Product Affinity: Analysis, Metrics and Cross-Sell KPIs
- Measure product affinity from order data with three numbers: support, confidence and lift.
- Rank product pairs by lift, not by confidence, because a bestseller looks like a match for everything.
- Set a minimum order count for each pair before you act, so small samples do not drive your recommendations.
- Use product relationships (complementary, substitute, alternative) to decide what to show beside each product.
- Apply the findings to cross-sell recommendations, bundles, product placement and segment-based ads.
- Track attach rate, items per order, average order value and margin to see whether affinity insights pay.
- Expect a measured gain: in one retail experiment, recommendations raised product sales by 11% [UF Warrington College of Business, 2018].
Product affinity is the tendency of customers to buy certain products together. A shopper who adds a camera to the cart often adds a camera bag or a lens, and your order history shows how often. Measure those patterns and you know what to recommend, what to bundle and whom to target.
This guide covers product affinity vs product relationships, how to run a product affinity analysis, its three metrics and the cross-sell KPIs that show whether it pays. Omniconvert Commander (formerly Nexus), the AI eCommerce growth engine, shows which customers are most valuable.
What is product affinity?
Amazon's "Frequently bought together" block is the familiar example. When you look at a product, it shows the items other shoppers added with it, and each accepted suggestion raises the average order value (AOV). A product recommendation engine runs on the same idea: what people bought together before predicts what the next shopper will want.
The reason to care is cost. New customers are expensive, and a store grows when customers come back and spend more. Product affinity lets you shape offers around real buying habits rather than guesses, and it supports tighter ad targeting: researchers at Kennesaw State University built a product affinity segmentation framework that groups customers by the products they buy together, so each pairing reaches the customer segments that buy it. The payoff is better retention, higher gross profit and a higher customer lifetime value (CLV).
Product affinity vs product relationship
Relationships describe what could go together; affinity data tells you which links customers act on. Someone who adds training weights to the cart may want resistance bands or a yoga mat (complements), or may want to compare other brands of weights (substitutes).
| Relationship type | What it means | Example | Where to use it |
|---|---|---|---|
| Complementary | Products used together | Camera and camera bag | Cart and checkout cross-sells, bundles |
| Substitute (replacement) | Products that do the same job | Two brands of training weights | Product page "similar items", out-of-stock pages |
| Alternative (competitor) | Rival options at a different price or spec | Entry-level and pro camera bodies | Upsells and comparison blocks |
How to run a product affinity analysis
The method comes from association rule mining, which Agrawal, Imieliński and Swami introduced for supermarket transaction data [ACM SIGMOD, 1993]. You need clean order data, not a data science team.
-
Gather your dataCustomer transactions are the primary source, and you already own them. Pull order history with order IDs, products, quantities, dates and customer IDs, plus purchase frequency and order values. A customer intelligence platform keeps this data in one place.
-
Analyze purchase patternsCount how often each product and each pair appears in orders over a set period, including pairs bought in sequence and customer groups with similar baskets.
-
Read support, confidence and liftCalculate the three metrics for each pair, drop pairs below a minimum order count, and rank the rest by lift. The next section shows the formulas.
-
Apply and test the findingsTurn the strongest pairs into recommendations, bundles, placement changes or ad audiences, and test each change before you roll it out.
Support, confidence and lift: the product affinity metrics
| Metric | Formula | Question it answers | How to read it |
|---|---|---|---|
| Support | Orders with A and B ÷ all orders | How popular is this pair? | Too low means too few orders to trust |
| Confidence (A to B) | Orders with A and B ÷ orders with A | When people buy A, how often do they add B? | High for any bestseller, so never use it alone |
| Lift | Confidence (A to B) ÷ share of orders with B | Does buying A make B more likely? | Above 1: affinity. About 1: no link. Below 1: bought apart |
Worked example: a store has 10,000 orders. A camera appears in 800 of them, a camera bag in 500, and both together in 240. Support = 240 ÷ 10,000 = 2.4%. Confidence (camera to bag) = 240 ÷ 800 = 30%. The bag appears in 5% of all orders, so lift = 30% ÷ 5% = 6. Camera buyers are six times more likely than the average shopper to buy the bag. Now compare batteries, which appear in 2,000 orders (20%) and with the camera in 168. Confidence = 168 ÷ 800 = 21%, lift = 21% ÷ 20% = 1.05. Batteries sell well with cameras only because they sell well with everything.
That second pair is where many teams go wrong. They rank pairs by confidence because it reads like a clear promise ("21% of camera buyers add batteries"), so the list fills with bestsellers customers buy anyway. The cross-sell slot shows batteries, the attach rate looks healthy, and nothing changes, because those batteries would have sold without the prompt. If the team also discounts the pair as a bundle, it gives away margin on purchases it already had. Rank by lift, and set a minimum order count so a pair seen in a dozen orders cannot top the list.
How to use product affinity analysis
- Cross-selling and upselling: when a customer adds a product to the cart, show its strongest complement or a better version. Keep it calm: an aggressive checkout prompt can cause cart abandonment. See these cross-selling and upselling strategies.
