In a good shop, somebody says “you will want the right fixings for that wall” and you are grateful, because you had not thought about it and you would have made a second trip.
That is all any of this is meant to be. Somebody who knows the products noticing what you are about to forget, or what you would be better off with. It has acquired a bad name because a great many websites do the opposite: they wait until you have decided, then interrupt you with something irrelevant, and call it increasing average order value.
The distinction is not whether you suggest things. It is whether the suggestion is useful to the person receiving it.
Three different things, often muddled
An upsell is bought instead of the thing they were looking at. A better version of the same item: the larger size, the model with the longer warranty, the thicker gauge. They leave with one product, not two. It only works when the difference is genuinely worth the money and you can say why in a sentence.
A cross-sell is bought as well as. Something that goes alongside what they have already chosen: the fixings, the filter, the batteries, the case. They leave with two things. This is the one most shops under-use, and it is by far the most likely to be appreciated, because it is usually the thing they would otherwise have come back for.
A bundle is a set sold together, usually for less than the parts. It works when the set is the sensible starting point for a beginner, or when the items genuinely belong together, and it falls flat when it is obviously three slow sellers tied to a fast one.
Most shops have a “related products” panel doing all three badly at once, filled by whatever the platform decided was similar.
The test worth applying
Before any suggestion goes on the page, ask whether a decent member of staff would say it out loud.
They would say “that needs a bracket, and it is not included”. They would say “for ten pounds more you get the version that fits the standard rail, which is what most people want”. They would not say “customers who bought this also bought a garden hose”, because no person has ever said that to another person.
That test filters out most of the damage on its own, and it is the reason automatic recommendations are so often worse than a short list you chose yourself. A platform can see what sold together. It cannot see that one of those pairings was a coincidence and the other is the reason the product works.
Where each one belongs
On the product page, two things belong. The upsell, shown as a plain comparison between the versions so somebody can see what the extra money actually buys, rather than as a nudge towards the dearer one. And the cross-sell that makes the purchase work: the consumable, the fitting, the accessory without which it will disappoint. Put that one near the add to basket button, phrased as a question the customer would have asked.
In the basket, small additions do well. Somebody who has decided to buy is comfortable adding a five pound item to a sixty pound order, and this is the natural home for the “do not forget” prompts.
After the order, in the confirmation or a follow up, suggest the things that come next rather than the things they should have bought. Refills, the matching item, the service. This is the most under-used placement of the three and the least intrusive, because nothing is standing between anybody and a purchase.
In the checkout, nothing. More on that below.
The opportunity most shops miss entirely is the second order. A customer who has just bought is the warmest audience you will ever have. The email confirming their order is opened far more often than any campaign you will send, and it is a perfectly reasonable place to mention the thing that goes with what they just chose.
Where it quietly does damage
Anywhere inside the checkout. Once somebody is entering card details, every additional decision is a chance to stop. Suggestions belong before that point, not during it.
Interrupting with a pop-up. A panel on the page is an offer. A modal that has to be dismissed is an obstacle, and on a phone the close button is often the smallest thing on the screen.
Suggesting what they already chose. If somebody is looking at the large one, do not offer them the large one. It happens constantly with automatic recommendations and it makes a shop look as though nobody is home.
Too many at once. Three good suggestions beat twelve. A long grid of vaguely related products reads as filler, and it pushes the things that matter, like your delivery information and reviews, further down the page.
Padding out a thin page. If a product page has one photograph and a borrowed description, a recommendation panel is not what it needs. The photographs and the words come first, because a suggestion only lands once somebody is convinced about the thing they came for. That groundwork is covered in photographs and descriptions that do the selling for you.
Bundles that people actually want
The bundles that work tend to be one of three things.
The starter set, where somebody new to the activity does not know what else they need and would rather you decided. The genuine pairing, where the two items are almost always used together. And the refill or multi-pack, where buying more is simply cheaper per unit and the customer can see that plainly.
Two things to get right. Show the saving honestly, because an inflated “was” price on a bundle is the fastest way to lose trust, and it strays into territory where the rules about how you present prices matter. And let people buy the parts separately as well, because a bundle that is the only route to one component is a source of quiet irritation.
How to tell whether any of it is working
The number worth watching is not average order value on its own, because you can raise that while making fewer sales overall and end up behind.
Look at two things together. Whether the number of completed orders is holding steady, and whether the value of those orders is going up. If order value climbs while completed orders fall, your suggestions are getting in the way rather than helping.
It is also worth looking at returns. A suggestion that pushes people into the wrong version shows up weeks later as a parcel coming back, and nobody counts that against the panel that caused it.
Where to start
Take your ten best sellers and, for each, write down the one thing a knowledgeable member of staff would mention. Not three things. One. Put that on the product page, next to the button, in plain language.
Then look at what your platform is currently recommending on those same pages, and turn off anything you would be embarrassed to say out loud.
That is usually a morning’s work, it costs nothing, and it tends to do more than any amount of tuning a recommendation engine. Nobody knows which suggestion belongs with which product better than you do.
The part worth handing over is what sits underneath: where the suggestions appear, what the shop does automatically when nobody has chosen anything, and making sure none of it turns up in the middle of the checkout. That is a one-off piece of setup rather than a standing job, and it is the difference between a shop that suggests well everywhere and one that does it properly on ten products.
If your product pages are currently recommending things you would never say out loud, get in touch, or read more about how we build and look after an online shop.