Product recommendation software selects which products to show alongside what a shopper is doing: related items on a product page, add-ons in the cart or at checkout, products in an email. The picks may be personalized to one shopper, based on the product being viewed, drawn from what other shoppers bought, or chosen by hand. It is rarely the first tool a store buys, and it is often one it already has, because commerce platforms, search vendors, personalization suites and email platforms all ship recommendations of their own. The buyers who run them describe a category defined less by vendors than by a choice between those sources.
What is product recommendation software?
Product recommendation software selects and ranks products for placements such as related-item carousels, cart add-ons, checkout offers, home-page modules and email product blocks. It can use catalogue relationships, the product being viewed, browsing or purchase history, patterns across shoppers, or rules the merchant sets. Some recommendations are personalized to one shopper; others show the same related or promoted products to everyone.
The choice is the job. The slots vary by surface, and the logic behind them runs from hand-picked pairings to models trained on what other shoppers bought; most tools mix the two, with an algorithm underneath and merchant rules on top.
Unlike site search, it needs no query from the shopper, and unlike the wider personalization of copy, banners and offers, its output is a list of products.
The second half of the question is the one the interviews answer:
Among interviewed companies that rate an identifiable recommendation tool, a dedicated recommendation or upsell app is the most often named of five kinds, at about a third of them. The others are capabilities inside commerce platforms, search and discovery software, personalization or testing platforms and email platforms. A company can rate more than one kind, so these are overlapping shares of the companies interviewed, not market shares.
About two in three of those companies rate no dedicated app for the job; their documented recommendations run on broader platforms, though that does not show they have no other recommendation software. Buyers move both ways between a platform’s built-in recommendations and a specialist, and their complaints split two ways: recommendations still set by hand, and algorithms they cannot steer.
What is the difference between product recommendations, personalization and merchandising?
Recommendations pick products for particular placements, by contextual, behavioural, algorithmic or manual logic; personalization adapts what a shopper or audience sees; merchandising governs which products appear, and in what order, across category pages, collections and search results. They overlap because a recommendation module can be personalized and merchandising rules can constrain what an algorithm recommends.
Personalization is the broader job: adapting recommendations, banners, copy, offers or email to a shopper or audience. A recommendation module is one thing it can adapt, though a related-products row can also show the same items to everyone. That overlap is why personalization and testing platforms such as Dynamic Yield, Monetate and Adobe Target run recommendations too. What a personalization engine covers beyond products is set out in what is a personalization engine?
Merchandising applies commercial rules and controls to every product list, browsed or searched, by sorting, pinning, boosting and burying. It meets recommendations in the controls: the rules a merchandiser lays over an algorithm to exclude an item, favour a promotion or show new arrivals first. The wider job is set out in what is merchandising software?
That shared ground is where buyers spend most of their words. A specialty home retailer that mostly lets the algorithms run asked its vendor to exclude storage components that make no sense on their own, and says “we have a way to do that in the console, but for whatever reason, items keep showing.” A specialty retailer rating the AI engine that now sorts its category pages, and is set to take over its recommendations, says “there’s not enough ability for a human to intervene.”
Site search is the other neighbour
Search responds to a query the shopper typed; recommendations infer what to show without one. Many discovery platforms do both, and buyers often shop for them together. How far the two overlap, and which tools lean toward each job, is on site search vs. product recommendations.
Who buys product recommendation software?
Ecommerce, merchandising and digital teams at consumer brands and retailers, from small Shopify stores to national chains. The companies describing it here are mostly apparel and footwear, beauty, home goods, sporting goods and specialty retail, where a shopper browsing one item is a chance to sell a second.
The interviews also show where the job stops paying. Several small-catalogue brands say a dedicated tool is not worth it. A food brand weighing one says “I don’t think it’s really the best fit for our company as of right now with less than 100 SKUs,” and a consumer-products brand rating its platform’s recommendations as basic still runs no other tool, because “we have a very small set of products, so there’s not a lot of magic behind the recommendations.”
