The question everyone asks
What AI does, and what it can't.
Half of this problem is being commoditised very quickly. The rest isn't an information problem at all — it's permission, and the living network the data moves through.
The half that's getting cheaper
Reading a messy spreadsheet. Pulling specs out of a PDF line sheet. Normalising units. Writing a product description from a spec table. Matching a product to a category. Cleaning up an image.
All of it is getting cheaper every quarter, and we use it — AI does the first pass and a human approves it. Anyone telling you their AI enrichment is a moat is selling you the half of the category that's deflating.
One — permission
Every one of these is an agreement between two identified parties. None of them is a fact that can be found, inferred or generated.
Authorisation
A brand deciding that this dealer may sell this line, in this territory, at this tier. No model can grant it.
Attestation
A value carrying a named approver with the standing to approve it, under a rights basis. A copy of an attested record isn't attested.
Standing to correct
When a dealer says the dimension is wrong, somebody has to be the one whose answer counts. That's a relationship, not a confidence score.
Commercial terms
A price that exists because two companies agreed it. It cannot be inferred from public data, and if it could, it shouldn't be.
Consent
Permission to use an image, in this channel, for this purpose, until this date.
Two — the network it lives in
Permission is only half of it. The other half is the shared network the data lives in — every authorised relationship between two businesses that had to trust each other, and the living record they keep current together, down and up.
A model can draft a description in a second. It can't grant a thousand authorisations, it can't make a dealer send a real photo back from the shop floor, and it can't make a brand accept the fix. That exchange — authoritative data down, real-world content and corrections back up — is the moat, and it's built one relationship at a time. Copy the software in an afternoon; you still don't have the network.
What that means in practice
A scraper can copy every value on a product page. It cannot acquire the right to publish them, it cannot make a brand accept a correction, and it cannot produce a price that two parties agreed to — and it's looking in from outside a network it can't join, one that keeps the record current and enriches it from the field.
So the question isn't whether AI will do the gathering. It already does, and it'll do it better next year. The questions are who is allowed to say the value is right, and who's in the network that keeps it that way — and that's the part that takes years to build and can't be generated.
We'd rather you asked this early than late. If someone quotes you a lower price per SKU for enrichment, they're probably right, and they're solving a different problem.