Every year, succeeding in ecommerce gets harder. Traffic costs more, product catalogs keep expanding, and shoppers have more alternatives than ever. Users increasingly visit websites with one goal in mind — to find the right product, compare it with others, and make a purchase. If a customer gets lost in a massive catalog and doesn't see relevant offers, they close the tab and head to a competitor.

In this environment, the businesses that win are the ones that make purchasing as effortless as possible. That's exactly where AI-powered product recommendations come in. Because recommendations are no longer just static blocks labeled "People also buy." Modern algorithms analyze thousands of signals, account for behavioral history and page context, and select products for each individual user at the precise right moment. When implemented correctly, these tools directly impact key metrics.

In this guide, we'll cover where to place recommendation blocks on your website, how to customize algorithms to match business goals, and how these tools are already driving sales growth for leading ecommerce brands.

How Modern Recommendations Work: From Basic Rules to LLMs

Product recommendations can be grouped based on how they work. The first type includes manual configuration and simple rules. Here, a category manager manually sets up blocks, or simple static rules are defined — for example, "for products in group A, recommend products from group B." This approach is the least effective and the hardest to scale.

Algorithmic systems deliver far better results. They automatically analyze data — product attributes, user preferences, and behavior — and generate recommendations based on that analysis. Despite their overall effectiveness, these approaches still have limitations in analytical depth.

For instance, products are typically analyzed using basic attributes: product ID, name, and broad category. As a result, algorithms can miss the mark — recommending another washing machine instead of detergent, or ignoring newly added products because they don't yet have any click history.

The latest generation of recommendation systems, built on what's known as transformer architecture, has a structural advantage. Rather than analyzing individual events, they process the entire sequence of a user's actions. The system reads a customer's session, capturing deeper context and genuine intent.

But even that isn't the final breakthrough for ecommerce. The best results right now come from combining these algorithms with large language models (LLMs). AI can now understand the "meaning" of a product from its description and attributes.

What does this mean for businesses in practice?

• Solving the cold start problem. When you upload a new collection, the system no longer has to wait weeks for purchase data to accumulate. LLM analyzes the product description and immediately begins recommending new items to relevant users.
• Automatic intelligent categorization. AI independently identifies logical, functional connections between products, freeing category managers from tedious manual work.

Take a broad catalog category like "Toothbrushes." A classic algorithm would simply suggest other toothbrushes as alternatives. An LLM understands context and builds ideal cross-sell pairings: for an electric toothbrush, it automatically pulls in compatible replacement heads or a travel case.

It's precisely this semantic understanding of products that enabled the Foxtrot retailer to increase accessory sales by 1.5x and boost the conversion rate of its "Frequently bought together" block by 39% — without any manual intervention. AI started "thinking" about the product catalog the way a live sales associate would.

Where to Place Recommendations on a Website: Key Pages and Use Cases

Now let’s move to a purely practical question: where exactly should you display these smart blocks so they drive results rather than just take up space? To get results, keep one thing in mind: your recommendations must align with the user's intent at every stage of their path to purchase.

Homepage: Navigation for Cold Traffic

The homepage frequently attracts traffic with no clearly defined intent — people arriving from display campaigns, casually browsing the catalog, or returning after a long break.

Here, recommendations function as a smart navigator. The goal is to capture attention, prevent the user from leaving the site, and help them quickly find what they’re looking for.

Which algorithms to use:

• Personalized recommendations — the safest bet when a user is logged in or the system has already recorded their previous sessions.
• Bestsellers or popular items — for new or anonymous visitors. To avoid overloading the homepage, these can be combined with personalized selections in a single smart block.

Custom blocks also work well here — for example, "New arrivals" or "Sale items" to promote business-priority products. Another interesting option is a "Products of the week" block. Instead of submitting a task to developers every week to update the homepage, a marketer can spend five minutes setting manual rules or defining auto-rotation logic through the platform's UI.

Category Page: Helping Users Choose

The situation changes here. The user has already made a first choice — say, they've navigated to "Sports Footwear" or "Board Games" — but they're still looking at hundreds of options. The risk of choice fatigue sets in.

A recommendation block on this page acts as a filter, narrowing the selection and focusing attention — accelerating the click to a specific product page.

Which algorithms to use:

• Personalized recommendations from the current category. The primary algorithm for this page. These are the same user-data-driven personalized selections, scoped specifically to products within that category.
• If the user isn't logged in, bestsellers or popular items from the category are the appropriate fallback.

