What an underperforming launch was telling us about refunds
Bigblue does plug-and-play logistics for e-commerce. We pick, pack and ship orders for brands, and we build the software they and their customers live in.
Fashion brands get a lot of product back. This is why we built an automated return portal. A customer could start a return themselves, without emailing anyone. Yet most brands would ask their customers to contact support before accepting returns.
In late 2023, I led the return experience design with a product manager and two engineers. Our objective was to drive adoption up. The actual question was why a feature built to take work away from support teams was still gated behind a support request.
The brake was on refunds, not on returns
We sat down with brands using the portal and with fashion founders who were not. Almost all agreed with our starting assumption: easy, free returns sell more clothes than they cost in returned stock. They believed it. They still would not open the portal.
What they could not take was the risk of refunding more customers than they ever had before. So they added friction. The friction was not aimed at the return, it was aimed at the money leaving. A customer who has to email support is a customer someone can talk to first.
That friction had a second effect. Every return routed through a human, so the support inbox filled up and the product meant to remove that work looked like the thing generating it. Brands that reached that point gave up on the portal.
We had been treating the portal as a support-cost feature. The brands were treating it as a cash control.
Customers wanted a product back
I went into our own returns data. Close to 70% of returns carried a size-related reason. Not disliked, not faulty, not a change of mind.
That is what turned exchange into the obvious bet. Someone returning an item because it is the wrong size has already decided they want the product. A refund is the worst available ending for everyone in that story.
We could not afford the version we wanted
Pairing with the engineers, we costed in-portal exchange: reserving replacement stock, price differences, partial refunds, against whichever e-commerce platform each store ran on. Months of work. We had weeks.
So we made a bet. When a customer chose exchange, we issued a gift code for the value of their order. No inventory logic, no price-difference handling, no per-platform integration. The store's own checkout did the work we could not do. We called it Store credit.
It was not a smaller version of exchange, it was a different product. Real exchange hands the customer a replacement inside the return flow. We had to send them back into the shop, and bet that wanting the replacement badly enough would carry them through the detour.
I built and tested the flow end to end. People understood what the code was and told us it seemed fair. We shipped it on time.
Then it underperformed
A couple of weeks after the release, the share of customers picking Store credit over a refund came in well under what we had modelled.
Because we had shipped something we knew was compromised, the first explanation was sitting right there: a gift code is not a real exchange. It was wrong. Had we spent the quarter building the real thing, we would have hit the same wall.
A refund felt safer
I went back to customers who had been offered Store credit and taken the refund anyway. What they told me had nothing to do with the flow.
A refund felt safer. Not faster or easier.
Someone let down by a purchase is deciding under uncertainty, and cash back on their card keeps every option open, including never dealing with this brand again. Store credit asks them to commit to the brand at the exact moment it has disappointed them. They were not comparing return mechanics. They were pricing a risk.
What worked was money
I ran experiments on the decision rather than the mechanism: where the option appeared, what the copy committed to, how concrete we made the value. They moved the share of refunds down. None moved it enough.
What moved it was paying for the risk. Choose the exchange over the cash and you get the value of your order plus a 10% bonus on top.
Store credit asks the customer to take a bet and a refund does not. Better copy does not close that gap. Money does. It works for the brand too: a refund costs the full order and usually the customer with it, a bonus costs a fraction and keeps both.
Each of those tests took days, because underneath it was still a gift code, and the one that won was a cash bonus on a gift code. On the version we could not afford, the same test would have meant touching price logic on every platform we had integrated with.
