The Return You Can Predict Before it Ships

Most retailers treat a return as something that starts when the package comes back. By then, the sale is booked, the item has shipped, and the only question is how quickly to process the refund.

That's too late.

By the time an order ships, most of the information that can indicate whether it's likely to come back already exists. Retailers should start treating return risk the way they already treat chargeback risk: as something that can be assessed before the outcome happens.

Chargebacks typically represent less than 1 percent of gross merchandise value (GMV), yet retailers have spent years building models to flag risky transactions. Returns are a much larger issue. NRF projected $849.9 billion in retail returns for 2025, including 19.3 percent of online sales, while McKinsey estimates retailers spend roughly $200 billion annually recovering value from returned goods.

The Signals Are Already There

Predicting returns doesn't require entirely new data. It requires connecting signals retailers already collect.

Merchandising Signals

Return risk varies by product attributes and how merchandise is presented online. For example, I've seen that apparel photographed on a live model can return at a different rate than the same item photographed flat, making something as seemingly minor as product imagery a useful predictive signal. That tracks with the shopper behavior the industry is seeing.

DHL’s 2025 E-Commerce Trends Report found that 54 percent of online shoppers have returned an item because it was the wrong size, while 39 percent said the product didn't look like the image. Those are signals retailers can address upstream through sizing, imagery and product descriptions rather than treating every return as unpredictable.

Shopper Behavior

Order history can reveal patterns that materially change return probability. In the same DHL study, 37 percent of shoppers said they had spent more to qualify for free delivery and then returned the extra items, while 16 percent cited ordering multiple sizes or colors as a reason for a return. That last signal is actionable in real time: data on products that are often ordered in multiple sizes is a good signal to surface a sizing guide or fit tool on the product page to proactively guide shoppers.

Order Context

Product mix, quantity, timing, fulfillment choices and customer history become more useful when combined at the order level rather than treated as isolated red flags.

Move From Blanket Policies to Order-Level Risk

Most return policies are designed for the entire business: 30 days or 60 days, free returns, one set of rules for every order. A more precise approach ties return costs to the order itself: category, price point and shipping method.

One practical method is expected cost by category and order value: weighing the likelihood of a return against shipping, processing, restocking, depreciation, lost margin and support costs. A $40 T-shirt and a $400 jacket carry very different economics, and industry estimates put that cost at 20 percent to 65 percent of an item's value.

With that view, retailers can vary return label costs by category and order value, and make sharper fulfillment and inventory decisions without evaluating individual shoppers.

What Teams Can Change

• Merchandising: Recurring patterns can point to issues with sizing, descriptions, imagery or promotions that should be fixed at the source.
• Operations: Predicted return risk can inform fulfillment economics and the true expected cost of an order before it ships.
• Customer experience: The highest-leverage moment is often before shipment, when better product information or proactive guidance can prevent avoidable returns.

The goal is not simply fewer returns. It's identifying which returns are predictable, costly and preventable.

The Next Shift is Upstream

Return policies will always matter, but they're reactive by definition. The next step is to move some attention upstream.

Retailers already have the data to estimate return risk before an order ships. The opportunity is to use it to decide where to absorb risk, where to improve the buying experience, and where to reduce preventable costs before the return ever happens.

Zack Peng is the co-founder and CEO of Seel, an AI operations platform helping brands and marketplaces transform traditionally complex areas of e-commerce, including customer support, returns and resale.