up with a perfectly clean account hierarchy pointing at a pile of stale, unverified contacts.
The cross-verification advantage
The real power comes from resolving entities AND contacts simultaneously — and cross-verifying between the two.
When Salmon enriches a contact, it also resolves their employer against known entities. When it enriches an account, it cross-references contacts against the company record. This bidirectional verification catches errors that single-direction tools miss:
- A contact claims to work at "Oracle" — but their email domain and LinkedIn profile point to Oracle Investment Management, not Oracle Corporation
- An account shows "Mercury" — cross-referencing the contacts reveals they all have @mercury.com emails, confirming it's the fintech company, not the space startup
- A person's profile says "VP of Sales at Acme Corp" — but Acme Corp was acquired by BigCo eight months ago. The entity data catches the acquisition; the contact data catches the stale title
No single-level tool — entity-only or contact-only — catches all three of these.
What to look for
If you're evaluating entity resolution tools, ask these questions:
Does it resolve contacts too? If not, you'll need a second vendor for contact enrichment — and the two systems won't cross-verify against each other.
Does it work in real time? Batch resolution means your data is decaying between runs. Point-of-creation resolution catches duplicates before they fragment your pipeline.
Does it show its reasoning? Black-box merges create trust problems. You need confidence scores, source attribution, and human-readable justifications for every decision.
Does it map hierarchies? Knowing that Slack is owned by Salesforce is table stakes. You need the full tree: ultimate parent, subsidiary, division, brand, branch.
How long does implementation take? If the answer is "30 days with a dedicated FTE," that's a signal about architectural complexity that will follow you into production.
The punchline
Entity resolution matters. But treating it as a standalone capability — separate from contact enrichment, separate from verification, separate from continuous monitoring — means you're building on half a foundation.
The companies getting this right are the ones that resolve entities and contacts together, cross-verify between the two, and keep both layers fresh continuously. That's the complete picture.