How many people does it take to get a product live? 

At a glance

• Product induction is the work of taking product data from a third-party brand or supplier and getting it live on your site: mapping attributes to your taxonomy, rewriting copy to house style, matching images to SKUs. North American teams call it product onboarding.
• Most retailers do it by hand. A 50,000-SKU catalogue at 30 to 40 attributes per SKU is more than 1.5 million data points, and enrichment teams of ten to fifty people cost seven figures a year.
• AI handles the mapping and rewriting now, which is what made the headcount optional. The retailers we speak with have stopped asking whether it’s possible.

Ask a retail commerce leader where their product data comes from, and you’ll usually get an org chart instead of an architecture diagram. Somewhere in the business, often offshore and rarely visible to the board, sits a team whose entire job is retyping, reformatting, and chasing product information so that new items can go on sale. They don’t merchandise or price a single item. Their job is getting the data into the building. 

That work has a name: product induction. A third-party brand or supplier hands over everything that describes an item, from the physical facts (SKU, weights, dimensions) to the copy and images that do the selling. The retailer transforms all of it to match its own taxonomy and standards and gets the product live on site. The data feeds your PIM; the images feed your DAM. North American teams tend to call the same work product onboarding or catalogue management. The commercial terms behind the assortment vary (traditional wholesale, cost price, or commission), but the work does not. Every retailer that sells other people’s products does it. Almost none of them have industrialised it. 

The workaround became a department 

The pattern repeats across the industry. A brand signs. Someone emails over a spreadsheet template. It comes back in the brand’s own format, with its own attribute names and its own idea of what a “short description” is. Then a person opens that file and starts translating: mapping the supplier’s shirt-length field to yours, rewriting descriptions to house style, renaming image files by hand because none of them match SKUs. The organisation has built a human bridge: people standing between a third party’s data model and the retailer’s, carrying every product across by hand. 

Multiply that across hundreds of third-party brands and thousands of SKUs and the workaround quietly becomes a department. The retailers we speak with describe the same shape everywhere: dedicated enrichment teams of ten, twenty, sometimes fifty people, in-house or offshore, processing queues that stretch to weeks per batch, and backlogs that grow faster than anyone can clear them. The arithmetic explains why: a 50,000-SKU catalogue at 30–40 attributes per SKU is more than 1.5 million individual data points, every one of them touched by hand. 

The cost hides in three places. The first is direct headcount: whether you run the team in-house or outsource it, the bill runs to seven figures a year, and outsourcing is no longer the bargain it once was. The second is opportunity cost: the assortment you didn’t launch because the queue was full. The third is error cost: manually retyped data fails at a predictable rate, and every failure becomes a suppressed listing, a support ticket, or a return. 

Why AI changed the retailer’s maths 

For years this was tolerated as a cost of doing business, partly because there was no obvious alternative. No retailer can push thousands of brands onto one data standard. Each sends product information in its own format, and someone has to translate. There’s also a legal reason the templates ask for so much. In the UK and Europe, the retailer is on the hook for the accuracy and safety of everything it sells, whichever party is the seller of record, so teams capture every attribute in writing rather than answer later for a wrong one. 

What has changed is the translation work itself. The field-by-field mapping and rewriting that human bridges exist to do is exactly what AI now handles well, and it runs continuously rather than as a one-off migration. The retailers we speak with have stopped asking whether this is possible. The question is why they’re still paying a headcount tax for a data transformation problem. 

The questions to ask your own operation 

1. How many FTEs, internal and outsourced, touch product data between a brand’s handoff and go-live? 
2. What is your actual cost per SKU onboarded? Most retailers have never calculated it. 
3. What percentage of your catalogue was enriched in the last 12 months? If it’s a fraction, your backlog is compounding. 

And then the question that reframes all the others: if your onboarding cost dropped 80% tomorrow, what assortment would you launch that you can’t justify today? The gap between that answer and your current range is what the human bridge is really costing you. 

Rithum works with some of the world’s largest retailers on exactly this problem: getting products from third-party brands and suppliers live faster, whether the commercial terms behind them are traditional wholesale, cost price, or commission. If product induction is consuming a team where it should be consuming a workflow, talk to us. Talk to our team. 

This is part one of our product induction series. Parts two and three cover what bad product data does to your returns line, and why time-to-online is the growth metric missing from most dashboards