How cross-selling and upselling work
Cross-selling suggests something that goes with what the shopper has chosen: a case with a phone, a memory card with a camera, insoles with walking boots. Upselling suggests a better version of the same thing: more storage, a larger size, a longer warranty, the next tier of a service. The two are usually grouped together because they appear in the same places, but they answer different questions. One adds a need the customer had not thought about; the other upgrades the need they already have.
Placement decides whether either works. On the product detail page a suggestion still feels like shopping. In the basket it feels like a reminder. On the order confirmation it feels like a separate, low-risk offer. Suggestions inside the payment step almost always hurt, because the shopper has decided and any new decision is a chance to stop.
Why cross-selling and upselling matter
They raise average order value without buying any more traffic, which is the cheapest revenue a shop has. The visitor is already there and already converting; the only question is whether the shop is helping them buy the whole solution or just part of it.
Done honestly they also reduce returns and complaints. A customer who buys a printer without ink, or books a trek without the right insurance, is a customer who will be disappointed later. Pointing out the missing piece is service as much as selling.
Where cross-selling and upselling go wrong
Irrelevance is the usual failure. Recommendation blocks filled with whatever is popular, rather than with what genuinely pairs with the chosen item, teach shoppers to ignore that part of the page for good. Relevance beats volume: one well-chosen suggestion outperforms a wall of them.
Aggression is the other. Interstitials, pre-ticked add-ons and upgrades that are hard to decline lift the basket in the short run while costing trust, refunds and repeat purchases later. Pre-selected extras in particular sit close to dark patterns, and customers who notice them rarely come back.
Getting it right
Start from real pairings rather than from an algorithm. Look at what customers already buy together in your own order history, ask whoever handles support what people come back for a week later, and build the first set of rules by hand. Automated recommendations are worth adding once there is enough order history for them to learn from, not before.
Keep every upgrade explicable in a sentence — what it does and why it costs more — and make declining it one obvious click. Then measure the whole picture: basket value, order count, margin and returns together, because an add-on that lifts the basket and comes back next week has cost you twice.