TL;DR
Product recommendations are the suggested products a store shows a shopper: the "you may also like," the "frequently bought together," the "customers also viewed." Done well, they help a shopper find something they actually want and quietly lift the order value. Done badly, they clutter the page with irrelevant junk and erode trust.
The whole art is in two words from the title: the right thing. A recommendation is only worth showing if it is genuinely relevant to what the shopper is looking at or has done. An accurate, well-placed suggestion feels like helpful service. A random or repetitive one feels like a store trying to pad the cart, and shoppers can tell the difference instantly.
Getting it right comes down to a few decisions: which type of recommendation to show, where on the journey to show it, how it is generated, and, above all, keeping relevance ahead of volume. More recommendations are not better; better ones are. This guide walks through the types, the placements, how they are generated, how it differs by what you sell, and the specific ways recommendations backfire when the "right thing" turns out to be the wrong thing.
What Product Recommendations Actually Are
A product recommendation is any suggestion your store makes about what a shopper might want next, shown somewhere in their journey. The familiar examples are the row of products under a listing labelled "customers also bought," the "complete the look" on a fashion item, the "recommended for you" on a homepage, or the add-on offered in the cart.
The purpose is twofold, and both halves matter. The first is discovery: helping a shopper find products they would not have found on their own, especially in a large catalog where most items never get browsed. The second is order value: a good recommendation naturally leads a shopper to add something, raising the average order without any additional acquisition cost. These are related to but distinct from upselling and cross-selling: an upsell is a deliberate offer to trade up or add on, while a recommendation is a suggestion the shopper is free to explore. Recommendations feel like help; hard offers feel like selling, and the tone difference matters.
The reason recommendations get so much attention is that they scale personalization. A physical shop assistant who knows the stock can point each customer to the right thing; a recommendation system tries to do that for every visitor at once, automatically. When it works, it is one of the highest-leverage features a store has, because it improves discovery, order value, and the experience all at the same time. When it fails, it does the opposite of all three. The difference is entirely whether it shows the right thing.
The Number Everyone Quotes, and What It Really Says
Any discussion of recommendations eventually meets the famous statistic: that recommendations drive 35 percent of what people buy on Amazon and 75 percent of what they watch on Netflix. It is quoted everywhere to prove recommendations are enormous.
It is worth knowing where that number comes from, because it is shakier than its fame suggests. Both figures trace back to a single sentence in a 2013 McKinsey article, stated with no source, no dataset, and no footnote, and neither Amazon nor Netflix has ever published or confirmed them. Every version you see today is repeating that one unsourced line more than a decade later. That does not mean recommendations do not work, they plainly do, but the specific 35 and 75 percent figures are folklore, not measurement, and you should not build a business case on them.
The better-sourced version of the same idea is more useful anyway. McKinsey's later research found that companies that excel at personalization generate 40 percent more revenue from those activities than average performers, and that most consumers now expect personalized interactions and are frustrated when they do not get them. That is a real, grounded finding, and it points the same direction: relevant, personalized recommendations earn more, and their absence is increasingly noticed. The lesson to carry forward is not the number, it is the habit of judging your own recommendations by what they actually add rather than by what worked for Amazon.
The Types of Recommendation, and What Each Is For
Not all recommendations do the same job, and showing the wrong type in the wrong place is a common mistake. Here are the main types and what each is genuinely good for.
Frequently bought together. Products that are commonly purchased with the item being viewed: the case with the phone, the filter with the coffee maker. This is the highest-intent recommendation because it completes a purchase the shopper is already making. It belongs on the product page and in the cart.
Customers also bought or also viewed. Items that other shoppers who looked at this product went on to buy or browse. This aids discovery of alternatives and related products, and it works well on the product page for shoppers still deciding.
You may also like or recommended for you. Personalized suggestions based on the individual's browsing and purchase history. This is the discovery workhorse for returning visitors, and it belongs on the homepage, in email, and across the journey once you know something about the person.
