From CSV to passport: preparing your collection data
The minimum fields of a textile Digital Product Passport, the most common mistakes in real files, and how the AI structures them for you.
- A v1 passport needs **a few fields done well**, not an ERP.
- The minimum fields: identity, composition, origin, care, certifications, economic operator.
- The recurring mistakes in real CSVs are predictable — and therefore catchable.
- The AI **extracts and normalizes**; you **approve**. No value enters the passport without confirmation.
- An afternoon's work turns a whole collection into its QR codes.
You don't need a perfect system to start. You need the file you already have: the collection export, messy as it is. Here's how it goes from a confused spreadsheet to a compliant passport — and what the AI does (and doesn't do) in between.
The minimum fields of a textile passport#
A v1 passport doesn't need everything: it needs a few things, correct and traceable.
| Field | What it holds | Example |
|---|---|---|
| Identity | SKU and, if available, GTIN | DRE-001 · 08051234500012 |
| Composition | Fibers with percentages | Silk 82%, Cotton 18% |
| Origin | Where the garment is made | Made in Italy |
| Care | Washing symbols or instructions | 30°C, no tumble dry |
| Certifications | GOTS, OEKO-TEX, GRS… | GOTS (optional, valuable) |
| Economic operator | Who places the garment on the market | company-level, entered once |
Composition and origin are the core: they're the fields the consumer looks at and the ones regulation requires structured, not in free text.
The most common mistakes in real CSVs#
Real files are never clean. But the patterns repeat, so they can be caught:
- Compositions as free text ("100% organic cotton") instead of fiber/percentage pairs.
- Percentages that don't add to 100 — a classic to flag before generating.
- Countries written a thousand ways ("Italia", "IT", "Made in Italy") to normalize to an ISO code.
- One single column holding everything: it can be mapped, but it has to be recognized first.
- Certificates named but not attached: the name isn't enough, the document has to be uploaded.
What the AI does (and what you decide)#
The AI reads your CSV — or a spec-sheet PDF, or a label photo — and extracts the fields, normalizing them: the fiber becomes a code, the country an ISO, the composition a structure with percentages. For each field it shows you where it came from (the cell, the PDF line, the photo region) and how confident it is — the confidence.
What the AI does not do is decide for you. It's an extraction assistant, not a source of truth:
- no value enters the passport without your approval;
- if a value is missing, it's flagged as missing — never invented;
- low-confidence fields are highlighted so you verify those first.
That's the anti-hallucination principle: a passport is an artifact with legal weight, so every value must have a reviewable source.
The flow, in three steps#
- Upload the CSV (or PDF, or label photos).
- Review: the AI proposes, you fix the uncertain ones and confirm the right ones. The screen shows source and confidence for each field.
- Generate: a unique ID, JSON-LD, a branded passport page and a print-ready GS1 Digital Link QR.
An afternoon's work turns a whole collection into its passports. No integrations, no consultants.
Where to start today#
Don't wait for the perfect file: start from what you have. Structuring your data now is also the cheapest way to arrive ready for the 2028 ESPR obligation — one series at a time, with no final rework.
Frequently asked questions
What file do I need to start?
The one you already have: a CSV or Excel of the collection is perfect, even messy. Alternatively the AI reads spec-sheet PDFs or label photos. No special format required.
Does the AI invent missing data?
No, never. If a value isn't in your source, it's marked as missing and it's up to you to provide it. The AI normalizes and proposes only what it finds, always with the source traced.
How do you handle badly written compositions?
Fiber/percentage pairs are extracted and normalized; if the percentages don't add to 100 or the text is ambiguous, the field is flagged as low confidence so you verify it before generating.
Do I need GTINs?
Not to start. You can begin from the SKU and add the GTIN later. Dig into how garment identity works.
How long does a whole collection take?
It depends on how clean the file is, but the order of magnitude is an afternoon: upload, review the uncertain fields, bulk-generate pages and QR codes.
Generate your collection's passports
From product sheet to compliant, hosted, print-ready QR codes.
Get started