Product Page: Cross-Sell and Alternatives

This page is the highest-purchase-intent touchpoint on the site, where the user makes the final decision — whether to buy or not. Since this page generates the largest share of revenue, it's worth paying special attention to its setup and the quality of product attributes in the product feed.

On the product page, recommendations serve two main purposes:

1. Confirming the choice and driving upsell. A "Frequently bought together" block helps customers logically complete their order with useful accessories or build a set right away.
2. Offering an alternative. If a customer is still on the fence about price, color, brand, or specific features, a "Similar products" block provides relevant options and keeps them on the site.

Which algorithms to use:

• "Frequently bought together." The ideal cross-sell tool for adding relevant accessories and complementary items to the main product. Place this block as high as possible — ideally directly below the photo and main description.
• "Similar products." A tool for navigating choice, helping retain users when the current product doesn't suit them due to price, color, or being out of stock. This block should be among the first things a visitor sees.

Don't hide recommendations at the bottom of the page — it directly reduces their effectiveness. Blocks need to be easily accessible — for instance, directly below the product image. Users shouldn't have to scroll halfway down the page to find them.

Collection Block or "Complete the Look"

The most effective block for the fashion niche. Instead of making customers hunt for matching pieces, the system presents a ready-made outfit — for example, pairing a selected jacket with a hat and scarf from the same collection.

This significantly boosts ease of choice and engagement — the customer gets not a random assortment, but a complete and cohesive solution. And it substantially increases the number of items per order and average order value (AOV).

Cart: Soft Upsell/Cross-Sell

The final stage before checkout — the user is one step away from purchasing. The priority at the cart stage is: do no harm. Recommendations here should not distract from completing payment or introduce doubt about the main item already in the cart. This is a zone for subtle, almost impulse-driven upselling.

Which algorithms to use:

• "Frequently bought together" and "Accessories." Inexpensive consumables, batteries, or services — like extended warranties — work well here. Related items from other categories can also perform well to expand the order (without making users second-guess what's already in the cart).

Non-Obvious Places for Recommendations

Most marketers focus on the classic recommendation block placements. But the customer journey isn't always linear, and a purchase doesn't always follow the path "homepage → category → product." There are blind spots on websites — pages that drive traffic but don't convert it, simply because relevant offers are missing.

Account Dashboard: Working with Your Most Loyal Customers

Who visits the account dashboard most often? Your most valuable audience segment: loyal customers who have already purchased from you, have an order history, accumulated bonus points, and a high level of trust in the brand. The system has maximum behavioral data on them.

Which algorithms to use: Personalized recommendations based on the user's history work best here. The CTR of a recommendation block on the account dashboard is 2–3x higher than that of any other block on the site.

Blog: Conversions from Search Traffic

The blog is an excellent source of organic SEO traffic. People come here for useful information — say, reading an article titled "How to Care for Your Cat's Coat During Shedding Season" — but they're not always ready to buy right away. If the page ends with nothing but text, you're missing the opportunity to turn a reader into a customer.

Recommendations in the blog work as a native, logical bridge to purchase: reading about cats — here are grooming tools and vitamins right there for you.

Which algorithms to use:

• Products related to the article's topic.
• Personalized recommendations (if the reader already has a browsing history).
• Popular items from the category the article belongs to.

The MasterZoo pet supplies retailer encountered a routine issue. Content managers were manually curating product selections for each article — a process that could take up to 8 hours per piece, totaling around 120 hours per month. After automating this through AI recommendations, preparation time for an article's product block dropped from 8 hours to 5 minutes. But the headline result — smart selections increased sales directly from the blog by 9x.

404 Page: Rescuing a Lost Session

A 404 error page — a broken link or a deleted page — is typically treated as a dead end. The standard user behavior in this situation is to close the site and go back to Google.

But instead of leaving it at "page not found," you can turn this error into a navigation tool and help the user continue their journey.Which algorithms to use:

• If the user already has an interaction history, personalized recommendations are one of the most effective options. They'll see products they've already shown interest in, reducing the urge to leave the site.
• If the user is new, try bestsellers or new arrivals. This saves the session, holds attention, and brings a potential buyer back into the funnel.