Recently viewed. A simple, powerful reminder of products the shopper already looked at, helping them pick up where they left off. It is not really a recommendation in the algorithmic sense, but it converts well because the interest is already proven.
Bestsellers and trending. What is popular right now. This is the fallback for shoppers you know nothing about yet, new visitors with no history, because popularity is a reasonable guess when personalization is impossible. It also carries social proof.
Complete the look or build the bundle. Curated complementary items that go together, common in fashion and home. This is part recommendation, part styling, and it works because it helps the shopper picture the whole rather than the part.
The practical point is that each type answers a different shopper question. "Frequently bought together" answers "what else do I need with this," "you may also like" answers "what else might I want," and "bestsellers" answers "what should I look at when I have no idea." Matching the type to the question is most of showing the right thing.
Where to Show Them
Placement matters as much as the recommendation itself, because the same suggestion can help or annoy depending on where it appears. Here is where each earns its place.
The homepage is for personalized discovery, "recommended for you" for returning visitors, and bestsellers or new arrivals for first-timers. It sets the tone and pulls people into the catalog.
The product page is the workhorse. Below or beside the item, "frequently bought together" and "customers also bought" help a deciding shopper find complements and alternatives without leaving the page. This is usually the highest-value placement.
The cart is for gentle, relevant add-ons: the small complementary item that completes the order. Keep it light here, because the shopper is on the way to checkout and too much distraction risks the sale.
After purchase, on the thank-you page or in a follow-up, is an underused spot for recommending the next logical product or a replenishment, because the sale is already secure and the shopper is warm.
Email is one of the strongest recommendation channels, because it can be fully personalized to the individual. Recommendation blocks in emails genuinely lift engagement, which is why the best flows include them.
Search and empty states are quietly important: when a shopper searches and finds nothing, or lands on an out-of-stock product, a good recommendation rescues a dead end that would otherwise be an exit.
The rule across placements is that the closer the shopper is to buying, the more the recommendation should complete the current purchase rather than distract from it, and the earlier they are, the more it should open up discovery.
What "The Right Thing" Actually Means
Everything above serves one principle, and it is the heart of the title: relevance beats volume, every time. A store's instinct is often to show more recommendations, on the theory that more suggestions mean more chances to sell. The opposite is usually true. A few genuinely relevant suggestions convert and build trust; a wall of loosely related products reads as noise and erodes it.
The reason is psychological. When a recommendation is accurate, it feels like the store understands the shopper, and that feeling of being understood is exactly what makes an experience feel personal and worth returning to. When a recommendation is off, recommending winter coats to someone shopping for swimwear, or pushing an item they just bought, it signals the opposite: that the store is not paying attention and is simply trying to sell something. One or two bad recommendations can undermine the credibility of all of them.
Relevance also means context, not just category. The right thing depends on where the shopper is in their journey. Someone who just added a camera wants a memory card and a case, not another camera. Someone browsing a category wants to see the range, not a single hard push. Someone who has bought from you five times wants recommendations that reflect their taste, not generic bestsellers. Showing the right thing means reading that context and matching the suggestion to it, which is why the systems that generate recommendations matter, and why the cheap approach of showing the same popular products to everyone leaves so much on the table.
The honest test for any recommendation is simple: would a knowledgeable shop assistant, watching this exact shopper, suggest this exact product right now? If yes, show it. If not, showing it costs you more in trust than it earns in the occasional extra sale.
Rules, Algorithms, and AI: How Recommendations Are Generated
Recommendations are produced in three broad ways, and knowing the difference helps you choose what your store actually needs rather than buying more sophistication than you can use.
Manual rules are recommendations you set by hand: "show these three accessories on this product," "pair this shirt with these trousers." They are simple, fully controllable, and perfect for small catalogs or for curated pairings where you know best, such as complete-the-look styling. The limitation is that they do not scale or personalize; every rule is work, and they treat all shoppers the same.
