Recommending a case, screen protector, and wireless charger to a customer who just added a phone to their cart makes shopping easier for them and grows the cart value at the same time. A well-built product recommendation engine is essentially the "would you also like this?" question, asked automatically and personalized for every customer. In this article, we look at how to grow cross-sell without turning it into sales pressure, using recommendations that are genuinely useful.
Why cross-selling grows the average cart value
The customer has already made their purchase decision and is in the cart with the intent to pay; no other point on the site is a better moment to introduce a new product. A relevant product shown at the right time raises the cart total without any extra advertising cost. What's more, when the recommendation is a genuinely useful complement, the customer perceives it as a helpful service rather than sales pressure. Cross-selling shouldn't be considered separately from other cart-growth techniques like upsell; we covered how to design the two together in our upsell and cross-sell guide.
The key word here is "relevant." Squeezing an unrelated product into the cart damages trust; that's why a proper recommendation engine has to run on data, not randomness.
Recommendation types: which logic works in which situation
Product recommendation isn't a single method; different recommendation logics serve different purposes:
- Frequently bought together: the "most people who bought this also bought that" logic; it's derived from past order data and typically delivers the highest conversion.
- Complementary products: comes from product logic rather than category relationships (phone-case, printer-ink, for example); can also be defined manually by a category manager.
- Similar products: alternatives that meet the same need at a different price or with different features; usually shown on the product page rather than in the cart.
- Personalized recommendations: a customer-specific list generated from the customer's past purchases and browsing behavior.
The type that works best at the cart stage is generally "frequently bought together" and manually defined complementary products, because both speak directly to the need the customer already has in mind at that moment.
Comparing the algorithmic approaches behind these recommendation types makes it clearer which one to use and when:
| Method | How it works | When it works best |
|---|---|---|
| Rule-based | Fixed pairings manually defined by the category manager (phone → case) | For new products and cold starts |
| Collaborative filtering | "People who bought this also bought that" — learns from past order data | Once order volume has grown enough |
| Content-based filtering | Builds similarity from product attributes such as category, color, and brand | When order history is thin but product data is rich |
The data that feeds the recommendation engine
A recommendation engine is only as good as the data it uses. Three core data sources determine how accurate the recommendations are:
- Order history: which products frequently show up together in the same cart is the most reliable signal.
- Browsing behavior: which products the customer viewed in the same session captures trends that haven't shown up in order data yet.
- Category and attribute relationships: attributes on product cards such as color, size, and brand generate a starting recommendation for newly added products that don't have order history yet.
Because a newly added product has no order history, the recommendation engine may not surface it at first; this "cold start" problem needs to be overcome with attribute-based matching or rules manually added by the category manager.
"The best recommendation is the one that doesn't feel like a sale; the customer should see it as a convenience, not a reminder."
Placement and dosage: where, and how many products to show
Even accurate recommendations lose their impact when shown in the wrong place or in too great a number. Showing two or three products on the cart page under a clear but non-pushy heading like "would you like to add this to your cart?" is enough. A long list of eight to ten recommended products interrupts the customer's checkout flow and scatters their attention.
Prices should be clearly visible on recommendation cards, and adding to cart should take a single tap; sending the customer to a separate page for the recommendation kills the instant advantage cross-selling provides. This principle matters even more on mobile: the recommendation area should be presented as a short, scrollable strip that doesn't clutter the cart on a small screen.
Strengthening recommendations with a price threshold and incentive
Pairing the recommendation with a concrete threshold like "add X more for free shipping" noticeably increases the impact of cross-selling. If the customer is already trying to push their cart total toward a threshold, the likelihood of purchase rises when the recommended product serves that goal. Just keep the threshold realistic; setting a target far above the cart total discourages rather than motivates. Presenting the recommendation that closes the gap as a ready-made set instead of a single product also works well; we covered this approach in detail in our product bundling article.
Measurement: how to evaluate the recommendation engine
Measuring a recommendation engine's success by "how many people added the recommended product to their cart" alone gives an incomplete picture. What really matters is the extra amount the recommendation adds to the average cart value and the return rate of recommended products; a high return rate means the recommendation logic is off target and needs review. Periodically comparing the recommendation engine across different logics (for example, purely rule-based recommendations versus order-data-driven recommendations) clarifies which approach works better for your brand.
Check your recommendation engine
- Are at most 2-3 products recommended on the cart page?
- Do the recommendations come from "frequently bought together" logic or defined complementary products?
- Is there a "cold start" rule for newly added products?
- Can the recommended product be added to the cart with a single tap?
- Are recommendations presented on mobile without cluttering the cart?
- Do you measure the extra amount the recommendation adds to the average cart value?
- Do you separately track the return rate of recommended products?
Analyzing order history and recommending the right product at the right time means writing separate rules for every single product when done manually, which becomes unsustainable as the product catalog grows. Şimşek Software's product recommendation infrastructure automatically extracts products that sell together from your order data and generates personalized recommendations for your cart and product pages; you just monitor the results and fine-tune the rules when needed.