Most D2C product pages have a "you may also like" row somewhere below the fold. Almost none of them move the order. Not because cross-sell doesn't work — because the row is decoration, not logic. It shows the same four products to every visitor, usually picked once by whoever built the theme, and never touched again.
A cross-sell that actually raises average order value has to answer one question correctly: given what this specific person just put in their cart, what is the single most likely next thing they'd add? That answer changes by product, by cart contents, and often by time of year — a static row can't give it.
Why "customers also bought" alone isn't enough
Generic "customers also bought" widgets are popularity-ranked, not relevance-ranked. They surface whatever sells the most store-wide, which is usually your bestseller — already in most carts, or already rejected by this visitor for a reason the widget doesn't know. It's the same recommendation for a first-time visitor and a five-time repeat buyer, which is exactly backwards: those two people need completely different suggestions.
Three signals that actually predict a real cross-sell
- Co-purchase pairing, not co-popularity — which specific products actually get bought together, weighted by how often, not which two products are individually best-selling.
- Cart-state awareness — a starter kit in the cart should surface refills or the next tier up; a single refill in the cart should surface the accessory that makes it last longer, not another refill.
- Order history for returning customers— a second-time buyer has already told you what they like. Recommend the adjacent product in that category, not a random bestseller from a category they've never touched.
What this looks like on a real catalog
For a food-and-wellness catalog — honey, spice blends, pickles, herbal teas — the highest-converting cross-sell isn't "other honey." It's the specific pairing that shows up again and again in actual order data: a jar of honey with the ginger-turmeric tea it's usually stirred into, or a pickle with the specific spice blend used in the same regional cuisine. Those pairings aren't guessable from the outside — they only show up once you look at what real carts actually contain.
The fix, in order
Pull twelve months of order data and build a real co-purchase matrix — which SKUs appear together more often than chance would predict. Replace the static "you may also like" block with cart-state-aware logic driven by that matrix. Then split it by customer type: first-time visitors get the highest-confidence store-wide pairing, returning customers get a pairing from their own purchase category. Re-measure AOV after a full order cycle, not after a week — cross-sell impact shows up over repeat purchases, not the first one.
This is the exact process behind our Get Chosen pillar — read the full breakdown on the Get Chosen page.