Who this is for
Merchant or feed manager with 'missing attribute' items, where the facts sit in the description text.
A marketplace or shopping feed flags missing size, colour, material or pack data, and the answers are in the descriptions.
The result
For up to 20,000 products, named attribute columns are filled only where the description states the value, otherwise 'unknown', each with the quoted words as evidence. A script confirms every quote appears in the description. No barcode or part number appears that is not in the source.
What is included
- One product list or ERP export of up to 20,000 products with English descriptions
- Up to five named attributes, such as size, colour, material, pack count, volume
- A quote from the description for every filled value, checked by a script
- A held-back test on 50 products you label
- The filled file, the script and the steps to run it
You receive
- The filled product file with 'unknown' where the text is silent
- An evidence column with the quoted words
- The quote-check script and its output
- A per-attribute results table
- The script and run steps
What is not included
- Estimating values the description does not state
- Creating or correcting barcodes or part numbers
- Regulated categories such as alcohol, gift cards or medicines, unless you supply the rules
- Importing the file into your store or marketplace
- Images or non-English text
What we need from you first
- The attributes you want filled, with the allowed values if you have a list
- Roughly how many products and how many are missing each attribute
- No real records, personal data or credentials in the first enquiry
Never send passwords, keys, customer records or confidential code in the first enquiry. Secure handover is agreed after scoping.
How we check it is done
- A script confirms every filled value's quoted words appear word for word in the product description. A value without a matching quote is removed.
- Across the named attributes, of the values filled on the 50 held-back products, at least 95% are right and none is invented, and at least 80% of the values your labels say the descriptions state are filled. The report gives the counts and the 95% lower bound (for example, 95 right of 100 filled shows at least 89%). An attribute with fewer than 20 filled values is left out of the pooled count and reported. A file that says 'unknown' everywhere fails.
- On 20 descriptions that state no attribute, which you add and we do not see beforehand, all 20 return 'unknown'. A single invented value fails the whole test.
- No identifier appears in the output that is not literally in the source files, and every identifier we copy through passes the GS1 check digit or is listed as invalid.
You inspect the results and the planted-blank test, and sign off in writing. Payment follows sign-off. You import the file yourself.
When we would stop or decline
- You want values estimated rather than stated
- The products are regulated and you cannot supply the rules
- You want identifiers created
- Two people on your side label the same 20 products and agree on fewer than 17
Questions
Why 'unknown' instead of a best guess?
A guessed size or barcode causes returns and disapproved listings. Unknown is a legal answer; a wrong one costs you.
Does it change my store?
No. You get a filled file and import it yourself.
What does the evidence column show?
The exact words from the description that the value came from.
Price and terms
£245 · untested offer price. £245 after the agreed test passes on the held-back part and you sign off. No payment before sign-off.
This is a new service with no published client results. The price is a starting point we have not yet tested with buyers. Nothing is ordered or charged by the enquiry. The full specification is on the Synthetic Industry catalogue.