Bracket by Proxy: When Shoppers Order Sizes for Someone Else and Your Fit Data Lies

The order that is not what it looks like

Your bracketing report flags a customer who ordered the same dress in a small and an extra large. That is a four-size spread, the kind of gap that screams bracketing. Except this customer kept both. The small was for her; the extra large was for her mother. Neither size was a guess. Both were confident purchases. Your dashboard counted a bracket. The customer experienced a perfectly normal shopping trip.

Bracket by proxy is the purchase of sizes for someone other than the buyer: partners, kids, parents, friends. It shows up in multi-size order reports constantly, and it is not bracketing in any sense that matters. The shopper is not uncertain about fit. They are shopping for two bodies. Treat it like bracketing and you will misread your best gift-buyers as your worst offenders.

Why proxy orders poison fit data

The damage is not just a mislabeled order. Proxy purchases corrupt the fit signals you use to tune everything else. Size recommendation engines learn from purchase and return pairs: kept means it fit, returned means it did not. A proxy order breaks that logic. The buyer kept both sizes because both fit their intended wearers, so the model learns that this customer wears both a small and an extra large. Future recommendations for that account get noisier, and if the data aggregates into population-level fit models, it pollutes those too.

Return behavior gets distorted in the other direction. Proxy gifts are returned at different rates than self-purchases, and for different reasons. A gift recipient exchanges for fit far more often, and the return reason codes mix gift exchanges with genuine fit failures. If you feed that stream into your bracketing model without separating it, your fit-failure estimates drift upward and your interventions target the wrong customers.

How to spot proxy orders

Proxy orders have signatures, and they are different from bracketing signatures. The strongest is the size spread: self-bracketing clusters in adjacent sizes, because the shopper is uncertain between neighbors. Proxy orders jump sizes, because two different bodies are involved. A small plus an extra large, or a medium plus a 2X, is proxy-shaped. Adjacent sizes are bracket-shaped.

The second signal is in the return pattern. Bracketers return all but one and keep the winner. Proxy buyers keep everything, or return one piece with a gift-exchange reason. The third signal is timing and assortment: proxy orders spike around holidays and life events, include mixed categories or mixed gendered items that one body would never wear together, and often ship with gift messaging or gift wrap.

None of these is conclusive alone. Together, they separate proxy orders from bracketing with useful accuracy. A simple classifier on size spread, keep behavior, and category mix will catch most of them.

What to do with them

First, stop counting proxy orders in your bracket rate. Tag them at the order level and exclude them from bracketing metrics. Your real bracket rate drops, which is good: it was never real. More importantly, your fit models get cleaner, because you stop training on purchases that were never about the buyer's body.

Second, treat proxy buyers as a segment, not a problem. They are high-intent shoppers buying for multiple people, and they convert at strong rates. The right intervention is not a size nudge, it is gift tooling: easy recipient profiles, size memory per recipient, and gift-friendly return flows. Brands that build this find that proxy buyers are among their most valuable customers, precisely because they were never bracketing at all.

The deeper lesson is about what your data actually measures. A multi-size order is not a behavior, it is a pattern, and patterns need interpretation before they become metrics. Proxy ordering is one of the most common alternative explanations, and it is hiding in almost every bracketing report. Find it, label it, and your numbers will finally describe what your customers are actually doing.