Job product-attributes-from-descriptions-unknown-allowed · revised 12 October 2026
Fill missing size, colour, material and pack columns for up to 20,000 products
A product list or ERP export gets attribute columns filled only where the description states the value, else 'unknown', with the quoted words as evidence. Tested on products you labelled.
You might be seeing
- Items are flagged for missing attributes although the description says the value
- Filling columns by hand for thousands of products is not realistic
No passwords, keys, card details or admin invites needed to start.
What usually happened
A model asked to fill a blank column will fill it, even when the description does not say. A guessed size or an invented barcode causes returns or a disapproved listing. The safe version only fills what the text states and shows the words it relied on.
Who it’s for: Merchant or feed manager with 'missing attribute' items, where the facts sit in the description text.
Usually starts when: 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.
Check whether this job fits
Answer these without sending files or logins. Nothing is submitted unless you choose to contact us.
What you get
- 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
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
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
How we know it’s done
Agreed with you before work starts. Each check produces evidence you keep.
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.
Evidence: The script output.
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.
Evidence: The per-attribute results table beside your labels.
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.
Evidence: The planted descriptions and the output.
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.
Evidence: The script check.
Sign-off. You inspect the results and the planted-blank test, and sign off in writing. Payment follows sign-off. You import the file yourself.
If it fails. If the held-back test is not passed, you do not pay for this fixed scope. You keep the results, the failures and the test set, and we agree whether to stop or re-quote. No surprise work.
When it fits, and when we stop
It fits when
- Product descriptions are in English and hold no personal data
- You label 100 products: 50 for tuning and 50 held back
- A named person approves the attribute list and the allowed values
- You import the filled file yourself
- Two people on your side label the same 20 products first and agree on at least 17
We stop and tell you if
- 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
What could go wrong
The job reads copies you send and writes files. It changes nothing in your systems, so there is nothing to revert. Our copies of your files are deleted when the job ends.
Scroll the table sideways to read it all.
| Risk | How we handle it |
|---|---|
| A model fills a blank with a plausible guess. | Silence must give 'unknown'. 20 descriptions that state nothing, which you add, must all return 'unknown'; one guess fails the whole test. |
| An invented barcode causes a disapproved listing. | Identifiers are never created. A script checks that every identifier is in the source. |
| Your files go to a model provider outside your control. | Before any file is shared we name the provider, its country and its retention terms in writing. Entry jobs run on redacted, synthetic or public copies, or on your own provider account with a key you hold. |
A separate automated reviewer checks the quote check and the planted blanks before hand-over and does not mark the held-back test. A named person on your side labels the test set and signs off. No person on our side is included unless your agreement names one, and at launch the work is largely automated.
How we deliver
We arrange the work and independent review, then show you the result against the agreed checks. You keep authority over your systems.
- Agree the attributes, the allowed values, the pass mark and the severe errors
- You label 100 products; 50 stay held back until the test run
- Two of your people label the same 20 products; they must agree
- Build the extraction with a rule that silence means 'unknown'
- Run the quote check, your planted blanks and the 50 held-back products
- Independent review of the evidence column and the identifier check
- Hand over the filled file and script; delete our copies
This is a one-off job, not emergency cover or a subscription. We confirm eligibility, the price, a start window and a delivery date before you accept, and work starts only after inputs, a secure handoff route and permissions are in place. Model usage up to a cap agreed in writing is included, and an enquiry creates no charge or booking.
Need to keep it working?
Need new products filled as they arrive? Ask about a quoted re-run on each new file.
Ongoing work is separately scoped and quoted: no monitoring, response-time guarantee or automatic subscription is included in this job.
Explore an ongoing managed lane, or mention the responsibility you need in your enquiry.
What you can check
This is a new service. We have not delivered this job for a client yet.
Other ways to get this done
- For a Shopify store, with a metafield-ready import file, attribute sets by category and a dry-run difference in a test store, use our Shopify attributes job. This job is for any spreadsheet or ERP list and does not import anything.
- Ask the supplier for a fuller product data file. Many suppliers have size and material columns they did not send.
- Google marks its product category as optional, so check which attributes your channel really requires first. support.google.com
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.
Send an enquiry
Send us
- 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
Later, once you agree
- A redacted or sample copy of the product file by an agreed secure route
- Your labels for 100 products
- A named person who approves the allowed values
- A written cap on model usage
You keep your product file and your store. We work on a copy you send by an agreed secure route, return the filled file and a script, and delete our copies when the job ends. We never import anything to a store or marketplace.
Email fallback: open your mail app
If website submission is unavailable, review and send the fallback email yourself. An email fallback is not a website receipt. Or write to hello@syntheticindustry.ai with “product-attributes-from-descriptions-unknown-allowed” as the subject.