Can AI find the real reasons behind product returns?

How D2C brands can use AI to read return requests, reviews and chats together to find the real causes of returns, and which fixes to test first.

Anadoris Digital · · 4 min read

Question mark
Photo: TeroVesalainen / Wikimedia Commons (CC0)

Returns are expensive for D2C brands: shipping both ways, damaged stock, refunds and time spent on each case. Most brands track return reasons through a dropdown such as “size issue”, “not as described” or “damaged”. These codes are useful for counting, but they rarely tell you what to fix. “Not as described” could mean the colour looked different on screen, the size chart was confusing, or the product photo showed a different variant.

The real reasons usually sit in free text: the comment box on the return form, the WhatsApp chat before the return, the review left afterwards. AI is good at reading that text at scale and turning it into specific, actionable causes.

Gather the right data

  • The return record: product, variant, reason code, comment and date.
  • The original order: channel, payment method, courier and delivery date.
  • Any customer messages about that order, before and after the return.
  • Photos the customer sent, where you have them.
  • Reviews linked to the product in the same period.
  • Exchange requests, which often share the same causes as returns but are easier to miss.

Linking these records is often the hardest part, because they live in different tools. Even a simple sheet joined on order number is enough to start. Remove names, phone numbers and addresses before sending any of this to an external AI tool.

If your return comments are mostly empty, improve the input before the analysis. Make the comment box prominent, ask one specific follow-up question depending on the reason chosen, and invite a photo. Even a few weeks of better input gives far richer material than months of bare reason codes.

Ask specific questions

Rather than asking the AI to summarise all returns, ask it to sort each return into a specific cause, with a short quote as evidence. Give it a starting list of causes and let it propose new ones.

  • Sizing or fit different from expectation.
  • Colour or appearance different from the photos.
  • Texture, smell or taste not as expected.
  • Damaged in transit, or leaking.
  • Wrong item or variant sent.
  • Delivered late, after the customer no longer needed it.
  • Changed mind or ordered by mistake.
  • Suspected misuse or reseller behaviour.

Then ask it to compare causes across products, channels, couriers and time periods. A spike in damaged returns from one courier, or appearance complaints concentrated on one variant, is the kind of pattern that leads to a clear fix.

It is also worth asking what happened before the return. Did the customer message you with a question that went unanswered? Was delivery late? Did they buy during a sale with a steep discount? AI can read the order timeline alongside the return comment and suggest whether the cause was the product, the service or the expectation set by marketing.

Check the patterns before acting

AI can over-read small numbers. Five vivid complaints about a colour can look like a trend when they are not. Before acting, a person should check the quotes behind each cause, see how many returns it really covers, and compare it with the same period last month. Photos are especially useful evidence: a quick look at customer images often confirms or rules out a theory faster than any analysis.

Turning causes into fixes

Most return causes point to a small number of fixes, and many of them are cheap.

  • Fit and sizing: clearer size charts, measurements in the description and comparison notes.
  • Appearance: more accurate photos, including natural light and in-hand shots of each variant.
  • Texture or smell: honest descriptions that set expectations, plus usage tips.
  • Damage: better packaging for fragile items, or a conversation with the courier.
  • Wrong item: a packing check step or barcode scanning before dispatch.
  • Late delivery: realistic delivery estimates on the product page.

Test one fix at a time where you can, and watch the matching cause over the following weeks. If the cause does not fall, the analysis may have been wrong, or the fix not strong enough.

Use returns to improve the product page

Many returns are prevented on the product page rather than in the warehouse. Once you know the main causes, AI can help draft clearer size guidance, honest texture or shade descriptions, and FAQ entries that answer the questions returners wished they had asked. Review each change for accuracy, publish it, and watch whether the matching return cause falls over the following weeks. Small wording changes can make a real difference when they address the exact misunderstanding customers describe.

Where to start

Take the last three months of returns for your most-returned product. Join them to their orders and any customer messages, remove personal details, and run a cause analysis with quotes. Pick the largest specific cause and make one fix. Review again in a month. Anadoris Digital can help link returns, orders and messages so this analysis runs regularly instead of as a one-off.

Related service: AI & Automation · More on this topic: AI & Automation articles

Common questions

Should I change my return reason dropdown?

Often, yes. Keep it short but more specific, and always include a free-text box. Ask customers to add a photo for damage and appearance issues. Better input makes both manual and AI analysis more reliable.

Can AI predict which orders are likely to be returned?

With enough history, patterns such as certain products, sizes, regions or payment methods can indicate higher return risk. Use any such signal to trigger helpful steps, like a size check message, rather than to refuse or penalise customers.

How do I handle return fraud with AI?

AI can flag unusual patterns, such as frequent returns from one address or mismatched items returned, for a person to review. Treat flags as reasons to look closer, not as proof, and apply your return policy consistently and fairly.

Your next chapter

Let’s build what comes next.

A new website. A better system. A simpler way to work. Tell us where you want to begin.

Discuss your project