How product recommendations affect both AOV and return rates

Product recommendation implementation increased AOV by 18%. Customers added suggested items, building larger carts. But return rate also increased from 9% to 13%. Some recommended products didn’t match customer needs—they were added impulsively based on algorithmic suggestion rather than genuine fit. The AOV increase came with a return rate increase that partially offset the benefit.

Recommendations affect both what customers buy and whether they keep what they bought. Understanding this dual impact helps you design recommendation systems that maximize retained revenue rather than just initial transaction size.

How recommendations increase AOV

Suggestions drive larger purchases:

Cross-sell exposure adds items

“Frequently bought together” and “customers also purchased” expose products customers might not have found. Items they didn’t know they wanted get added to carts. Exposure creates purchases that wouldn’t otherwise happen.

Upsell suggestions trade up purchases

“Consider the premium version” encourages upgrading. Customers who might have bought the basic option see and choose the better option. Higher-value products replace lower-value selections.

Convenience reduces shopping friction

Recommendations bring relevant products to customers rather than requiring search. Easy discovery of complementary items makes adding them natural. Convenience enables purchases that shopping friction would have prevented.

Social proof through purchase data

“Others bought this too” provides social validation. Customers feel confident adding items that similar customers purchased. Social proof converts recommendation views into additions.

Completeness mentality

Showing what goes with a product triggers completeness desire. Customers want the full solution, not just one piece. Recommendations frame additional items as necessary complements.

How recommendations increase returns

The same suggestions can backfire:

Impulse additions are weakly considered

Items added from recommendations often receive less evaluation than items customers sought deliberately. Quick “add to cart” on recommendation doesn’t involve the consideration that direct search does. Less-considered purchases are more likely to be regretted and returned.

Algorithmic suggestions can miss actual needs

Algorithms optimize for conversion, not for customer fit. Products that similar customers bought might not match this specific customer’s needs. Good aggregate performance masks individual mismatches that become returns.

Urgency and scarcity in recommendations

“Only 2 left” or “selling fast” on recommended items creates purchase pressure. Customers add items to avoid missing out rather than because they truly need them. Pressure-driven purchases have higher return rates.

Add-ons don’t fit primary purchase

Accessories recommended for product A might not work with the specific variant customer chose. Size mismatches, compatibility issues, or style conflicts create returns when recommendations aren’t context-aware.

Discovery creates exploration purchases

Recommendations introduce customers to products they wouldn’t have sought. Some exploration purchases work out; others don’t. Exploration has inherently higher return risk than targeted shopping.

Calculating net recommendation value

Measure true impact:

Incremental AOV versus incremental returns

18% AOV increase sounds great. But with return rate rising from 9% to 13%:

Before recommendations:

$100 AOV × 91% kept = $91 retained revenue per order

After recommendations:

$118 AOV × 87% kept = $102.66 retained revenue per order

Net improvement: 13% in retained revenue, not 18%. Returns consumed about 30% of the AOV gain.

Include return processing costs

Each return costs money to process. If incremental returns cost $12 each to process and recommendations add 4 returns per 100 orders:

Additional return cost: 4 × $12 = $48 per 100 orders = $0.48 per order

This further reduces net recommendation value.

Track return rate by recommendation source

Are returns concentrated in recommended items? Track whether items customers found themselves versus items added from recommendations have different return rates. This identifies problematic recommendation types.

Optimizing recommendations for retained value

Design for keeping, not just buying:

Recommend compatible products specifically

Generic “you might also like” has higher return risk than specific “this accessory fits your selected item.” Context-aware recommendations reduce compatibility returns.

Show information that prevents regret

Include enough detail in recommendation display for informed decisions. Size, key specs, and use case information help customers evaluate fit before adding. Informed additions are kept more often.

Avoid high-pressure recommendation tactics

Urgency and scarcity drive immediate additions but increase regret. Recommendations that let customers consider rather than pressure produce better retention.

Weight algorithms for retention, not just conversion

If you can track which recommended items get returned, use that data to improve algorithms. Optimize for purchased-and-kept rather than just purchased.

Personalize based on return history

Customers with high return rates might need different recommendation strategies. Fewer recommendations, more conservative suggestions, or emphasis on fit information might reduce their returns.

Types of recommendations and their return profiles

Different recommendation types have different risks:

Complementary accessories: Lower return risk

Items designed to work with primary purchase usually fit. Cases for specific phones, batteries for specific devices—these have clear compatibility and lower return rates.

Similar alternatives: Moderate return risk

“You might also like” shows similar products. Customers might add these exploratorily without strong intent. Return rate depends on how well similarity matching works.

Trending or popular items: Higher return risk

What’s popular isn’t necessarily right for this customer. Trend-based recommendations attract impulse additions with limited fit evaluation. Higher exploration, higher returns.

Recently viewed: Variable return risk

Items the customer already viewed and didn’t buy might have been rejected for reasons. Re-recommending might overcome initial hesitation or might add items customer already decided against. Mixed outcomes.

Measuring recommendation performance properly

Track the right metrics:

AOV lift from recommendations: Compare AOV between customers who engaged recommendations versus those who didn’t.

Return rate by recommendation engagement: Do customers who add recommended items return at higher rates?

Return rate by recommendation type: Which recommendation algorithms or placements produce higher return rates?

Retained revenue per order: AOV × (1 - return rate) captures true value better than AOV alone.

Item-level return rate: Do specific recommended items have unusually high returns? Identify and improve or remove problematic recommendations.

Frequently asked questions

Should I turn off recommendations if they increase returns?

Only if net value is negative. If AOV lift exceeds return cost increase, recommendations still help. Optimize rather than eliminate.

How do I know if recommendations cause returns?

Track return rate on recommended items versus customer-found items. If recommended items return at higher rates, recommendations contribute to return increase.

What return rate increase is acceptable?

Depends on margin and AOV lift. Calculate net retained revenue change. If net revenue increases despite higher returns, the trade-off works.

Can I show recommendations without increasing returns?

Yes, with careful design. Context-aware, well-explained recommendations that help customers make informed decisions can increase AOV without proportional return increase.