Product Recommendations for eCommerce: Strategies & Examples

Picture this. John lands on your store looking for a motorbike helmet for his daughter. He clicks a basic one, then spots a carousel of similar helmets, including one with kitten prints she’ll love. It goes in the cart. At checkout, matching gloves pop up too. He buys both, happily spending more than planned.

That’s product recommendations doing their job: helping shoppers find things they didn’t know they wanted, while quietly lifting your average order value.

It’s simple in theory, but most guides skip the detail that actually makes it work: what a customer searches for is one of the strongest signals for what to recommend next. Ignore it and you’re guessing. Use it, and your carousels start acting like a salesperson who knows the shop.

This guide covers what product recommendations are, the strategies that work, where to place them, how to configure the logic behind them, and how to measure if they’re paying off.

What Product Recommendations Actually Do for Your Store

Before picking tactics, it helps to know what you’re aiming for. A recommendation strategy without a goal is just a random carousel of products, and random carousels rarely convert.

Most stores use recommendations to hit one or more of these:

• Reduce cart abandonment. Showing a relevant alternative or a complementary item can be the nudge that keeps a customer moving toward checkout instead of clicking away.
• Increase average order value. Cross-sells, upsells, and bundles are the fastest way to get customers to add one more thing to their cart.
• Improve retention. Recommendations built on purchase and browsing history make returning customers feel like the store remembers them, which keeps them coming back.
• Lift conversion rate. Well-placed suggestions, especially ones powered by real behavioral or search data, turn browsers into buyers more often than generic “you might also like” widgets.

Keep these in mind as you build your strategy. They’ll tell you which logic to use, where to place it, and how to measure whether it’s working.

This isn’t just a hunch. McKinsey’s research on personalization has found that the large majority of consumers now expect brands to personalize their experience, and a similarly large share say they get frustrated when a brand fails to deliver that.

The Types of Product Recommendations, Explained Simply

Not all recommendations are built the same way. Under the hood, most systems rely on a handful of approaches, and knowing the difference helps you pick the right one for the right spot.

• Content-based. Products are matched by shared attributes, things like category, brand, title, or description. If someone is viewing a red running shoe, this shows other red running shoes or the same brand’s other models.
• Collaborative filtering. This looks at what other customers with similar behavior bought or viewed. “Frequently bought together” and “people also viewed” both fall under this bucket.
• Popularity-based. Straightforward: show what’s selling or getting the most traffic right now. No personalization involved, but it works well for new visitors you know nothing about yet.
• Personalized (behavior-based). Built from an individual shopper’s own history: what they’ve viewed, bought, or added to cart.
• Search-driven. The one most guides leave out. This uses what a shopper actually typed into your search bar, and what they clicked on from those results, to shape recommendations shown elsewhere on the site. It’s arguably the richest signal available, because it’s the customer stating their intent directly rather than you inferring it from clicks and page views.
• Hybrid. Most real systems don’t rely on a single method. They blend two or more of the above, for example combining content-based matching with collaborative filtering, so a single carousel can fall back on attribute similarity when there isn’t enough behavioral data yet.
• Deep learning-based. More advanced, AI-powered systems use machine learning models to find non-obvious patterns in large catalogs and big traffic volumes, patterns that simpler rule-based matching would miss.

A strong recommendation setup usually blends several of these product recommendation algorithms rather than leaning on just one.

Product Recommendation Strategies and Techniques That Actually Move Sales

Here are the strategies worth putting into practice. We’re starting with search-driven recommendations on purpose, not because it’s the flashiest tactic, but because it’s the lens worth applying to everything else on this list. Once you see how a search query can sharpen a recommendation, it’s hard not to notice how much every other technique below improves when you feed it real signal instead of a guess.

Search-Driven Recommendations

This is where Doofinder’s approach differs from a generic recommendation widget. Instead of only looking at what a customer clicked or bought, it factors in what they searched for, since a typed query is the clearest statement of intent a shopper gives you. This is sometimes called customer intent-driven recommendations, built around what someone wants right now, not just what they’ve browsed before. Search-driven recommendations tend to outperform purely popularity-based ones because they’re grounded in explicit intent, not just a pattern pulled from aggregate behavior.

Use the power of social proof

How many millions of people could have read The Da Vinci Code only because it was a bestseller? In the end, it’s inevitable; nothing is as reassuring as social proof.

Let’s be honest; if we see that a product is popular we think “If that many people are buying it, there must be a reason”. And that increases our chances of including it in our cart.

So go ahead and take advantage of this to give your most popular products visibility:

• Including a “Most Sold” section on your homepage or even in categories sites.
• Adding a “Bestseller” label to the product card.

In this case, this store has chosen to include them in its “related products” section, adding a “Best Seller” tag.

Offer and recommend the user complementary products

Do you remember John (from the beginning of this post) ended up buying a more expensive helmet and some gloves he hadn’t even considered?