Customization: Making AI Work for Your Business Goals

Personalized recommendations aren't just about serving each visitor a unique selection — they're also about tailoring algorithms to a specific business's objectives. Whether it's promoting priority products or adjusting block logic, the system doesn't just determine what a user is interested in; it works in the business's interest.

Even without additional configuration, the system already has baseline rules that make recommendations smart and practical. For example:

• The system accounts for the price segment the user gravitates toward. Premium shoppers won't see low-cost offers, and vice versa — price-sensitive users won't see products outside their budget.
• The system factors in product popularity. Priority goes to "live" products that people are actually buying.
• Out-of-stock products never appear in recommendations. This may seem obvious, but with manual recommendation setups, it's frequently an issue.

When a business has specific needs, these can be addressed through custom rules, for example:

• Boosting priority products. A marketer can artificially increase the display frequency of specific items — high-margin products, private-label goods, or partner-brand merchandise. Budynok Ihrashok, for instance, successfully used a boost to promote charitable coloring books. This allowed the brand to communicate its values directly at the point of decision-making, without compromising the overall relevance of the block.

• Restrictions and exclusions. Sometimes it's critical to NOT show certain products. For example, hiding products from the lowest price tier or sensitive assortment categories. A clear example is the Liki24.com marketplace, where regulatory requirements prohibit recommending prescription medications. A custom logic was implemented: if a customer is browsing exclusively prescription drugs, the system doesn't leave the recommendation block empty. Instead, it displays permitted products from the "universal medicine cabinet" category (bandages, vitamins, etc.) — and even that fallback selection is personalized for a user.
• Attribute-based affinity. The algorithm can be configured to prioritize specific attributes from the product feed. For example, so that alternatives suggest footwear only in the same color, brand, or size as the main product.

Omnichannel: Recommendations Beyond the Website

The customer journey is more than a browser tab. Limiting recommendations to the website alone means leaving a significant share of revenue on the table. Personalization needs to work across every customer touchpoint.

Mobile App: The Big Retail Trend

For many large brands, the mobile app has stopped being just an additional channel and has become the primary point of sale. App users purchase more frequently and leave richer behavioral data. The key advantage is that you don't need to reinvent the logic. The app uses the same algorithms and the same key placement points — homepage, product page, cart — as the web.

The most important element here is synchronization. When web and app data are unified through a CDP, customers get a seamless experience. For example, they browsed sneakers on a laptop, then opened the app on their phone that evening and immediately saw those sneakers — along with relevant additions — in the recommendation block.

Direct Channels: Email, Viber, Push

Direct channels are a powerful environment for integrating product recommendations. They should be used not only in classic trigger sequences — such as "Abandoned Cart," "Abandoned Browse," or a Next Best Offer — but also in regular mass promotional campaigns.

Instead of sending every user the same selection of a few products chosen by a category manager, it's far more effective to build an individual offer for each customer.

Modern platforms make it straightforward to generate dynamic recommendation blocks directly inside messages — and the results are significant.

The Comfy retailer added personalized AI recommendations to its mass promo campaigns and saw +10% to average order value, with 51% of all clicks in the email going to the algorithmically generated product recommendation block. In other words, people are more inclined to click on content tailored to their interests than on generic promotional offers.

The Budynok Ihrashok team ran an A/B test in Viber, splitting the audience into three segments. The first received a classic promo with a single static banner. The second got a banner plus recommendations. The third received only a carousel with AI-generated product recommendations from Yespo. Across all metrics — Open Rate, CTR, Conversion Rate — the recommendations-only campaign won decisively, outperforming the static banner by 20–30%.

---

Today, personalization is the baseline for ecommerce. Without it, the gaps in your sales funnel will be wider than the holes in Maasdam cheese. But simply placing recommendation blocks isn't enough. Modern algorithms have long outgrown simple rules. They work with greater precision, analyze more deeply, and surface connections faster than any manual category review could.

The numbers from market leaders make this clear. What it takes: a clean product feed, correctly configured behavioral data tracking, and the right algorithms matched to each step of the user journey.

If you're seeing sales dip in your online store, users abandoning carts at the final step before payment, or you simply want to free your team from the routine of manually assembling product selections — we have a solution.

Fill out the form below — we'll analyze your current funnel, identify hidden growth opportunities, and walk you through exactly how to implement AI recommendations in your business so they start generating additional revenue from the very first weeks.

Get professional expertise