AI in Footwear: Sizing, Returns and Production Planning

Footwear has the highest return rate in apparel retail, and almost all of it is fit. Customers order two sizes because they cannot predict how a style will fit, and the economics of that behaviour dominate everything else in the category. This guide covers where AI helps, drawing on FISTA Solutions' AI agents work in retail. It complements the retail and commerce operations whitepaper and how to build a returns processing agent. This article is general guidance, not legal advice.

Why are returns the central problem?

Because each one costs outbound shipping, return shipping, processing, inspection, and frequently the entire margin on the sale — and at footwear return rates, that applies to a large share of orders.

Reducing fit-related returns is therefore the largest economic lever in online footwear, larger than acquisition efficiency or merchandising improvements, and it is addressable because the cause is knowable.

Lever	Effect on returns	Difficulty
Fit guidance from purchase history	Substantial	Moderate
Style-level fit notes from reviews	Moderate	Low
Size-curve accuracy in buying	Indirect but real	Moderate
Honest uncertainty signalling	Positive	Low
Multi-size ordering incentives	Negative	—
Better product photography and detail	Moderate	Low

Why is sizing inconsistent?

Because fit depends on the last, not the nominal size. Two styles from the same brand in the same stated size fit differently because they are built on different lasts with different volumes and widths.

Customers know this, which is why they order multiple sizes. No amount of size chart improvement fixes it, because the chart describes a nominal measurement rather than how the shoe fits a foot.

What is the best fit signal?

The customer's own history. What they bought, what they kept, what they returned, and why, across styles and brands, is far more predictive than self-reported measurements — which are inconsistently taken and frequently aspirational.

For a returning customer, that history plus the style's fit characteristics relative to styles they have already judged is a genuinely useful basis for a recommendation. For a new customer it is weaker, and the system should say so.

Why does uncertainty need stating?

Because a confident wrong recommendation produces the return it was supposed to prevent, plus a loss of trust in the guidance. A customer who follows a size recommendation and has to return the shoe will not follow the next one.

Saying that a style runs small, that sizing on this last is less predictable, or that customers with similar history split between two sizes is more useful and more honest than a single definitive number. See what is abstention in ai.

What is size-curve forecasting?

Forecasting demand by size within a style rather than for the style in aggregate. Buying the right total units with the wrong curve means selling out of the middle sizes early and marking down the extremes at season end.

That markdown is a recurring, quantifiable loss, and the curve varies by style, channel, and market in patterns that historical data supports.

What about production and materials?

Lead times are long and materials are committed early, which makes the initial forecast consequential. Improving it, and improving the ability to react within the season through better early-signal detection, both reduce the terminal markdown that defines footwear profitability.

Who should own it?

Merchandising and planning, with digital owning the customer-facing guidance. Fit guidance is a customer experience decision with merchandising consequences, and splitting ownership tends to produce guidance that is either commercially naive or unhelpfully cautious.

How is it evaluated?

Return rate by reason code, return rate among customers who used the guidance versus those who did not, sell-through by size, terminal markdown, and guidance acceptance rate. Recommendations served measures usage, not effect.

What goes wrong?

Fit guidance based on self-reported measurements. Confident recommendations where the data does not support them. Size charts improved instead of fit predicted. Forecasting at style level while markdowns happen at size level. And measuring guidance usage rather than return outcomes.

What does it cost to run?

Low per interaction. The value is in returns avoided, which is measured in shipping and margin rather than in operational cost, so the business case should be built on return rate rather than on efficiency.

What should you do first?

Break your returns down by stated reason and by size relative to what the customer ordered. The proportion that are fit-related, and whether customers size up or down, is usually clearer than expected and points directly at which styles need attention.

What about in-store?

The same fit knowledge serves store colleagues, who currently rely on personal experience of how styles run. Making style-level fit characteristics available at the point of a conversation improves conversion and reduces the returns that follow a store purchase, which are often excluded from the returns analysis entirely.

How FISTA Solutions helps

FISTA Solutions builds footwear systems with fit guidance grounded in purchase and return history rather than self-reported measurement, explicit uncertainty where the signal is weak, size-curve level demand forecasting, and returns analysis by reason, through AI agents, AI enablement, and web and mobile engineering. The record behind the approach is 150+ projects for 50+ companies with 47% efficiency gains.

To reduce fit returns rather than merely serve recommendations, message FISTA on WhatsApp, or read the retail and commerce operations whitepaper.