Well, that example contains 2 of the most popular strategies to increase any eCommerce’s average ticket price:

• Up-selling: you recommend a customer a better product than the one he has in mind at a higher price. For instance, you’re looking at an iPhone 7 product card and among the recommended products there’s an iPhone 10.
• Cross-selling: products that complement the one a customer has bought (or is about to). In the case of the iPhone, we could suggest the user to buy a matching case or a pair of AirPods.

And… no; we did not come up with this idea ourselves.

If you look closely, Apple’s product cards include a “Compatible Accessories” section.

By the way, if you’d like to learn more about this strategy, make sure you check out this post.

Special offers recommendations

Another really interesting option is to recommend the products you have on sale.

Very frequently, it’s the case that a customer hadn’t considered getting a specific product, but then they see it’s on sale and think: “Well, I guess I could take advantage of this offer and get it now at a cheaper price; I don’t want to miss out on this offer.”  

The scarcity trigger in action. 😉

The best thing is that product recommendation is a very adjustable resource. Recommendations can be displayed:

• On the homepage (on a banner, on a specific tab of the browsing menu bar…).
• On product listings.
• In newsletters.

Try different options to see where you get a higher conversion.

Recommend Seasonal products

A way of anticipating your customer’s needs is including seasonal products in your recommendations.

For example, think about the famous “Back to school” one.

If you sell backpacks or school supplies, surely a fair amount of your customers will come searching for those products. So, why not make it easy for them and include them among your featured products on the homepage?

Another pretty curious example.

See how Amazon Spain has made the best of the Camino de Santiago pilgrimage season to highlight product categories that relate to the journey (backpacks, shoes, foot care products, etc.)

“Other users have also viewed…”

This recommendation indicates which product sites have been visited by other buyers.

Something similar to what we see on IKEA’s website when we go on the product card of one of their planters.

This strategy reinforces social proof; the user sees that certain specific products are attractive to other users and that builds up trust.

But, what if we take it up a notch?

For example, in addition to showing them that other people are also viewing such products, you can let them know there are only a few items left in stock.

It’s like saying: “Watch out, this is a very popular product and if you don’t hurry up, you could miss out on it”.

Once more we’re taking advantage of the scarcity’s psychological trigger we mentioned above.

Product Recommendation: Bundling

Let’s see another frequent and effective way of recommending products: kits.

This strategy, known as product bundling, consists in grouping different complementary products and offering them in a single bundle.

This works well because:

• The user doesn’t have to choose: by offering them a “closed pack”, they don’t have to spend a while comparing product cards to decide which to get.
• It makes the purchase process more efficient: by only clicking once, they can add all the products in the bundle to their cart.
• It reduces the price: even if it’s not a rule of thumb, the usual thing is getting the full kit at a cheaper price than getting all the products separately.

For example, if you’re looking at a desktop on Amazon, right below you’ll be offered a mouse, trackpad and keyboard bundle. 

Personalized product recommendations

So far, everything we’ve seen involves general recommendations, identical to all users.

But now, we go one step beyond with personalized recommendations.

These suggestions, as you can tell, vary automatically from customer to customer (which is known as dynamic content).

Every user is offered personalized results based on:

• Their purchase history.
• The products they’ve viewed.
• Their demographics.
• Etc.

The following is an example of dynamic recommendations on Amazon, based on previous purchases made by the user and on the products they’ve viewed in the past.

The Data Behind Good Recommendations

Every technique above depends on having the right data feeding it. Here’s what to pull from and why each source matters:

• Contextual data. Location, time of day, even the weather. Recommending sunscreen on a hot day or sweaters in winter makes suggestions feel timely rather than generic.
• Past interactions. What a customer has already bought or browsed. This is the backbone of most personalized recommendations.
• Search data. What a shopper types into your search bar, and what they click on afterward. This is the signal most stores completely ignore, and it’s often more reliable than browsing behavior alone because it captures explicit intent rather than passive interest.
• Real-time behavior. What a customer is doing on the current visit: which products they’re hovering over, adding to cart, or spending time on. This lets recommendations react within the same session instead of waiting for historical data to build up.

Combining an on-site search engine with your recommendation logic means these two systems reinforce each other instead of working in isolation. What people search for informs what gets recommended, and what gets clicked in a carousel can, in turn, sharpen future search relevance.

Where to Place Recommendations (and What to Expect From Each Spot)

Doofinder groups these touchpoints into stages of the shopper’s journey, and it’s a useful way to think about placement: what you recommend, and how confidently, should shift as someone moves from just browsing to about to buy.

Discovery

This is where a shopper is still exploring, so lean on broad, low-risk suggestions rather than anything narrowly personalized.

