Capability
Resolve Every Row to a Real Product Before a Word Is Written
Identifier-based and evidence-based matching, confidence recorded with the reason, ambiguity routed to a person with candidates side by side, and unmatched rows treated as data problems.
The Question This Answers
How do I know the content is about the product I actually sell?
There is one error in catalog automation materially worse than the others: excellent content about the wrong product. It is fluent, specific, internally consistent, and it passes casual review precisely because it reads so well.
How Matching Works
Manufacturer together with model or part number is the strongest available signal, so identifiers are used first. Where identifiers are incomplete or ambiguous, broader evidence contributes. Every candidate carries a confidence indication and the reason it was proposed.
| Outcome | Meaning | What happens |
|---|---|---|
| Confident | Identifiers resolve to exactly one product. | Proceeds automatically. |
| Ambiguous | Several products fit comparably well. | Held for confirmation, candidates shown side by side. |
| No match | Nothing fits the identifiers provided. | Routed to exceptions as a probable data problem. |
Why Medical Equipment Is Hard
- Model numbers differ by one character across variants with materially different specifications.
- Internal SKUs encode condition or configuration and were never meant to be public identifiers.
- The same device is sold under different names by region, and under a legacy name after an acquisition.
- Accessories and consumables share most of a base unit's model string.
- Spreadsheets accumulate transcription drift from being retyped across systems.
Ambiguity Is Information
Confirming in Groups
Ambiguity clusters around systematic causes, so the ambiguous queue is grouped by the pattern producing it. Resolving the pattern typically clears dozens of rows at once, which turns what looks like hundreds of decisions into a handful.
Boundaries
What This Does Not Do
Stated plainly, because knowing the edges of a capability is more useful than another list of benefits.
- It does not quietly pick a winner when two candidates are close. That is the exact error it exists to prevent.
- It does not use price, condition, or commercial data as a matching signal.
- It cannot resolve a row whose identifier is genuinely not specific enough; that is reported rather than guessed.
- A confident match is not a guarantee. You can always inspect and override.
In Sequence
What Comes Before and After
Each capability hands something specific to the next. Following the chain is usually more useful than reading them in isolation.
Questions
Common Questions
What if the match is wrong?
You can inspect and change it. Because everything downstream inherits the matching decision, correcting it re-bases the evidence and content that followed from it.
How many rows typically need confirmation?
It depends entirely on identifier hygiene in your source file, and the count is less meaningful than it looks. Uncertainty clusters, so a large number of uncertain rows usually clears through a small number of decisions.
Private Beta
See This Working on Your Catalog
Capability pages describe the mechanism. A working session on your own products shows you how it behaves with your data, which is a more useful basis for a decision.