Too Tight, Too Long, or Not as Expected? AI Can Organise Fashion Returns

“Didn’t fit” can hide several different problems. The collar may feel tight. The sleeve may be long. A parent may have expected a roomy boys’ kurta and received a closer fit. Or the garment may fit perfectly, but the occasion changed.

When every comment is placed in one broad “fit” bucket, the business loses the reason behind the return. AI can help organise that language, but the useful outcome is not a dashboard full of sentiment. It is a shorter list of specific issues that a person can investigate.

Start with a return-reason map

Before applying AI, define a small, practical taxonomy. For fashion, it might include:

• too small at a named point: collar, shoulder, chest, waist or armhole;
• too large or too long;
• colour or pattern differed from expectation;
• fabric or hand-feel differed from expectation;
• damage or construction concern;
• ordered more than one size;
• occasion, delivery or personal reason unrelated to the product.

Shopify’s Admin API defines standard return reasons such as size too small, size too large, style and unwanted, while also supporting an “other” category. That is a useful starting structure, not a substitute for the customer’s exact words. See Shopify’s ReturnReason reference.

Where AI adds practical value

Group similar phrases

“Neck is snug,” “collar pinches” and “top button is tight” can be grouped for review without forcing customers to use identical wording. Clustering is designed to group related examples, but its quality depends on the similarity measure and the data. Google’s clustering guide highlights the importance of choosing a similarity measure and evaluating the clusters rather than accepting them blindly.

Attach the issue to the correct product detail

A complaint about sleeve length should be linked to the size and product involved. A comment about a set should not be assigned automatically to the kurta if it actually concerns the pyjama. Preserve SKU, variant, order date and original comment alongside the AI label.

Summarise patterns for action

The useful weekly note is not “negative fit sentiment increased.” It is “several size-20 comments mention tight collars; check the measurement chart and inspect the current batch.” That statement still needs human verification before a page or pattern is changed.

Boys’ and men’s feedback cannot share one model casually

For a boys’ item such as the TryBuy Boys Black Floral Embroidered Printed Kurta Pajama Set, parents may discuss age, height, growth allowance and ease of dressing. Men’s shirt feedback is more likely to reference chest, shoulder, sleeve or collar measurements. Combining both without category context can create misleading clusters.

Keep separate labels where the buying decision differs. “Room to grow” belongs in children’s apparel analysis; it should not become a default goal for a fitted men’s shirt.

Use a human review queue

AI should flag patterns, not decide that a garment is defective. A practical queue can prioritise clusters that are new, growing, tied to a particular variant or based on clear construction language. Reviewers should see examples, not only the summary.

Also watch for false certainty. A small number of comments can look dramatic, and a return spike can reflect a promotion, delivery delay or duplicate-size ordering. Compare comments with units sold, exchanges and inspection records before acting.

Turn analysis into better information

Once verified, return themes can improve size-chart notes, product photography, detail copy, packaging checks or manufacturing inspection. The change should be specific: add a collar measurement, show the garment length, clarify whether a boys’ product is a set, or photograph a texture close-up.

Browse TRYBUY.IN boys’ kurtas and review each product’s stated size and set contents before choosing.

FAQ

Can AI decide whether a customer’s return is valid?

It should not. Policy and customer-service decisions require the applicable rules and human oversight.

Does sentiment analysis explain why a garment was returned?

Not reliably. A specific reason taxonomy is more actionable than a positive-or-negative score.

Should the original comment be deleted after classification?

No. Keep it with the label so reviewers can audit and correct the interpretation.