Skip to content
Documentation
All docs

Analytics Reference

Return Reason Clusters

How free-text return reasons are grouped into actionable categories.

Reason clusters take the messy, free-text reason a customer typed into their return request and group it into one of a handful of structured categories. This is what turns “dosent fit lol” and “Size too small for my shoulders” into the same actionable bucket: SIZING_FIT.

The clusters

RefundSentry uses a fixed taxonomy of reason clusters so you can compare like-for-like over time and across products:

  • SIZING_FIT — Wrong size, doesn’t fit, runs small/large.
  • QUALITY — Defective, damaged, broken, poor materials.
  • EXPECTATIONS — Looks different from the photo, color is off, not what I expected.
  • CHANGED_MIND — Don’t need it anymore, ordered by mistake, no reason given.
  • WRONG_ITEM — Received the wrong product or wrong variant.
  • LATE_DELIVERY — Arrived too late to be useful.
  • OTHER — Anything that doesn’t fit the above.

How clustering works

For each return, RefundSentry takes the customer’s reason text and runs it through a classifier that maps the free-text input to one of the fixed clusters. The classifier is multilingual — you can write “trop petit” or “zu klein” and both land in SIZING_FIT.

Returns with no customer reason at all (Shopify lets the merchant skip this) get bucketed as OTHER. If your store has a high OTHER share, your return form is probably not asking for a reason — that’s a setup issue, not a clustering issue.

Why this matters

Most stores discover one or two cluster patterns once they look at the data:

  • SIZING_FIT > 35% — Your size guides are wrong, your photos don’t show fit, or you’re carrying too many size variants per SKU. See Sizing-related returns.
  • QUALITY > 15% — A supplier or QC issue concentrated in specific SKUs. Cross-reference with the Reason × product heatmap to find the culprits.
  • EXPECTATIONS > 20% — Your product photography or copy oversells. Common in fast-fashion and dropshipping.
  • CHANGED_MIND > 25% — Often legitimate. But if it’s very high and concentrated on a few customers, it’s a wardrobing signal — investigate the customers, not the products.

Trend chart

The Reasons tab shows cluster shares over time. Sudden spikes are the most actionable insight here — a QUALITY spike on a specific week usually traces to a single bad batch from a supplier.

Next steps