AI is increasingly influencing what consumers buy online. But what happens when an AI shopping tool recommends the wrong product?
The customer may request a refund or replacement, but it’s the retailer — not the AI company — that typically handles the return, pays the additional shipping costs, and deals with an unhappy customer. For small e-commerce businesses operating on thin margins, those expenses can quickly add up.
I talked to Kyle Henzel, COO of Ship.com, a shipping and logistics platform, about the hidden costs of AI-influenced returns, why traditional conversion metrics don’t tell the whole story, and what small online sellers can do to protect their profits as AI plays a bigger role in shopping decisions.
The Real Cost of AI-Influenced Return
Rieva Lesonsky: What does an AI-influenced return actually cost a small e-commerce seller?
Kyle Henzel: A lot more than the refund. Say a shopper asks an AI tool for a good hiking boot, and it points them to yours, but it’s the wrong boot for what they needed. You refund the sale.
You may pay for the return label. You pay to ship the replacement, in a new box with new packing material. Someone on your team spends time on emails and processing. And the item that comes back isn’t always sellable at full price, or at all.
Then there’s the cost that never shows up on an invoice. When an AI picked the product, the customer typically blames you, not the tool. That customer often doesn’t come back, and they may leave a review. For a small seller, one order like that can erase the profit from several good ones.
Lesonsky: How do returns, replacement shipments and customer-service demands erode already-thin margins?
Henzel: Shipping is already one of a small seller’s highest costs. A return makes you pay for it two or three times on the same order: the original shipment, the return, and the replacement. You pay all of that on a sale where the revenue is now zero or close to it. So, it isn’t a break-even order. It’s a loss.
Customer service is the part people forget. Every return means emails, a refund, inspecting the item, and restocking it. In a small business, that time comes from the owner or one or two employees who should be doing something else.
The bigger problem is that these costs are scattered. The return label is on one bill, the replacement is on another, and the labor isn’t on any bill. Most sellers never add them up per order, so they don’t see the leak until the quarterly numbers look wrong.
Lesonsky: Are small sellers more financially exposed to bad AI recommendations than
larger retailers? Why?
Henzel: Yes. A large retailer has return centers, volume pricing on return shipping, outlets for returned goods, and enough sales to absorb a bad week. A small seller has none of that. The returned item comes back to the same garage or small warehouse, and the owner decides what to do with it.
Concentration matters too. If an AI tool keeps sending the wrong buyers to one product at a big retailer, it barely shows up in the numbers. A small seller often depends on a handful of products, so the same problem with one listing can hit a real share of the business.
Small sellers also have less control over how AI tools read their products. Big brands have teams feeding clean product data to every platform. A small seller may have a listing it wrote once, years ago, and AI is now making recommendations from it.
Measuring What Really Matters
Lesonsky: Why is conversion the wrong measure of success for AI commerce?
Henzel: Conversion stops counting at checkout, and the costs start after it. A tool that gets more people to click buy looks great on a dashboard. If a big share of those orders come back, you didn’t gain sales. You paid for returns.
A sale isn’t really a sale until the customer keeps the product. The better measure is how many orders stay sold and what you made on them after shipping, returns, and replacements. I’d track kept orders and true profit per order. An order that turns into store credit is still a kept order. An order that turns into a refund isn’t. This matters even more with AI, because these tools get better at whatever they’re rewarded for. If they’re rewarded for clicks, they’ll get better at clicks. If they learn from what customers kept, they’ll get better at matching the right buyer to the right product.
That’s the result sellers actually need.
Protecting Profits as AI Shopping Grows
Lesonsky: What can merchants do now to reduce the cost of AI-influenced returns?
Henzel: Start with your product listings. AI tools read your descriptions very literally. Be specific about size, weight, materials, and fit, and say plainly who the product isn’t for. A line like “not waterproof” or “runs small” can stop a return before it starts.
Second, track why things come back. Most small sellers process a return and move on. Add a simple reason to each one and review them every month. If the same reason keeps showing up on one product, fix that listing.
Third, catch shipping mistakes before the label. A bad address or the wrong box creates returns and extra charges that have nothing to do with the product.
Last, set your return rules with the full cost in mind. On some low-cost items, refunding and letting the customer keep the product is cheaper than paying for return shipping.
And give customers options besides a refund. An exchange keeps the sale. Store credit keeps the money in your business and brings the customer back to shop again, often for the product that was right for them in the first place. Many shoppers will take the credit if you make it easy, especially if you add a little extra, like a few dollars more than the purchase price. A refund ends the relationship. Credit keeps it going.
Lesonsky: How can post-purchase and fulfillment data improve future AI recommendations?
Henzel: Most AI shopping tools learn from what happens before the sale: searches, clicks, and purchases. The most useful information comes later. Which products came back, why they came back, which ones arrived damaged, and which ones showed up when promised.
That’s the information that tells you whether a recommendation was good. If a product keeps coming back for “too small,” any tool recommending it should know that. If a seller hits its delivery dates every time, that should count in the seller’s favor.
Sellers can act on this now without waiting for the platforms. Put return reasons back into your listings. Keep your delivery promises accurate and published in a form software can read. As AI agents start buying for shoppers, they’ll favor sellers whose costs, delivery dates, and tracking they can trust.
Lesonsky: As commerce becomes more automated, where should merchants keep human oversight in the purchasing and fulfillment process?
Henzel: Let automation do the repetitive work: filling in orders, comparing rates, checking addresses, and answering basic tracking questions. Software is good at that, and it doesn’t get tired at 11 at night.
Keep a person on three things. The first is spending that falls outside your normal pattern, like an overnight replacement or a large refund. The second is any customer who has already had a bad experience. When a return comes from a bad recommendation, a real person can still save the relationship, and an automated reply usually can’t. The third is what you tell the world about your product. Descriptions, sizing, and return policies now feed AI tools, so a mistake there gets repeated thousands of times.
That’s how we built Ship AI. It prepares every order, and nothing gets bought until the seller says yes. I treat any new tool like a new hire. Check its work for a while, and give it more room as it earns it.
My Take
AI may help small online retailers attract more customers, but getting someone to click “buy” is only part of the equation. If a recommendation leads to a return, the merchant can end up paying for shipping, packaging, customer service, and a lost sale.
What struck me most about Henzel’s advice is that small businesses don’t have to wait for AI shopping platforms to improve. They can take steps now to protect themselves.
Start by making product descriptions as accurate and detailed as possible. Track why customers return items, and use that information to improve listings. And don’t judge success solely by how many orders come in. Pay attention to how many customers keep their purchases and how much profit remains after fulfillment and returns.
As AI becomes more involved in shopping decisions, the goal shouldn’t just be more sales. It should be more sales that stay sold.
Rieva Lesonsky is the founder of Small Business Currents, a content company focusing on small businesses and entrepreneurship. You can find her on Twitter @Rieva, Bluesky @Rieva.bsky.social, and LinkedIn. Or email her at Rieva@SmallBusinessCurrents.com.
Photo courtesy Getty Images for Unsplash+