Search "do product recommendations increase sales" and the same handful of numbers turn up across dozens of blog posts, usually with no link back to an actual report. Some of those numbers are real, sourced, and still checkable today. Most have been stretched, mislabeled, or separated from the study that produced them somewhere along the way. Before deciding what to test on a Shopify store, it's worth knowing which is which.
Why most recommendation stats don't survive a Google search
The pattern is almost always the same: a number appears in a blog post with no citation, gets copied into the next blog post with the same missing citation, and after a few years reads like established fact. Two failure modes account for nearly every bad stat in this space.
The first is segment collapse: a lift measured only among shoppers who already clicked a recommendation gets rewritten as if it applied to every visitor on the site. The second is source loss: a number gets attributed to a company or "a study" with no link, no sample size, and no year attached — at that point there's no way to check whether it still holds, or ever did.
The "Amazon gets 35% of revenue from recommendations" case
The clearest example of both problems together is the claim that Amazon generates 35% of its revenue from product recommendations. It shows up in pitch decks, blog posts, and app listings across ecommerce, almost always presented as a live, current fact about Amazon today.
It isn't a current fact — it's a repeated one. In 2013, McKinsey reported that 35% of what consumers purchased on Amazon.com came from recommendations. McKinsey never published a methodology behind that number, and Amazon has never released a primary source confirming it. Thirteen years later, the same line is still being copied — sometimes with "purchases" quietly upgraded to "revenue," which isn't the same claim.
None of this means recommendations don't work at Amazon's scale — they almost certainly do. It means this specific number belongs in a sentence like "In 2013, McKinsey reported that..." — cited as a piece of ecommerce history, not repeated as a current benchmark to plan a store around.
The numbers that do hold up
A smaller set of stats survive a closer look because they come from named vendors, disclosed sample sizes, and reports you can still find today. They're worth citing carefully — with the segment they measured attached, since that's the part that usually gets dropped.
According to Salesforce, which tracked more than 150 million shoppers across 250 million visits, the 7% of visits where a shopper clicked a product recommendation accounted for 26% of the store's revenue, across $550 million in tracked orders (Product Recommendations report, 2017). The same report found that shoppers who clicked a recommendation spent 5 times more per visit than average, and that 52% of their orders included a recommended product. Notice what the number actually says: revenue concentrated among visits that clicked — not a lift applied to every visitor who saw a recommendation.
According to Nosto, in a 25-day A/B test, shoppers at Bandier who saw cart recommendations spent 10.2% more per order than shoppers who didn't (Bandier case study). It's one retailer and one test, but it's a controlled before/after comparison — rarer in this space than the marketing copy around it suggests.
According to Rokt Aftersell's 2026 Revenue Leak Report, across roughly 50 million cart sessions, cart-based upsells generated 4 times more revenue per session than the same offers shown at checkout. This is one of the few data points that directly compares placements rather than measuring one in isolation.
According to Barilliance, across 300 stores, an average of 12% of sales were attributed to product recommendations — ranging up to 31% on some sites. The range matters more than the average here: it tells you the real number depends on catalog, traffic, and placement, not a fixed formula that applies to any store.
Two more data points round out the picture, with their scope stated plainly. McKinsey's broader personalization research (2019 and 2021) puts typical revenue lift in the 5–15% to 10–15% range — but that figure covers personalization generally (pricing, email, on-site content, and recommendations together), not recommendations alone. Dynamic Yield's published case studies show smaller, placement-specific results: a 4.2% increase in revenue per visitor from product-page recommendations for e.l.f., and a 15% increase in revenue per visitor from add-to-cart recommendations for LUISAVIAROMA.
How to apply this: what to test first
The data above points in a consistent direction: cart placement has the strongest, most directly comparable evidence — Rokt's 4x figure and Nosto's 10.2% test — so it's the reasonable first placement to test on a Shopify store, followed by the product page.
Placement alone isn't enough, though. Product page and cart drawer cross-sells reach shoppers at different points in the decision, and Baymard's UX research — a separate, non-revenue data point — found that 52% of sites fail to show relevant products in cart cross-sells, and 58% show only one type of suggestion. That's a design and relevance problem, not a placement problem: a technically well-placed recommendation still underperforms if it's the wrong product.
How to measure it in your own store
None of the numbers above will match a specific Shopify store exactly — they're a planning reference, not a guarantee. The only reliable answer is what a store's own dashboard shows. Three numbers matter, roughly in this order:
- Click-through rate (CTR) — how often a recommendation gets a click relative to how often it's shown. This tells you whether the pairing and placement are relevant enough to notice.
- Attach rate — of those clicks, how many turn into an add-to-cart or a purchase. This is where relevance either pays off or doesn't.
- Attributed revenue and AOV change — the actual dollar impact, tracked over a defined window rather than assumed from click volume. Dropr's attribution dashboard reports this from impression to completed order, so a store can see its own number instead of relying on an industry average.
A couple of weeks of data is usually enough to see whether a placement is working, provided the store has enough order volume to make the sample meaningful.
Honest expectations for a small store
Most of the numbers above come from larger retailers with high traffic and deep catalogs — worth knowing before assuming they transfer directly to a smaller store. Rokt's 2026 data found that stores under 1,000 orders a month saw $0.20–$0.28 in revenue per transaction from their recommendations. That's a modest, realistic number for a smaller catalog — not the double-digit percentage lifts quoted for enterprise retailers, but a real, compounding return that costs nothing to test. Current Dropr pricing is built around that reality: flat and predictable, regardless of order volume.
The honest takeaway from all of this: recommendations work, the vendor data behind that is real and traceable, and the "35%" version of the claim was never about the results a typical store will actually see.
Related reading
- Product Page vs. Cart Drawer Cross-Sell: Which Placement Converts Better?
- How Dropr Tracks Revenue Attribution: A Step-by-Step Explanation
Frequently Asked Questions
How much revenue do product recommendations generate?
It depends heavily on placement, catalog, and traffic — there's no single site-wide number. Barilliance's 300-store study found an average of 12% of sales attributed to recommendations, ranging up to 31% on some sites. Salesforce found that the 7% of visits that clicked a recommendation drove 26% of tracked revenue. The most reliable answer is the one from a store's own attribution dashboard, measured over a few weeks.
Is the Amazon 35% statistic real?
Not as it's usually presented. It traces back to a single, unsourced line in a 2013 McKinsey report stating that 35% of what consumers purchased on Amazon.com came from recommendations. There's no published methodology and no primary source from Amazon confirming it. It's fine to cite as ecommerce history ("In 2013, McKinsey reported...") but not as a current benchmark.
Should I test product page or cart recommendations first?
Cart placement has the strongest comparative data — Rokt found 4x more revenue per session from cart upsells versus checkout offers, and Nosto's 25-day A/B test found a 10.2% AOV lift from cart recommendations. Start there, then add product-page recommendations once the cart placement is live and performing.
How do I know if product recommendations are actually working on my store?
Track three numbers in order: click-through rate (is anyone clicking), attach rate (are clicks turning into adds or purchases), and attributed revenue with AOV change (the actual dollar result). A couple of weeks of data is usually enough to tell if a placement is working, assuming the store has enough order volume for the sample to mean something.