Supplier product data arrives in the least glamorous formats in the business: emailed spreadsheets with renamed columns, changed units, new part numbers and prices whose effective dates are buried in a note. People merge those files into a catalogue or ERP by hand, and the errors appear later as margin problems, order failures or arguments with customers.
This composite workflow covers catalogue and price-list updates. The aim is not to let a model edit the product master freely. It is to make the proposed change structured, validated and reviewable, then load only approved records with a complete record of where each value came from.
The workflow as it usually runs
- 1A supplier emails a spreadsheet or CSV with products and prices.
- 2An operator works out whether it is a full file or a set of changes.
- 3They rename columns and convert formats to the internal template.
- 4Supplier part numbers are matched to internal SKUs.
- 5Descriptions, units, pack sizes and categories are cleaned.
- 6Price differences are checked and sometimes approved.
- 7The file is imported, errors are fixed and the import is repeated.
- 8Nobody retains a clear field-by-field record of what changed and why.
The eight opportunities
1. Control intake and versions
Save every source file, supplier, received date and effective date under one batch. Detect whether a newer file supersedes an earlier one. This prevents two operators from loading different versions and creates the provenance needed to explain a later price or specification change.
2. Map columns to a canonical model
Known supplier templates should map deterministically. For unfamiliar files, a model can propose that “dealer cost” means supplier price or that “case qty” means pack size, but a person approves new mappings before they become reusable rules. The mapping itself is saved for the next file.
3. Match supplier items to internal SKUs
Use exact supplier identifiers and maintained cross-reference tables first. Then compare manufacturer codes, barcodes and normalized descriptions. Ambiguous or new products become a review queue with likely matches. A wrong merge is much harder to unwind than a delayed new item.
4. Normalize units, packs and formats
Convert dates, currencies, dimensions and units through explicit rules. Validate pack size against known ranges and do not infer a conversion when the source is unclear. “12” may be twelve units, twelve cases or a twelve-pack; the workflow should surface the ambiguity instead of deciding silently.
5. Validate required and dependent fields
Check required identifiers, category rules, valid tax codes, positive prices and relationships such as unit cost multiplied by pack quantity. Compare the file with prior supplier data to catch dropped columns or a full catalogue suddenly containing a tenth of its usual records.
6. Separate ordinary changes from commercial exceptions
Calculate price movement, margin impact and effective date. Changes inside approved boundaries can proceed; large increases, retrospective dates and affected customer contracts go to purchasing or finance with the impacted products and amounts attached.
7. Load through a staged, repeatable process
Write approved changes to a staging area, run ERP validations, then promote them as one batch. Failed rows remain attached to the batch with exact reasons. Re-running the batch must not create duplicate products or apply the same price change twice.
8. Publish a change report and retain provenance
Show products added, changed, rejected and waiting for approval. For each field, retain source file, original value, normalized value, rule or reviewer and load result. This turns a fragile spreadsheet import into an auditable product-data process.
What we would build first
Choose one high-volume supplier and build intake, mapping, matching and validation through a staged preview. Let operators approve the proposed changes and compare the batch with their current import. Once the mapping stabilizes, automate only the changes inside clear commercial and data-quality boundaries.
The rule for AI here
Use AI to propose a mapping or match from messy text. Use deterministic validation and human approval to change the product master. The model handles ambiguity; it does not own the catalogue.
Return to the seven distributor workflows and choose the next process to inspect.