That matters for the definition. Our interpretation, not a measured finding: catalogue size changes the problem an engine has to solve. A small assortment can be managed with manual pairings or native recommendations, while a large catalogue makes automated selection and upkeep worth more. Size alone does not settle the return: a small store can still gain from relevant cross-sells, and a large retailer may find an advanced engine adds little.
Where do product recommendations come from?
Five kinds appear among interviewed companies that rate a recommendation tool. Each company counts once within a kind, but a company that rates tools of two kinds appears in both, so the shares below do not add to 100% and are not ecommerce market shares. From most to least common:
- A dedicated recommendation or upsell app
- A search and discovery platform
- The commerce platform’s built-in recommendations
- A personalization or testing platform
- The email platform
These are overlapping shares of the companies interviewed, not market shares or a full inventory of every tool they use. Most companies appear in only one kind. About two in three rate no dedicated recommendation app for the job. A small number of large retailers describe systems they built themselves, which a count of rated products does not consistently capture.
A dedicated recommendation or upsell app
Tools whose main job is the product slot: recommendation engines such as Nosto, Algonomy and Stylitics for outfits, and the cart, checkout and post-purchase upsell apps that sit inside Shopify stores. This is the route buyers take when they want more than the platform gives them.
An apparel brand says the out-of-the-box carousel designs from its app look poor on its home page, but “we have seen incredible performance benefits from it, especially in our mini cart.”
A search and discovery platform
Algolia, Bloomreach, Constructor, Coveo and Searchspring, usually bought for search first, with recommendations running off the same catalogue and behaviour data. Buyers choosing this route describe wanting one vendor for discovery.
A beauty retailer chose a vendor with “a search and a recommendation tool built into one,” and says “the recommendation engine is fed by what people search as well.”
The commerce platform’s own
Shopify’s built-in recommendations and Salesforce Commerce Cloud’s Einstein are the two named most here, with Adobe Commerce and VTEX behind them. No new vendor, and buyers rate it on whether it is good enough rather than best.
A DTC brand: “We’re currently using out of the box Shopify search and recommendations. And it’s like it’s working well enough.” A luxury retailer on Einstein says its setup is “very rules-based” and “too underwhelming.”
A personalization or testing platform
Dynamic Yield, Monetate, Adobe Target and Optimizely. Recommendations sit beside the experiments and audience targeting the platform was bought for, which makes testing one recommendation against another part of the same tool.
A specialty retailer that moved its recommendations from Einstein to Monetate says “Monetate beat them out of the park with their recommendation.”
The email platform
Klaviyo, Bluecore and other email platforms fill product blocks in messages. For many brands this is where recommendations live outside the site, and some compare it directly with their site engine.
A baby-products brand values that “you don’t have to get another tool to integrate with Klaviyo just to get product recommendations.” A DTC food brand on the same platform says “I specifically do not like the product recommendations on Klaviyo.”
Companies move between these sources in both directions. Two completed moves took recommendations off a commerce platform’s own tools: the specialty retailer above from Einstein to Monetate, and a regional retailer that tried its platform’s recommendations, “just weren’t impressed,” and runs them through Searchspring, as it still did in its latest interview. One went the other way: a fashion brand on Salesforce Commerce Cloud dropped its third-party recommendation vendor once Einstein improved, saying “We’ve actually gone back to using the Salesforce one,” and in its latest interview Einstein still runs its site recommendations, with Bloomreach handling email. A beauty brand stopped exploring a recommendation app in 2024 when Shopify expanded its own; that is an evaluation ended, not a migration.
Other changes are still plans. A specialty retailer intends to replace Einstein’s recommendations with Cimulate, the AI engine that already sorts its category pages, and its interview does not describe the change as done. Salesforce completed its acquisition of Cimulate on March 3, 2026, so both engines now sit in one company’s portfolio; that by itself does not show the two have been combined, or that this buyer has switched.
Which product recommendation tools do buyers use?
The dedicated kind spans broad recommendation engines such as Nosto and Algonomy, use-case tools such as Stylitics for outfits, and Shopify apps for cart, checkout and post-purchase upsells. Algonomy, which its site now says is part of ADA, lists its Recommend product for product recommendations and content personalization; Stylitics describes styled outfits and product bundles that adapt to each shopper. An outfit or post-purchase tool is not automatically a substitute for an engine running every storefront and email placement. Most other recommendation tools buyers name were bought for search, personalization, email or commerce first. The split below is by each product’s main job, not by whether it can recommend products, since nearly all of them can.
