Clean a business list before importing it into the CRM, while you can still inspect and reverse decisions easily. Standardizing text is only part of the job. You also need to decide what counts as the same company, which values are trusted, and which existing records must never be overwritten.
Keep an untouched copy of the source and give every incoming row a stable import identifier. Those two steps make later investigation much easier.
Define the entity you intend to store
One row per company is different from one row per location or contact. A chain with ten branches may need one account and ten location records. Removing nine rows as duplicates would destroy useful information.
Write the intended CRM structure before matching records. Preserve the relationship between companies, branches, and people instead of forcing all three into a single flat identity rule.
Normalize without losing original values
Trim unnecessary whitespace, normalize domains, and standardize agreed location fields. Preserve original names and source values alongside normalized matching fields. A normalization rule can be wrong, and you need a way to inspect it.
Treat phone numbers and postal codes as identifiers, not arithmetic values. Separate extensions when relevant, and do not strip meaningful leading zeros. Record missing fields explicitly rather than replacing them with plausible-looking defaults.
Also check export safety before opening untrusted text in spreadsheet software. Data should not unexpectedly execute as a formula when staff review the import.
Use several signals for duplicate candidates
Exact source IDs or trusted existing CRM IDs can provide strong matches. Domains, company names, addresses, and phone numbers can support candidate matching, but each has exceptions.
A shared domain may belong to several branches or franchise operators. Similar names can identify unrelated companies. One phone number may serve several locations. Fuzzy matching should produce review candidates when identity remains uncertain, rather than automatically merging everything above an unexplained score.
For an illustrative review, two rows with the same domain and address could be a likely duplicate. Two rows sharing a brand name but carrying different branch addresses might be intentionally separate records.
Establish field ownership
Decide which incoming fields may update existing records. Fresh research should not replace sales notes, customer status, suppression flags, or a manually verified contact without an explicit rule.
When sources conflict, retain provenance and a checked date. Prefer a review queue over silently choosing whichever value happened to arrive last. Importing a new dataset should not make an unsubscribed or excluded record eligible again.
Rehearse the import and reconciliation
Test a small subset in a suitable test environment or controlled import. Check record creation, updates, branch relationships, character handling, and duplicate behavior. Confirm that the import identifier appears on the resulting records.
After the full import, reconcile accepted, rejected, merged, and unchanged counts. Keep an error report and a rollback or correction plan. A successful upload message does not prove the CRM now contains the intended entities.
Our lead data service can deliver structured company research with agreed identity rules. Describe your CRM and an example input file to scope preparation and delivery.
