For a long time my checkout offered the same upsell on every product. It was a high-margin accessory I liked, and I dropped it onto the confirmation step for everything. Conversion on it was fine on the one product it actually made sense for, and near zero everywhere else. The problem was not the copy. It was that I had paired one product with every other product.
The pairing problem is a data problem
Good cross-sell is about fit, not margin. A phone case belongs with a phone, not with a book. A refill belongs with the device that uses it. I had the order history that proved this — people who bought A also bought B — but I never turned that into a per-product pairing list. Doing it by hand for a few hundred SKUs is a week of spreadsheet work, and the list goes stale the moment I add inventory.
The deeper mistake was treating upsell and cross-sell as one thing. An upsell is the next tier of the same product; a cross-sell is the thing that complements it. They belong in different places in the funnel, and lumping them together is why my single suggestion felt random to shoppers.
What I changed in the flow
I now generate pairings per product instead of one global suggestion. UpsellIdeas takes a product and returns 3–5 cross-sell and upsell pairings, each with the reason it fits and ready-to-use sample copy. The "reason per pair" is the part I underestimated: it forces the suggestion to be defensible. If I cannot say why a pairing fits, it probably doesn't, and showing the reason in my admin also helps a teammate trust the suggestion.
The output gives me two shapes: cross-sell pairings (buy this with that) and an upsell ladder (the next tier up). That distinction changed how I place them — cross-sell near the cart, upsell on the product page where the upgrade is still a click away. If you want to see the shape of the output first, the UpsellIdeas page shows the pairing and reason format I rely on.
Where it stops being helpful
A generator can only pair from what you tell it about the products. If my product descriptions are vague ("great gadget"), the reasons come back vague too. I had to clean up the seed input first.
It also won't know my margin rules or my stock. I still filter pairings by what is actually in stock and what I am willing to discount. The tool is a suggestion engine, not a pricing system, and I treat its sample copy as a draft I adapt to my voice.
A caveat about catalog data
One more thing I learned: pairings are only as varied as the catalog data you feed in. Early on I passed it two products that were nearly identical and got near-identical suggestions back, which taught me nothing. Giving it real attribute differences — material, use case, price tier — is what makes the pairs distinct and useful. The tool reflects the structure of your catalog; it won't invent differences your products don't have. I now keep the seed attributes clean specifically so the output stays honest, and I drop any pair that reads as a stretch rather than a fit.
FAQ
Does it write the checkout copy for me?
It gives sample copy per pairing that you can drop in or adapt. I treat it as a first draft, then match the tone to the rest of my store.
How many pairings do I get per product?
Three to five per run, split across cross-sell and upsell, each with a one-line reason it fits.
Can it use my own sales data?
It works from the product details you provide. If you want data-driven pairs, feed it the attributes that actually drive your repeats; it won't pull order history on its own.
Is this a guarantee of higher revenue?
No. Better pairings remove a known gap, but conversion still depends on price, stock, and the page itself. It is decision support, not a revenue promise.
Top comments (0)