Algorithmic recommendations, often called collaborative filtering, use the behavior of many shoppers to infer what goes with what: people who bought this also bought that. This is what powers "customers also bought," and it scales across a large catalog without manual effort. Its weakness is the cold-start problem: it needs data to work, so it struggles with brand-new products that no one has bought yet and with brand-new visitors it knows nothing about. This is why new stores and new items often fall back to bestsellers.
AI personalization goes further, using an individual's full behavior and broader patterns to tailor recommendations to each person in real time, and increasingly to generate them dynamically. This is the most powerful approach and the one behind the strongest "recommended for you" experiences, but it needs enough data and traffic to learn from, and it is only worth its cost once a store has the volume to feed it.
For most stores the honest answer is a blend: manual rules for the high-value curated pairings you want to control, algorithmic recommendations for scale across the catalog, and bestsellers as the fallback for cold-start moments. You do not need the most advanced system to show the right thing; you need the right approach for your catalog size and traffic. A small store with good manual pairings often out-performs a large store running a poorly-tuned algorithm on irrelevant data.
Recommendations by Category
What "the right thing" looks like depends heavily on what you sell, so the same recommendation strategy does not transfer across categories. A quick tour of how it shifts.
Fashion and apparel. The dominant pattern is "complete the look": showing the full outfit around a single item, because clothing is bought as combinations and shoppers respond to seeing the whole styled together. Size and fit signals also matter, and recommendations that respect a shopper's previous sizes feel far more relevant.
Electronics and high-consideration goods. The winning recommendation is compatibility and completion: the memory card, the case, the cable, the compatible accessory. Because these purchases are researched, "customers also bought" and comparison-style recommendations help a deliberating shopper, while a random unrelated product just adds noise to a careful decision.
Beauty. Recommendations work as routine-building: the cleanser that goes with the serum, the shade that matches, the next step in a regimen. Beauty shoppers respond to recommendations framed as expertise, and personalization by skin type or previous purchases lifts relevance sharply.
Food, CPG, and consumables. The most valuable recommendation is often replenishment and pairing: reminding a shopper of what they buy repeatedly, or suggesting what goes with it. Because the purchase recurs, recommendations based on past orders are unusually accurate and welcome.
Home and furniture. Like fashion, this is a "goes together" category: the lamp with the desk, the cushions with the sofa. Because items are high-consideration and visual, curated room-style recommendations and complementary pieces outperform generic suggestions.
The unifying idea is that some categories are about completion (electronics, food) where the right thing is what you need alongside your purchase, and some are about combination (fashion, home) where the right thing is what goes with it to make a whole. Knowing which your category is tells you what kind of recommendation to lead with.
When Recommendations Backfire
Recommendations are not a free win, and it is worth naming the specific ways they turn negative, because each is common and avoidable.
Recommending what the shopper just bought or returned. The classic failure: someone buys a sofa and gets shown sofas for weeks, or is recommended the exact item they returned. It signals the store is not paying attention, and it is one of the fastest ways to make personalization feel broken.
Showing too many. A page buried in recommendation rows becomes noise, dilutes the actual product, and can slow the page down. More suggestions past a small number reduce, not increase, the odds any one is acted on.
Irrelevant or off-context suggestions. A recommendation that has nothing to do with what the shopper is looking at reads as a random sales push and undermines trust in every other recommendation on the site. Relevance is not optional; a wrong recommendation is worse than none.
Distracting from the purchase. Aggressive recommendations at checkout can pull a shopper away from completing the order they came to make. Near the point of purchase, recommendations should complete the sale, not open new decisions that stall it.
Over-discounting through recommendations. Constantly recommending discounted or clearance items trains shoppers to expect deals and can cannibalize full-price sales. The right thing is usually the relevant thing, not the cheapest thing.
Slowing the site. Heavy recommendation widgets that load slowly hurt the experience and conversion more than the recommendations help. Performance is part of relevance; a suggestion that arrives after the shopper has left is worthless.
None of these argues against recommendations. Each argues for restraint and relevance: fewer, better, more contextual suggestions, and a willingness to show nothing when you have nothing genuinely relevant to show.
