Cross-sell related products that actually fit
Relevant cross-sells lift order value; random ones add clutter. Base them on real behavior.
In short
- Recommendation clicks are ~7% of visits but ~26% of revenue, and relevance is where that gap lives.
- Tag-based widgets show substitutes; real cross-sells show complements pulled from actual co-purchase data.
- Put a price, image, and review count on each add-on. A bare thumbnail is a cold ask.
Random 'related products' that shoppers ignore.
Illustrative. Real lift is measured on your traffic first.
A cross-sell block earns its place on the page or it's just visual noise the shopper scrolls past. The recommendation slots that actually pull weight are a small slice of all the action and a fat slice of the money. Salesforce found visits where someone clicks a product recommendation are only 7% of visits but drive 26% of revenue. The difference between those two numbers is relevance, and relevance comes from what…
What's the problem?
Your 'related products' widget shows items that don't really go together, so shoppers ignore it and your average order value stays flat.
Why does this happen?
- Recommendations are based on tags or randomness, not real co-purchase patterns.
- Cross-sells appear where they distract from, rather than complement, the buy decision.
- There's no measurement of whether the widget adds revenue.
- The block is the same on every product page. A 'related products' module pulling from the same collection means the carry strap, the cleaning kit, and the spare part all show the exact same eight items, so it reads as…
- It recommends substitutes, not complements. Tag- and collection-based widgets surface more of the thing you're already looking at: three other yoga mats under a mat. That's a comparison shopper's exit ramp, not an add-…
- The add-on has no proof of its own. A $14 carry strap shown bare next to a $90 mat is a cold ask. Buyers de-risk small companion purchases the same way they de-risk big ones, with a star count. Spiegel's research puts…
- Take rate gets mistaken for value. A widget that 'converts' can still be cannibalizing, moving a sale that was already coming, or pulling attention off the main buy. Without revenue-per-visitor on the page version vs.…
What does the research show?
Independent researchFigures below are from independent studies, not StorePilot data. They're why this problem is worth testing on your own store.
-
Visits where a shopper clicks a product recommendation are just 7% of all visits but drive 24% of orders and 26% of revenue, across 150M+ shoppers and 250M+ visits.
Salesforce (Commerce Cloud), 'Personalized Product Recommendations Drive Just 7% of Visits but 26% of Revenue' ↗ -
35% of what consumers buy on Amazon comes from algorithm-driven product recommendations.
McKinsey & Company, 'How retailers can keep up with consumers' ↗ -
A product showing five reviews has 270% greater purchase likelihood than the same product with none, so attaching review counts to your cross-sell items matters.
Spiegel Research Center, Northwestern University ↗ -
Personalization typically lifts revenue 10 to 15 percent, with company-level results ranging 5 to 25 percent depending on execution.
McKinsey & Company ↗ -
Only about 1 in 7 (roughly 14%) of A/B tests produces a meaningful winner, which is why a cross-sell change needs to be measured rather than assumed to work.
VWO ↗
How does StorePilot AI fix it?
- StorePilot uses real co-view and co-purchase signals to suggest genuinely complementary products.
- It tests placement and framing of the cross-sell so it adds order value without hurting core conversion.
- Revenue-per-visitor measurement confirms the cross-sell is incremental, not cannibalizing.
How do you fix it, step by step?
-
Pull your real co-purchase pairs first
Before touching the widget, look at what actually shares a cart or gets bought back-to-back over the last 60-90 days. The pairs that show up repeatedly (mat and strap, grinder and filters, frame and mounting kit) are your real cross-sells. Tags and 'same collection' are a guess; order history is the answer.
-
Recommend the next thing, not another of the same thing
Swap any logic that surfaces more items from the current product's collection for logic that surfaces the complement. If a yoga-mat page is showing three other yoga mats, it's helping shoppers comparison-shop their way out. Show the strap, the towel, the cleaner, the stuff that rides along.
-
Reframe and reposition the block as 'Pairs well with'
Drop the generic 'Related products' carousel and use a small, named block, like 'Pairs well with' or 'Complete the set', placed near the buy box or in the cart, not buried below the fold where it reads as filler. Two or three specific items beat a scrolling rail of eight.
-
Put a price, an image, and a review count on each item
A bare thumbnail is a weak ask. Show the companion's price (so the math is obvious), a clear product image, and its star rating where it has one. The review signal does real work on low-consideration add-ons. One-tap add-to-cart removes the last bit of friction.
-
Test it on revenue per visitor, not take rate
Run the new block against the old widget (or no widget) and judge it on revenue per visitor for the whole page, plus core conversion as a guardrail. A cross-sell that lifts attach rate but dents the primary sale, or just moves a purchase that was already coming, isn't a win.
-
Keep what proves out, cut what doesn't
Most pairings won't beat the original, so expect to kill some. Promote the cross-sells that show a real revenue-per-visitor lift to more pages, retire the ones that flatline, and re-pull co-purchase data periodically as your catalog and buying patterns shift.
An illustrative example
Demo data- What StorePilot detects
- Buyers of the yoga mat frequently buy the carry strap together, but the related-products widget shows unrelated apparel.
- The fix it builds & tests
- Replace the widget with a 'Pairs well with' block featuring the carry strap on the mat PDP.
- The projected outcome
- Example projection: higher attach rate and order value. (Illustrative demo figure.)
Key takeaways
- Recommendation clicks are ~7% of visits but ~26% of revenue, and relevance is where that gap lives.
- Tag-based widgets show substitutes; real cross-sells show complements pulled from actual co-purchase data.
- Put a price, image, and review count on each add-on. A bare thumbnail is a cold ask.
- Judge the block on revenue per visitor, not take rate, so you don't reward cannibalization.
This guide is part of the StorePilot average order value playbook. If this is costing you sales, look at Increase average order value with bundles and Add a post-purchase upsell that shoppers welcome next.
