We do not collect structured return reasons today. Where do we start?

Start at the point of return, not with a data project. Replace the free-text reason field with a short structured flow: fit or not fit, then too small or too large and where, then the exchange size if any. You will have usable data within one return cycle, and the historical unstructured reasons were not going to help anyway.

Does this work for new products with no return history?

Partially. A new product inherits the fit profile of its silhouette family: similar garments from the same pattern block carry similar keep-rate curves until the new product earns its own data. Flag the recommendation as based on similar styles, then let the product's own returns take over as volume builds.

NimbleTailor answers

Returns data as a fit signal: using what came back to recommend what fits

Your returns are the most honest fit data you own: each one records a garment, a size, a body, and a verdict. Stores that structure return reasons, track size exchanges, and compute keep rates by size curve can recommend sizes that measurably cut the next return. The data is already in your warehouse; the work is making it structured enough to learn from.

Why returns beat every other fit signal

Size charts describe garments. Reviews describe opinions. Returns describe outcomes: this size on this customer did not work, and here is the replacement size that did. A size exchange is a labeled training example with the ground truth attached, and most stores process thousands of them a year without ever feeding them back into recommendations.

The signal is also self-correcting. Fit preferences drift with trends, seasons, and silhouettes, but return patterns drift with them automatically. A recommendation model trained on last year's exchanges is already adapting to this year's fit reality, with no manual recalibration required.

Structuring return reasons so they are usable

The reason most returns data is useless is the reason field itself: a free-text box or a dropdown with one option called 'does not fit.' Split fit returns into the dimensions that matter: too small where, too large where, length wrong, and the exchange size chosen. Four structured fields turn a return from a cost center into a data asset.

Also capture the direction of the exchange, not just the return. A customer who returns a medium for a large is telling you the garment runs small; a customer who returns a medium for a small is telling you the opposite. The exchange pair is the atomic unit of fit intelligence, and it only exists if your returns flow records both sides.

Keep rates by size curve

For each product, compute the keep rate per size: the share of units sold in that size that were never returned for fit reasons. A size with a 92 percent keep rate is a safe recommendation; a size with a 64 percent keep rate is a warning sign that the grading or the size chart is off for that specific garment.

The curve matters more than any single number. If keep rates dip at both ends of the size range, the garment's grading is the problem, not the customers. If one size dips while its neighbors hold, that size is likely mislabeled or cut inconsistently. Keep-rate curves turn vague fit complaints into specific manufacturing feedback.

Turning the data into recommendations

The simplest version needs no machine learning: when a shopper views a product, show the keep rate by size and nudge toward the size with the best outcome for customers like them. 'Customers with your purchase history kept the large 89 percent of the time' is more persuasive than any size chart, because it is evidence, not instruction.

The stronger version matches the shopper to exchange patterns. If customers who bought and kept size M in three similar garments all exchanged to L in this silhouette, recommend L with the reason attached. Shoppers trust recommendations that cite other shoppers' outcomes far more than recommendations that cite a brand's own chart.

Closing the loop with the product team

Fit data should flow upstream, not just into recommendations. A monthly report of the lowest keep-rate sizes by product gives designers and technical teams a prioritized fix list: regrade this, relabel that, adjust the size chart copy here. Every fix compounds, because better-cut garments generate better data for the next round of recommendations.

Measure the loop the way finance measures everything: track return rate for shoppers who saw a data-driven size recommendation versus those who did not. The gap is the ROI of the whole program, and it is the number that justifies investing in the returns-flow changes that feed it.