• Homepage. Bestsellers, new arrivals, special offers, and whatever you want to push that week.
• Category page. Best-selling products from that category at the top of the page, guiding the shopper before they even start scrolling.
• Search results. When a search returns few or no exact matches, a recommendation carousel saves the visit instead of leaving the shopper with a dead end.
• 404 page. Error-404 pages can be personalized too. A best-sellers carousel gives a visitor who hit a broken link a reason to stay instead of leaving.

Consideration

• Product page. Similar products, visually similar products, and frequently bought together all belong here. This is typically the highest-traffic, highest-converting spot for recommendations.

Conversion

The shopper has already committed to buying something, so this is where complementary and bundled suggestions earn the most.

• Cart. Frequently bought together or complementary items work well while everything’s still fresh in the shopper’s mind.
• Checkout. Right before checkout, show complementary products to whatever’s already in the cart, without adding enough friction to risk the sale itself.

Retention

• Post-purchase / order confirmation. The customer is already in a buying mindset with no cart friction left. A “complete the look” or “frequently bought together” carousel here can pick up a second sale without any extra ad spend.
• Email. Use email marketing to recommend products by season, or send dynamic, personalized picks: “Recently bought these shoes? Here’s what pairs well with them.”

Search + Assistant

This one cuts across every stage above rather than sitting at a single point in the funnel. Doofinder’s search layer, autocomplete, and AI Assistant all feed and surface recommendations as a shopper types, searches, or asks a question directly, so the same intent signal that powers a great search result can power a smarter recommendation right alongside it.

Real-time pop-ups are worth a mention here too. Exit-intent pop-ups can offer a deal or an alternative right as someone is about to leave, and cart-based pop-ups can suggest an upgrade or add-on the moment something gets added to the basket.

How to Configure Recommendations That Convert

This is the part most articles on this topic skip, because most of them are written by people who’ve never actually built a recommendation engine. Here’s what configuring this well actually looks like.

Recommendation carousels run on Logics, which are the rules that decide what products show up and turn a static widget into automated product recommendations that adjust on their own. Here’s the full set available, what each one does, and how it updates:

Logic	Where it works	What it recommends	How it updates
Most Popular	All pages	Most-visited products site-wide over the last 15 days	Refreshed daily
Best Sellers	All pages	Top-selling products based on your sales data, over the last 15 days	Refreshed daily
Browsing History	All pages	The shopper’s own recent interactions, per session	Live per session
Custom Picks	All pages	A manually defined selection, built with include/exclude rules	Whenever you update the rules
Similar Products	Product pages	Items sharing attributes like title, brand, or category	Based on your feed
Visually Similar	Product pages	Items matching visual features via image recognition	Real time
People Also Viewed	Product pages	Products viewed by shoppers who also viewed this one	Retrains every 7 days
Frequently Bought Together	Product pages	Items commonly purchased in the same order, from the last 90 days of sales data	Refreshed daily

A few things worth knowing before picking a Logic:

• Best Sellers and Frequently Bought Together need sales data configured first, or they won’t be selectable at all.
• People Also Viewed needs a short training period before it displays anything. Enable a fallback to Most Popular during that window so the carousel isn’t empty while it learns.
• Visually Similar only evaluates the exact image shown, so with multiple color variants, it matches the specific one a shopper is looking at, not the parent product.

Filters narrow what a Logic pulls in, from general rules (in-stock only, exclude a category) to IF/THEN conditions (show pricier items on a premium product page, hide winter stock in summer). Same as Viewed is worth calling out: it auto-scopes a carousel to match whatever category a shopper is browsing, so you don’t need a separate carousel per category.

Fallback kicks in when a Logic can’t fill the carousel, swapping in a backup set instead of a half-empty widget.

Dynamic cart carousels adapt in real time to whatever was just added to the basket, using logics like Frequently Bought Together or People Also Viewed.

Most setups allow up to 15 active carousels, 5 to 25 products per carousel, styled and placed with a visual CSS selector rather than a developer.

Measuring Whether Your Recommendations Are Working

None of this matters if you’re not tracking it. At minimum, keep an eye on:

• Impressions, how often a carousel actually loads and gets seen.
• Click-through rate (CTR), the ratio of clicks to impressions.
• Conversion rate (CR), tracked when a click on a recommended product leads to a completed purchase in the same session.

Beyond the basics, test and refine over time:

• A/B test your logics. Compare how a “Best Sellers” carousel performs against a “Frequently Bought Together” one in the same spot, and let the data decide which stays.
• Compare placements. The same logic can perform very differently on a product page versus in the cart. Don’t assume; check.
• Send events to GA4 so recommendation performance sits alongside the rest of your analytics instead of living in a silo.
• Revisit regularly. A recommendation strategy isn’t something you configure once. Sales data shifts, catalogs change, and a logic that converted well last quarter might need a refresh.

Barilliance’s research is a useful outside benchmark here: sessions that included a click on a recommended product showed average order values several times higher than sessions without one. Your own numbers will vary by catalog and traffic, but it’s a reasonable yardstick to check your results against once you have a few months of your own data.