The same supplier under three names: how to reconcile messy records

An ERP vendor account, a legal entity, a corporate group and a manufacturing site are different things. One legal entity can have several vendor accounts. A group can own several legal entities. A company can operate plants with different products and exposure to disruption.

Write down the question before matching. For payment controls, the legal entity and vendor account matter. For a fire affecting a casting line, the operating site matters. A correct corporate match does not prove that an event affects the site supplying your part.

Model these relationships explicitly where the evidence supports them. Connecting a site to its parent company is often more accurate than merging both into one supplier record. Preserve the relationship’s dates when ownership or operating responsibility changes.

Retain the source system, key name and record ID for every imported supplier. “ERP / vendor_id / 10442” is a reference you can return to. A cleaned name is a comparison aid, not a reliable replacement for that identity.

Normalize casing, spacing and common punctuation for comparison, while keeping the original value. Do the same for addresses and identifiers with rules appropriate to the source. Do not strip country, legal suffix or location information so aggressively that distinct entities become indistinguishable.

A website domain or phone number can be shared across a group. An address can be a head office, registered agent or shared industrial site. These fields help establish a match, but their meaning matters as much as their apparent similarity.

Northfield has three records that might refer to the same casting supplier. The ERP and buyer spreadsheet share a tax identifier and operating address. The article names a company at that address but does not include the tax identifier. That supports two different levels of certainty.

A matching process should be able to say that a candidate needs review. An uncertain link is useful when its uncertainty remains visible; it becomes dangerous when the downstream answer presents it as a settled fact.

Give a reviewer the original records, the attributes that support the match, conflicting attributes, and the business consequences of merging them. Start with candidates that could change an important answer, such as attaching a disruption to a critical supplier, rather than reviewing solely by matching volume.

For ambiguous cases, identify what would settle the question: a verified legal identifier, a supplier confirmation, a site address in the source article or a contract linking the vendor account to the legal entity. Keep the candidate separate until your review policy permits the decision.

A buyer may know that Acme moved the casting line last year. Record which site was involved, when the change took effect, what supports it, and who reviewed it. That correction should change the relevant relationship; it should not imply that every past article was wrong.

When you discover a false merge, use a supported split or reversal workflow rather than manually changing database IDs. Check the impact on linked orders, claims and historical references. A clean-looking supplier list can still leave an old report pointing to the wrong entity.

Rerun ingestion with a small known dataset and check whether stable identities and reviewed corrections remain intact. Treat preservation as something to verify, not something guaranteed simply because an import finished successfully.

In a separate fictional case, report NEWS-12 describes a shutdown at Acme Packaging’s Riverside site. A name-only match attaches it to Acme Castings, vendor 10442, and the answer warns that Northfield’s casting orders may be affected. The report describes a real event within this example, but the relationship carrying that event into the answer is wrong.

A buyer checks the report’s address against the vendor records. NEWS-12 names 7 Riverside Way and Acme Packaging, vendor 20881. Northfield’s casting order PO-107 names Acme Castings at 18 Foundry Road. The different company identifiers, addresses and supplying-site record support separating the two companies and removing the incorrect site link.

Record the correction with those references and the reviewer, apply it through the supported review workflow, and ask the question again. Inspect both the changed answer and the older answer’s history. The casting delivery concern may still warrant a call; the unrelated shutdown no longer supports it.

Create a small review set containing known matches and known non-matches. Include difficult cases: shared names, group companies, renamed sites, different vendor accounts for one entity and similar businesses in different regions. Keep the expected decisions available for individual review.

Count wrong merges and missed matches separately. A run that produces the expected total supplier count can still contain both errors. Inspect the cases where the decision changed, and consider their operational impact: missing a duplicate is inconvenient, while attaching a shutdown to the wrong critical supplier can cause an expensive response.

Repeat the check when source data, mapping configuration or resolution behavior changes. Keep the source snapshots and the version used so that a later difference can be explained.

Semogram’s role is to connect and reconcile the records your business already holds, then support answers with evidence. Map stable source identities and relevant relationships into your business model. Review proposed matches against their source records and use the supported correction workflow when an identity is wrong.

After reconciliation, ask a specific question: “Which purchase orders and captured signals relate to the site supplying our casting line?” Check the entity and site links as well as the answer’s evidence. The important result is a supplier picture that can be questioned and corrected, not just a reduced row count.

That same approach applies beyond suppliers. Customer accounts, facilities and assets also appear differently across systems. The entity types and evidence change, but the discipline remains: preserve sources, establish identity with evidence, and keep corrections reviewable.