Fashion Intelligence Series | Module 4 of 6

Returns analysis:
Why jeans come back - the real reasons

How NLP analysis of 210,000 free-text comments uncovers the 5 true root causes behind "doesn't fit" - and which 3 measures cut the return rate from 28% to around 26% - together with the size guidance from module 3 to around 23%.

Model scenario - fictional fashion company

This module is part of a six-part model calculation about a fictional mid-sized fashion company (1,800 SKUs, 18 markets). All figures, datasets, model metrics and euro amounts are entirely fictional - they illustrate the methodology and the kind of effect achievable. They are not results of real clients and not a performance promise. What values are achievable on your real data is established only in a pilot project.

01 The problem - €26 per return, 537,600 times a year

Why "does not fit" as a return reason is not sufficient

With 1.92 million units sold online (60% online share) and a return rate of 28% (a blend across 18 markets - the German market is higher), that amounts to 537,600 returns per year. Each costs on average €26 (€19.51 process cost per shipment, University of Bamberg returns research, plus a share of the value loss - assumption). Official statistics show "does not fit" at 42% - but the NLP model reveals: behind that single label lie 5 distinct problems with 5 distinct solutions.

02 Model - NLP root-cause analysis on 210,000 free-text comments

5
Root-cause clusters
€0,81M
Savings potential / year
28% → 26%
Return rate
2 wks
Fastest quick win
Root causes: official dropdown selection vs. NLP analysis
↳ The 5 clusters and their solutions

The total potential is around €1.4M - shown here: €0.81M from the two most quickly implementable measures: product-photography and colour-fidelity improvements plus more honest product descriptions. Derivation: 31,181 avoided returns × €26, set at a 58% non-size share (assumption; deliberately below the cluster 2-5 sum of 69%), a relative reduction of -10%. Cluster 1 (31%): Incorrect size run → Fix: size recommender (Module 3) - accounted for there, not counted here. Cluster 2 (22%): Colour/material differs from expectation → Fix: better product photography + fabric close-ups. Cluster 3 (19%): Quality defect → Fix: tighter QC requirements with suppliers. Cluster 4 (16%): Expectation vs. reality → Fix: more honest product descriptions + customer reviews prominently placed. Cluster 5 (12%): Serial returner → Fix: checkout prompts + order limits.

03 Next steps for your brand

① NLP analysis of your returns

All free-text comments from the past 2 years through the topic model. Output: root-cause map + top 20 problem SKUs.

② Implement quick wins

Improve product photography, sharpen descriptions, review integration. Combined: €0.81M/year.

③ Prediction API

Return probability per order in real time. Targeted interventions for high-risk orders.

All 6 modules: AI in fashion