When Agents Match Customers Better, Returns Drop
Returns are expensive. For the average online retailer, processing a return costs between £10 and £20 in logistics alone, before accounting for restocking, potential markdowns, and the customer service time involved. In fashion, return rates regularly exceed 30%.
A quiet shift is changing those numbers. Purchases made through AI agent recommendations are showing measurably lower return rates compared to purchases from other channels. The reason is straightforward: agents are better at matching products to stated needs.
Why Agent-Recommended Purchases Stick
When a human shops online, they browse, get distracted, make impulse decisions, and sometimes misread specifications. A customer searching for a laptop might buy one with impressive specs without noticing it weighs 2.8kg, making it impractical for the commute they planned to carry it on.
An agent operates differently. A user says "find me a laptop under £800 that weighs less than 1.5kg and has at least 16GB RAM." The agent checks those exact criteria against every option it can access. It doesn't recommend a 2.8kg laptop, no matter how good the reviews are.
This precision filtering reduces the most common cause of returns: the product didn't match what the customer actually needed.
The Numbers So Far
Early data from retailers tracking agent-referred purchases separately shows return rates 15-25% lower than their site average. The effect is strongest in categories with measurable specifications: electronics, appliances, sporting equipment, and tools.
In fashion, where fit and subjective preference drive returns, the improvement is smaller but still present. Agents that cross-reference size guides, fabric descriptions, and user-provided measurements can reduce "didn't fit" returns, though "didn't like the colour in person" remains an unsolvable problem for any online channel.
One UK electronics retailer that restructured their product data for agent readability reported a 19% reduction in return rates on agent-referred orders over a four-month period. Their average order value was slightly lower, which suggests agents were matching customers to products they actually needed rather than upselling them to pricier options. But net margin per order was higher because the return cost vanished.
The Margin Impact
The margin improvement from lower returns compounds in ways that aren't immediately obvious.
Direct logistics savings. Every return avoided saves the shipping, processing, and restocking cost. At scale, this adds up quickly.
Reduced markdown pressure. Returned items often can't be resold at full price. Some are repackaged, some are sold as open-box, and some are written off. Lower return rates mean more units sell at full margin.
Better inventory planning. High return rates create uncertainty in demand forecasting. If 30% of your sales come back, you need to over-order to maintain stock levels. Lower returns mean more accurate forecasting and less capital tied up in safety stock.
Customer lifetime value. A customer whose first purchase meets their needs is more likely to return. A customer who goes through the hassle of a return is more likely to switch to a competitor next time. Agent-matched purchases build a foundation for repeat business.
What This Means for Product Data
The catch is that agents can only match products precisely if your product data is precise. Vague descriptions, missing specifications, and inconsistent sizing information all degrade the agent's ability to make good recommendations.
If your laptop listing says "lightweight" but doesn't state the weight in grams or kilograms, the agent can't filter on weight. If your clothing sizes don't include specific measurements, the agent can't cross-reference them against the user's dimensions.
The retailers seeing the biggest return rate improvements are the ones with the most complete product data. They've invested in:
- Exact measurements for every relevant dimension
- Standardised specification fields across product categories
- Accurate and current stock information
- Honest descriptions of limitations, not just features
This aligns with broader agent readiness work. If you want agents to recommend your products accurately, you need to structure your HTML so agents can parse it correctly.
Sizing the Opportunity
Calculate this for your own business. Take your current return rate, your average return processing cost, and your monthly order volume. If agent-referred orders could reduce returns by even 15%, what does that save annually?
For a retailer processing 10,000 orders per month with a 25% return rate and a £15 cost per return, a 15% reduction in return rate on agent-referred orders (assuming those make up 10% of total orders) saves roughly £27,000 per year. That's a conservative estimate that doesn't include markdown savings or customer retention effects.
As agent traffic grows, so does the savings. The businesses that make their product data agent-readable now will benefit from this compounding effect earliest. Running an agent readiness audit is the fastest way to identify where your product data falls short and what to fix first.