Common Mistakes With Product Recommendations
- Prioritizing volume over relevance. More recommendations are not better. A few accurate ones convert and build trust; a wall of loose ones erodes it.
- Ignoring context. The right recommendation depends on where the shopper is and what they just did. Recommending another camera to someone who just bought one misses the point.
- Recommending the already-bought or returned item. Nothing signals a broken system faster. Exclude what the shopper already owns or rejected.
- Using one approach for everything. Manual rules, algorithms, and AI each fit different needs. Small curated catalogs and cold-start moments need different tools than a large personalized store.
- Over-discounting via recommendations. Constantly surfacing clearance trains shoppers to wait for deals and cannibalizes full-price sales.
- Letting recommendations slow the page. A heavy widget that hurts load time costs more in conversion than it earns in suggestions.
A Simple Way to Start
If you are adding or fixing product recommendations, work in this order. Most of the value is in the first two steps.
First, get the product-page recommendations right. Add "frequently bought together" and "customers also bought" on the product page, the highest-value placement. Make sure they are genuinely relevant and exclude the item being viewed.
Second, handle the cold-start fallback. For new visitors and new products with no data, show bestsellers or trending items rather than nothing or something random. This covers the moments personalization cannot.
Third, add recommendations to email. Personalized recommendation blocks in your flows are one of the strongest, cheapest wins, because email can be fully tailored to the individual.
Fourth, respect context everywhere. Exclude already-purchased items, match the recommendation type to the placement, and keep the cart recommendations light so they do not stall checkout.
Fifth, measure relevance, not just clicks. Watch whether recommendations lift order value and repeat visits without raising bounce, and be willing to show fewer, better suggestions. If a placement is not helping, remove it rather than tuning noise.
Most stores add every recommendation widget their platform offers and call it done. The ones that show the right thing do the opposite: fewer placements, higher relevance, and constant exclusion of what does not fit.
FAQ
What are product recommendations in ecommerce?
They are the suggested products a store shows a shopper, such as "you may also like," "frequently bought together," and "customers also bought." Their purpose is to help shoppers discover relevant products and to lift the average order value without extra acquisition cost.
Do product recommendations actually increase sales?
Yes, when they are relevant. Personalization leaders generate meaningfully more revenue from these activities, and well-placed recommendations lift order value and discovery. But irrelevant recommendations can hurt trust, so the gain depends entirely on showing the right thing, not on showing more.
Where should I put product recommendations?
The product page is the highest-value placement, using "frequently bought together" and "customers also bought." Add personalized suggestions on the homepage and in email, keep cart recommendations light so they do not stall checkout, and use recommendations to rescue empty search results and out-of-stock pages.
How are product recommendations generated?
Three main ways: manual rules you set by hand, algorithmic recommendations based on what many shoppers do together (which power "customers also bought"), and AI personalization tailored to the individual. Most stores blend manual rules for curated pairings, algorithms for scale, and bestsellers as a cold-start fallback.
Why do some product recommendations feel bad?
Because they are irrelevant or out of context, recommending an item the shopper just bought, showing too many, or pushing unrelated products. An accurate recommendation feels like helpful service; an off one signals the store is not paying attention and undermines trust in every suggestion.
Is the Amazon 35 percent recommendations statistic true?
It is unverified. The figure, along with the Netflix 75 percent claim, traces to a single unsourced 2013 McKinsey sentence and has never been confirmed by either company. Recommendations clearly work, but that specific number is folklore rather than measured fact.
Where to Go From Here
Product recommendations reward restraint and relevance over volume. Get the product-page suggestions right, fall back to bestsellers when you have no data, personalize email, respect context everywhere, and always ask whether a knowledgeable assistant would really suggest this to this shopper right now. Show the right thing and recommendations feel like service; show too much and they feel like noise. Recommendations help the shopper find and choose; for the moments where a real conversation closes the sale or answers the question holding them back, Kovax handles voice, WhatsApp, and support for Shopify stores.
