Case
AI-assisted product enrichment: from image chaos to a sellable range
Products in stock were missing images in the web shop. With AI support I matched the right image to the right item, so the range could be sold.
Got images
90% of priority items
Reviewed
4,700 images
Ready to publish
1,850 images
Time
58 hours
The challenge
The brand had just over 500 products in stock, but many of them had no images in the web shop. The images usually existed somewhere, spread across folders and seasons, with names that couldn't be tied to the right item.
Doing it by hand would take weeks. And every product without an image is a product nobody can buy, however good the advertising is.
What I did
1
Started in the product data, not the image folders
Which products have stock, belong to the right season and lack images? That gave 380 priority items out of 527 in stock.
2
Flipped the question: from product to image
Many active products weren't in the new photo shoots. I searched for images for each product, including older seasons. As long as the product looks the same, an older image works.
3
Mapped every image
Just over 14,500 files were reviewed and matched to item number and colour.
4
Built a clear naming scheme
Each image got a name with item, colour and image type (front, back, side, detail). Rules were set per product category.
5
Let AI do the heavy lifting, with me in control
The AI tool reviewed the images and suggested names. I approved before anything changed, then scripts renamed and sorted everything.
6
Worked in rounds with the client
Season by season, with a check-in with the client between each round.
How I used AI
The AI support made the project possible in 58 hours. But it only worked because it was clear who did what.
AI did
Decided what each image shows (front, back, side, logo detail, inside). Matched 14,500 files to stock and product data. Caught easy-to-miss errors, like wrong item numbers in folder names (confirmed by reading the hang tag in the image) and front and back swapped for a whole category. Produced suggestion lists in Excel and wrote the scripts that renamed and sorted.
I did
Set the rules: naming scheme, image codes and exceptions per category. Made the calls in unclear cases, such as whether an older image still shows the right product. Approved every round before anything changed. Kept the dialogue with the client and set priorities.
Close to 4,700 images reviewed in 58 hours, including building the method. By hand the same job would have taken many weeks.
The flow
- Product data
- Selection: active products in stock
- Search across all seasons
- Image mapping
- Naming scheme
- AI review and approval
- Web shop
Before and after
Example file names (made-up numbers).
Swipe sideways to see the full table.
| Before | After | Image type |
|---|---|---|
| IMG_4471.jpg | 10234_BLK_011.jpg | Front |
| Jacket-blue (2).jpg | 10234_BLU_012.jpg | Back |
| DSC_0093.png | 10234_BLU_017.png | Logo detail |
Results
90% of the priority products got images and can be sold online. 14,500 files were mapped, 4,700 reviewed and 2,500 renamed in four rounds.
Older seasons became an asset: images that would otherwise sit unused now show active products. The client got a clear list of what was missing, and the method can be reused for every new photo batch.
The fastest way to grow sales wasn't more media. It was making sure what was already in stock could actually be bought. AI made it possible in days instead of weeks, but the rules and decisions made the result reliable.
What the case shows
- Starting in the product data instead of the image folders decides what gets done first.
- AI does the heavy lifting fast, but a human approval before every change keeps the quality.
The client is anonymised. File names and item numbers are made up.
Want to talk about this case?
Happy to walk you through the details.