How Company URL Matching Works

When you paste a company name into a URL research tool, a lot happens between "input" and "here's your match score." This article breaks down, in plain terms, how our matching approach turns a raw company name and a set of candidate results into a single 0–100 confidence score.

Step one: normalize before comparing

Raw input is messy. Before any comparison happens, both the input name and each candidate name are normalized: trimmed of extra whitespace, lowercased for comparison purposes, and stripped of common legal suffixes like Inc, Ltd, Corp, GmbH, or Pvt Ltd. This step alone resolves a large share of apparent mismatches — "Salesforce" and "Salesforce, Inc." should score as identical, not merely similar.

Signal one: exact match

If the normalized input and normalized candidate name are identical, that's the strongest possible signal, and the match scores at the top of the range. This handles the common case cleanly — most company names, once cleaned, either match exactly or don't.

Signal two: name similarity

For names that aren't an exact match, we measure character-level similarity using an edit-distance calculation (how many single-character changes it takes to turn one name into the other, relative to the length of the names). This catches close variants — typos, minor formatting differences, abbreviations — without requiring an exact string match.

Signal three: token overlap

Edit distance alone can penalize reordered or multi-word names unfairly. Token overlap looks at the individual words in each name and measures how many are shared, regardless of order. "Global Acme Logistics" and "Acme Logistics Global" would score well on token overlap even though their raw character sequences differ more.

Signal four: domain-to-name similarity

Finally, we compare the cleaned input name against the candidate's domain itself (specifically, the part before the extension — for example, "salesforce" in salesforce.com). A domain that closely resembles the company name is a useful corroborating signal, especially for smaller or lesser-known companies where the "name" field returned by a data source might be formatted differently than expected.

Combining the signals

These signals aren't used in isolation — they're blended into a single weighted score, with name similarity weighted most heavily, token overlap next, and domain similarity contributing a smaller share. The result is a single 0–100 number that's more robust than any one signal alone: a candidate can still score well even if one signal is weak, as long as the others are strong.

Reading the score bands

• 85–100 (High): multiple signals agree strongly — a good starting point, with light verification still worthwhile for anything important.
• 50–84 (Medium): a plausible match with some divergence between input and candidate — worth a manual look before relying on it.
• 0–49 (Low): weak overall similarity — treat as a lead to investigate, not a confirmed result.

Why alternatives matter

A single top score doesn't tell the whole story. If the top two candidates score close together, that's a sign the input name is genuinely ambiguous — multiple companies share a similar name, and the "right" answer depends on context the matching algorithm doesn't have (industry, region, or which specific company you meant). That's why the tool surfaces up to three alternatives alongside the top match, rather than hiding the runner-up results.

What the score deliberately doesn't claim

It's worth being explicit: this score measures name similarity between your input and a candidate result. It does not verify legal ownership of the domain, does not confirm the business is currently operating, and does not guarantee the domain hasn't changed hands since the underlying data was last updated. Treat a high score as "these names look like the same company," not as a legal or operational confirmation.