- Product bundles: when repeat customers keep buying the same items together, offer them as a set. The product bundling guide covers bundle types and pricing.
- Product placement: place high-affinity products near each other on category and product pages.
- Segment-based ads: target each pair at the customer segments that buy it, and check the margin of each pair against your most profitable products before you spend on it.
Two variants of the analysis add depth. Predictive market basket analysis looks at items bought in sequence, so you can time the next offer. Differential market basket analysis compares the same rules across stores, sales channels, customer groups or time periods, and when a pair sells well in one place but not another, you look for the cause, as TechTarget defines the two types [TechTarget]. It is often a weak spot in the customer journey.
Test each change before it goes live. AliveCor used Omniconvert to run a structured A/B testing program and achieved a 21 percent lift in conversion rate, a 5 percent lift in revenue per visitor, and 94 percent statistical relevance across their experiments [Omniconvert, AliveCor case study]. That program did not test product affinity itself, but the lesson applies: an affinity-based offer is a hypothesis until a test shows it earns more per visitor.
Cross-sell KPIs to track product affinity results
Measure these before and after each change:
- Attach rate: the share of orders with product A that also include product B. It is confidence, tracked over time.
- Items per order: average units per order. It rises when cross-sells work.
- Average order value: revenue ÷ number of orders. Our guide to increase average order value covers more levers.
- Recommendation revenue: the share of revenue from products added through a recommendation or bundle.
- Gross margin per order: revenue minus cost of goods sold, so a discounted bundle does not hide a loss. See how to calculate eCommerce profit margin.
- Repeat purchase rate and CLV: whether cross-sell buyers come back and spend more. See the repeat purchase rate.
- Cart abandonment rate: a guardrail. If shopping cart abandonment rises after you add a checkout offer, the offer costs more than it earns.
Keep your expectations realistic. In a large experiment on a fashion retailer's website with more than two million product page views, product recommendations raised overall product sales by 11%, well below popular industry claims, researchers at the University of Florida and Wharton found [UF Warrington College of Business, 2018]. For wider store metrics, see these important eCommerce metrics and the eCommerce metrics that increase user engagement.
Omniconvert Commander adds the customer view: it calculates CLV and True Profit per order and per customer, and pushes RFM segments directly to Meta Ads, Google Ads and Klaviyo.
Product affinity FAQs
Product affinity is the tendency of customers to buy certain products together, such as a camera with a camera bag. You measure it from order data with three metrics: support, confidence and lift. Online stores use it to choose cross-sell recommendations, bundles and product placement.
Yes. Product affinity analysis and market basket analysis are two names for the same method: mining past orders for products that appear together more often than chance. Market basket analysis is the older retail and data science term, and product affinity analysis is the term most eCommerce teams use.
Count the orders that contain product A, product B and both. Support is the orders with both divided by all orders. Confidence is the orders with both divided by the orders with A. Lift is confidence divided by the share of all orders that contain B. A lift above 1 means the pair sells together more often than chance.
Any lift above 1 shows a positive affinity, and the higher it is, the stronger the link. Treat values close to 1 as no real link, and check support too: a lift of 8 on a pair that appears in 12 orders is noise, not a pattern. Set a minimum order count before you rank pairs by lift.
Track attach rate, items per order, average order value, the share of revenue from recommended products and gross margin per order. Over the longer term, watch repeat purchase rate and customer lifetime value, and keep cart abandonment rate as a guardrail so cross-sell offers do not push shoppers out of checkout.
Product affinity is the evidence: data that shows which products customers already buy together. Product bundling is one action you can take on that evidence, by selling those products as a single offer, often at a combined price. Affinity analysis also feeds recommendations, product placement and ad targeting, not only bundles.
Omniconvert Commander brings your order and customer data together, calculates customer lifetime value and True Profit per order and per customer, and groups customers into RFM segments. Omniconvert Commander shows which customers are most valuable, so you can run your affinity analysis on the orders of those segments and push them directly to Meta Ads, Google Ads and Klaviyo for targeted cross-sell campaigns.
Export the last 12 months of orders and count every product pair that appears in at least 50 orders. Calculate support, confidence and lift for each, then keep the three pairs with the highest lift. Put the strongest pair in your cart recommendation slot, test it against a control, and judge it on attach rate, average order value and gross margin per order, not on clicks.
Then ask who buys those pairs. If your highest-value segments buy them most, build your next cross-sell campaign for those segments first. Omniconvert Commander calculates customer lifetime value per customer and pushes RFM segments directly to Meta Ads, Google Ads and Klaviyo.
If you want a team to run this analysis for you, see Omniconvert's managed services.
Point your cross-sells at the customers who pay
Omniconvert Commander calculates customer lifetime value and True Profit per order and per customer, groups customers into RFM segments and pushes them directly to Meta Ads, Google Ads and Klaviyo, so product pairs reach the segments worth targeting.