Built for recommendations
Nosto · Algonomy · Stylitics · Octane AI (quizzes) · cart and post-purchase upsell apps
Recommendations inside a broader platform
Shopify · Salesforce Commerce Cloud · Algolia · Bloomreach · Constructor · Coveo · Searchspring · Dynamic Yield · Monetate · Adobe Target · Klaviyo · Bluecore
The platform column is not a lesser tier. Each of those products ships recommendation logic as a documented feature. Shopify’s help center, checked in November 2026, says related products are generated automatically for every product, while complementary products, the pair it with kind, are chosen by hand in its Search & Discovery app. Salesforce’s help center describes Einstein recommenders built on up to three strategies, with up to thirty rules to show, hide, promote or demote products. Klaviyo’s help center lists personalized feeds, including one that uses collaborative filtering to predict what a recipient will buy next, beside classic best-seller and newest-product feeds. A fashion label complained in late 2024 that its email recommendations offered only best sellers or most viewed; that described its setup at the time, and the personalized feed is in Klaviyo’s documentation now.
What separates the buyers is who sets the rules. Several say their recommendations are still built by hand: a beauty brand rates its platform’s recommendations low because “it’s very manual yet,” and a footwear brand says “I have to manually put in what products I want to feature in the upsell at checkout.” Others want the reverse, more say over an algorithm. A specialty manufacturer chose hand-picked pairings on purpose, so a new product can be recommended the day it launches; otherwise, it says, “the recommendation algorithms haven’t learned about it yet.” The arrangement buyers praise is a mix: a footwear group says its recommendations are “predominantly AI driven, but we do have some manual control there.”
The largest retailers sometimes build rather than buy a finished engine. A large retailer runs a recommendation platform “built in-house” on Google’s cloud AI services; another company builds its product recommendations on Kumo; a grocery retailer moved to an engine of its own even though it narrowly lost an A/B test against its vendor, because it wanted to own it. Building is an ownership choice, not a separate kind of AI product recommendations: the commerce, search, personalization and email tools above run models too.
Conversational shopping is a different interface. A large retail chain runs a Google-powered shopping assistant that recommends products in conversation, first tested on a limited assortment and then expanded once the results were good. That differs from an engine filling a fixed carousel, though both can draw on the same behaviour data.
How do you know if product recommendations are working?
The strongest evidence comes from controlled experiments, not the revenue a vendor credits to its widgets. A tool can report purchases made after a shopper viewed or clicked a recommendation, but many of those would have happened anyway; a beauty brand that likes its upsell app says the app claims credit for nearly every sale on the site. Widget-attributed sales and incremental sales are different measurements.
Buyers who measured describe the gap. A national retailer gives its engine a middling rating because “from an ROI perspective, it’s been hard to quantify what that ROI is.” An apparel retailer says its outfit recommendations show strong associated demand, but “when we A/B test it on and off, it’s hard to prove the incremental value that it has to the business.”
Head-to-head tests settled several switches, in both directions. A multinational beauty group stopped using a specialist recommendation vendor after testing it against its email platform’s own recommendations, which won. A women’s fashion label that moved its recommendations to a new vendor ran an A/B test and found “there was no measurable difference between the two vendors,” which is one company’s result, not a verdict on either tool. And the grocery retailer above switched despite losing its test, to own the engine.
Want a review of your own stack?
Get my review →Should you buy product recommendation software?
Check what the tools you already own can recommend before you add one. In these interviews most companies’ recommendations run inside a commerce, search, personalization or email platform, and buyers move back to those as well as away from them.
The rest is our recommendation, not a practice buyers describe. Start with the surface: list the slots you want filled (product page, cart, checkout, home page, email) and which of your current tools already fills each one, since the commerce platform, the search vendor and the email platform may each be running a different engine today. Then decide where control has to sit. If merchandisers need to exclude items, pin new arrivals or favour a promotion, run those exact requests through the trial, in the vendor’s console, as your team would make them, since the buyers above found those gaps only after go-live.
