01
The same item can arrive under different descriptions
A supplier describes an item as “ACME bolt, M8 × 30, stainless.” Your catalog stores it as “Stainless steel bolt, M8, 30 mm.” The descriptions differ, but they may refer to the same item.
Record matching connects an incoming record to a specific existing record. StayCharted AMT learns from approved examples of those connections so your team can review suggested matches using its own business data.
02
Start with examples of confirmed matches
Use records your team has already matched to a trusted catalog or reference set. Retain the details people use to distinguish one item from another.
| Incoming description | Approved internal item |
|---|---|
| ACME bolt, M8 × 30, stainless | BOLT-M8-30-SS |
| Stainless steel bolt, 8 mm thread, 30 mm long | BOLT-M8-30-SS |
| ACME bolt, M8 × 40, stainless | BOLT-M8-40-SS |
The third record looks similar to the first two, but its length identifies a different item. Examples like these help make the distinction clear. Stable internal identifiers keep the intended match unambiguous.
03
Include the details that make a match meaningful
The relevant evidence depends on the task. Product matching may depend on size, material, manufacturer, or part number. Customer record matching may require business identifiers and location to distinguish companies with similar names.
- Pair source records with their approved destination records.
- Keep the reference identifiers consistent across examples.
- Include similar-looking records with different correct matches.
- Set aside reviewed examples to test the model separately.
A shared word or a similar company name is not enough to establish that two records represent the same thing. Prepare examples around the distinctions your team actually uses.
04
Plan for records that have no suitable match
Some incoming records represent new items. Others lack the detail needed to distinguish between possible matches. A suggested match should not force either situation into an existing record.
During evaluation, include unmatched records and close alternatives. Check whether the suggested destination is valid and whether its important attributes agree with the source. Your workflow should let a reviewer leave a record unmatched when the evidence is insufficient.
Focus on incorrect matches as well as correct ones. Linking a record to the wrong item can create more cleanup work than leaving it for review.
05
Put matching into the workflow you already use
Review suggested matches in a batch file, or use the StayCharted AMT API from the application that receives the incoming records.
- Incoming item or record
- Your application or batch workflow
- StayCharted AMT model
- Suggested matching record
- Validate and accept, or send for review
Your application controls what happens next: attaching an internal identifier, preparing an import, or sending an uncertain result to a reviewer. A prediction alone does not need to merge or overwrite records.
06
Begin with one reference set
Start with a well-maintained catalog or a reviewed set of customer records. Test new examples against it before expanding to additional sources.
- Are suggested matches pointing to the correct existing record?
- Which near matches still need human review?
- What happens when the correct item is absent from the reference set?
Use the results to improve your examples and review process. Re-evaluate as your reference data changes so the model continues to reflect the records your business uses.