How we accidentally ended up deep in ecommerce product enrichment

Hi, just wanna share our story - might be useful for someone stuck in similar mess.

We started out doing general AI automation as an agency for ecommerce teams. But whatever the project was, we kept running into the same quiet blocker: catalogs. New supplier sends over some messy excel, someone disappears into spreadsheet hell to fix attributes, fill in missing details, and get everything into a usable format. Then you’ve got titles and descriptions that don’t actually match the underlying data, or differ across channels, so any automation on top of that feels shaky.

Everyone treated it as “just annoying ops” but across clients it was clearly the bottleneck. People tried own scripts and custom ai agents- which is cool at the beginning but once supplier formats start changing and volumes go up, you either drown in maintenance or pay a lot for compute if you’re not careful.

Because we kept seeing the same pattern, we ended up focusing on the algorithm side: how do you map/enrich/generate consistently while keeping data quality high and compute low. That work eventually became our own service productlasso, and in practice we’re usually able to run this cheaper and more reliably than the DIY script setups we saw.

If you’re thinking about building your own pipeline for ecommerce product enrichment , my only suggestion would be: spend a lot of time on how you structure the algorithm and compute, not just the prompt or script. The difference between a naive setup and an optimised one is huge in both cost and stability.

Curious how you’re approaching this- happy with your own scripts, using a tool, or still living in spreadsheets and copy‑pasting into chatgpt?