Then decide which question the test answers. An engine comparison puts the proposed engine against the current one: assign comparable shoppers or sessions at random, keep the placements, design and promotions the same, and run it across a normal buying cycle. An incrementality test asks whether the module earns its place at all, against a control that shows no recommendations or a simple baseline such as best sellers.
Either way, judge on revenue, contribution margin and conversion per assigned visitor, counting everyone in each group rather than the shoppers who clicked, and allow for traffic, stock and seasonality when sizing the test. A small catalogue may find manual pairings or native recommendations do as well; a pricier engine should earn its place through better outcomes or less work to run, not more attributed sales.
For how recommendations overlap search, see site search vs. product recommendations and what is ecommerce site search? For the platforms that run them as one part of personalization, see what is a personalization engine? and what personalization buyers wish they’d known. Head-to-heads include Nosto vs. Dynamic Yield and Nosto vs. Bloomreach.
Common questions
What is product recommendation software?
Product recommendation software selects which products to show in a placement alongside what a shopper is doing: related items on a product page, add-ons in the cart or at checkout, products on the home page or in an email. It chooses them from catalogue relationships, the product being viewed, browsing or purchase history, patterns across shoppers, or rules the merchant sets. Some recommendations are personalized to one shopper; others show the same related products to everyone. Unlike site search, they need no query.
What is the difference between product recommendations and personalization?
Recommendations pick products for particular placements; personalization adapts what a shopper or audience sees, including copy, banners, offers and email content. A recommendation module can be personalized, but it need not be: a related-products row often shows the same items to everyone. Many personalization and testing platforms also run recommendations, which is why the terms blur, but a tool that is strong at one is not automatically strong at the other.
What is the difference between product recommendations and merchandising?
Merchandising governs which products appear, and in what order, across category pages, collections and search results, by business rules such as pinning, boosting and burying. Recommendations pick products for particular placements. They overlap in the controls: merchandising rules can constrain what an algorithm recommends, most recommendation tools let merchandisers add exclusions and manual picks, and buyers in these interviews ask both for more of that control and for more automation.
Does Shopify have built-in product recommendations?
Yes. Shopify’s help center, checked in November 2026, says related product recommendations are generated automatically for each product, and its Search & Discovery app lets merchants add up to ten related and ten complementary products by hand; complementary picks are manual, and the generated ones cannot be edited. In these interviews some Shopify brands call the built-in recommendations good enough, others call them basic, and moves run both ways between them and dedicated apps.
What are AI product recommendations?
Recommendations chosen by a model trained on shopper behaviour rather than by rules a person writes. Most recommendation tools now describe themselves this way. In these interviews the term covers three things: an algorithm inside a bought tool, a recommendation model a large retailer builds on cloud AI services, and shopping assistants or chatbots that suggest products in conversation. Buyers whose recommendations are still set by hand are the ones asking for it most directly.
How do you know if product recommendations are working?
With a controlled experiment. Vendors report the sales that pass through their widgets, and one buyer says its app claims credit for nearly every sale; a sale that touched a recommendation is not a sale the recommendation caused. Comparing two engines and asking whether recommendations add sales at all are different tests: the first splits comparable shoppers between engines, the second compares against shoppers shown no module or a simple baseline. In these interviews head-to-head tests gave some buyers a clear winner and another no measurable difference.
Which product recommendation tools do buyers use?
Five kinds, which overlap. Dedicated recommendation and upsell apps such as Nosto, Stylitics and Algonomy; search and discovery platforms such as Algolia, Bloomreach, Constructor, Coveo and Searchspring; commerce platforms with built-in recommendations, such as Shopify and Salesforce Commerce Cloud with Einstein; personalization and testing platforms such as Dynamic Yield, Monetate and Adobe Target; and email platforms such as Klaviyo and Bluecore. A smaller group of large retailers build their own.
A tech stack review, powered by the buyers who've been there
What matters most to you, a shortlist for your stack, and the next steps to take — built from hundreds of verified buyer interviews.
Get my stack reviewFree · no password · your answers are anonymized before they inform research like this.