A sender may upload a document twice, forward a copy by email or replace it with a corrected version. An AI intake process that treats every arrival as new work can create duplicate records. One that rejects matching names can lose legitimate revisions. Define identity before connecting extraction to a destination.
Related service: AI automation development- 01Check file identity
- 02Compare business identifiers
- 03Identify revisions
- 04Route or link
SECTION 01
Give the agency three different questions
First ask whether the exact file was already received. Next ask whether another file refers to the same business document, using an approved combination of fields rather than a filename alone. Finally ask whether the new content changes an earlier version and what that change should do to downstream work.
Document type matters. A repeated delivery reference may be a duplicate, an additional page or a separate partial delivery. Classification and extraction can provide candidate fields, but the business owner must define their meaning. Microsoft's classifier documentation is useful background for separating document instances inside one file; upload guidance covers validating the files before processing.
SECTION 02
Illustrative example: a corrected supplier quotation
A supplier sends quote Q-41, then emails a PDF named Q-41-final with a changed delivery date. The first file has already produced a draft purchasing review. The new file is not an identical upload, but it refers to the same quotation. Automatically creating a second review would confuse ownership.
A useful intake screen links the new revision to Q-41, highlights the changed field and marks the earlier draft as requiring recheck. If the previous proposal was already approved, the change does not silently inherit that approval. This hypothetical example makes version decisions explicit without assuming every supplier uses dependable revision numbers.
SECTION 03
Test collisions and ambiguous matches
Ask for tests using identical bytes with different names, matching names with different contents, reused reference numbers from different suppliers and multi-page documents arriving separately. Confirm that a low-confidence extracted identifier does not trigger automatic deletion. Uncertain matches should retain both sources while an authorized operator decides.
Handoff should include identity rules, matching evidence, revision states and the recovery path for a mistaken merge. Compare automatic matching with a review-first release when identifiers are inconsistent. Keeping a few ambiguous items visible is usually more manageable than hiding document loss behind an aggressive deduplication promise.
SECTION 04
Frequently asked questions
Can filenames identify duplicates reliably?
Only in a tightly controlled process. Outside it, names may change for identical content or remain unchanged for revisions.
Should duplicate uploads be deleted immediately?
Not by default. Preserve a receipt and relationship to the accepted item, then apply the approved retention policy to redundant copies.
Can AI decide which revision is authoritative?
It can suggest differences, but authority should come from documented business rules or an authorized reviewer, especially after prior approval.
PRIMARY REFERENCES
Sources and further reading
These references cover the standards, platforms, or published prices discussed in the guide. Worked examples and checklists are our editorial guidance.
- Custom document classificationMicrosoft Learn
- File Upload Cheat SheetOWASP
EDITORIAL METHOD
About this guide
We use AI to assist with drafting and editing. Catapult AI Work is responsible for the published content. Examples illustrate possible approaches; they are not client case studies unless identified as such.
Budget examples are not Catapult package prices. Check linked provider pages for current fees and plan limits before making a purchase.
Read the editorial policy