CASE STUDY · 2025 · DTC
8 WEEKS · DATA SCIENCE CONSULTING

A size model that cut fit-based returns.

Most of the returns weren't about taste, they were about fit. I built a size-recommendation model from purchase-and-return history that nudged each shopper toward the size they'd actually keep, cutting fit-based returns without hurting conversion.
FIT-BASED RETURNS
−22%

On covered styles

TRAINED ON RETURN HISTORY
MODEL

Purchases, sizes, kept vs sent back

CONVERSION LOST
0

Guidance, not a barrier

Figures are approximate and rounded to protect client confidentiality; exact numbers are withheld.

The problem.

A large share of returns came down to fit, not preference, shoppers guessing between sizes and sending back the miss. Every one of those returns carried shipping, processing, and lost-margin cost, and the site gave no guidance to prevent them.

The approach.

I built a size-recommendation model trained on purchase-and-return history, which sizes each style ran to, and which orders came back as fit returns. The model translated that into a simple, per-style recommendation surfaced at the point of choice, so the guidance helped rather than blocked the purchase.

The outcome.

Fit-based returns on covered styles fell 22% with no measurable hit to conversion, the shopper got to the right size before buying, not after. A data-science answer to a problem the industry usually accepts as unavoidable.

+
ARTIFACTS

From the work.

Returns that are really a fit problem?

If most of your returns are size misses, there's a model for that, trained on your own history. One slot open for Q3 2026.

BOOK A 20-MIN CALL