Most stores treat returns as a cost of doing business online. The box comes back, you process the refund, you eat the shipping, and you move on. The numbers have quietly grown large enough that this acceptance is now one of the biggest unmanaged drains on a growing store's margin.
In 2024, U.S. shoppers sent back roughly $890 billion in merchandise, about 16.9% of everything retailers sold. Online, the rate runs close to double the in-store figure, because buying something you can't touch or try on produces more mismatches. For a store doing real volume, the difference between a 15% return rate and a 25% one is often the difference between a profitable year and a painful one.
Here's the part most owners miss: a large share of those returns were avoidable, and AI is genuinely good at preventing the ones that aren't.
Returns rarely show up as one line you can stare at. They're scattered across refunds, return shipping, restocking labor, and inventory that comes back in no condition to resell. That spread is exactly why they get ignored.
Processing a single online return costs most retailers somewhere between $15 and $30, before you account for the sale you lost. According to national returns figures from the NRF and Happy Returns, total returns reached around $890 billion in 2024, and the rate has climbed steadily as more buying moved online. Apparel takes the worst of it, with return rates commonly in the 25% to 40% range. Footwear isn't far behind.
None of that is fixed cost. It moves when you change what happens before the order ships.
Sort returns by reason and they fall into a few buckets. Damaged goods and honest change-of-mind are real, but they aren't the big one. The largest group, roughly 45%, comes down to fit, size, or color not matching what the shopper pictured.
That's a product-page problem, not a warehouse problem. A customer guesses at a size because the chart is vague, orders two, and sends one back. Multiply that across a catalog and bracketing (ordering several variants intending to keep just one) becomes a planned cost baked into every launch. Nearly two-thirds of shoppers now do it.
The point is that these returns are created upstream, in the listing, by an expectation gap you can close. That gap widens when the same product is described three different ways across your site, a marketplace, and a shopping feed, which is the argument for holding product information in one system rather than maintaining it channel by channel.
AI earns its place by closing the fit-and-size gap that drives the most returns, and the evidence is no longer theoretical.
Tools that predict a shopper's correct size from their past orders, measurements, and how similar customers fit have measurably cut size-related returns. One peer-reviewed study across 120,000 shoppers and six European retailers found a 22% average reduction. A fashion brand under the H&M group reported a 32% drop in knitwear returns after pairing AI fit guidance with richer 3D product views, a result documented in Stord's 2026 report on AI in e-commerce.
The same engine that recommends a size also powers better product recommendations, which is the other half of the story. A shopper who lands on the right item in the first place rarely sends it back. We wrote about that side of it in our piece on AI-driven personalization for e-commerce stores.
The common advice gets this backwards. Owners hear "AI fit tool" and reach for the expensive, multi-week integration before doing the cheap, obvious work that captures most of the same gain.
Real measurements instead of vague S/M/L. Model height and fit notes on the page. A return policy shown on the product page itself, where most shoppers look for it and where nearly half of stores fail to put it. These cost almost nothing and attack the same expectation gap.
Do those first. Then layer AI fit prediction and 3D views onto the categories where the math justifies it. A store that buys advanced tooling on top of a broken product page is paying for sophistication it can't yet use.
Not every catalog has a returns problem worth solving with AI. If you sell objective-spec products like electronics or hardware, your return rate is probably already in single digits, and your budget is better spent elsewhere.
If you're in apparel, footwear, furniture, or anything where the buyer is guessing at fit, feel, or scale, this is one of the highest-return investments on the table right now. The more volume you do, the faster it pays back.
The reason these projects stall usually isn't the AI. Running a fit demo on a handful of sample products is easy. Connecting it to your live catalog, real order history, and customer data, safely and without slowing the site, is where the time and budget actually go. The same gap decides whether any customer-facing AI works, which is why a support chatbot is only as good as the order and returns data behind it.
That gap is the difference between a pilot that demos well and something that shows up in your monthly numbers. It's also worth getting right, because a fit tool fed partial or messy data will quietly recommend the wrong size and make your returns worse. Done properly, this is squarely the kind of AI integration work that can earn back its cost inside a season or two.
If you take one thing from this, make it this: track returns as a margin metric, not an operations afterthought. Look at return-adjusted margin per product, find the handful of items driving most of your returns, and fix those first. The fixes tend to lift conversion while they're at it, since the same clarity that stops a return also helps a hesitant shopper commit to the order.
If returns are eating into your margin and you're weighing whether AI fit or recommendation tools are worth it, this is the sort of problem we help with at Encomage. Our work usually starts by looking at where your returns actually come from, then connecting the right tool to your store's real data so it moves the number instead of just demoing well.
Let’s build something powerful together - with AI and strategy.
