Global e-commerce crossed a quiet threshold between May 2024 and April 2025: the average online order contains 4.52 products, up from a December low of 3.27. Shoppers are consolidating purchases into fewer, larger baskets, driven by shipping economics, bundle promotions, and the trust that comes with established marketplaces. Mobile accounts for roughly 77% of retail site traffic and 68% of orders in the first quarter of 2025. Yet in digital fashion, the average accessory attach rate is approximately 0.4 items per order — meaning six out of ten orders leave the matching piece behind.

Attach rate is the share of a primary product's sales that carry a complementary item: units of product B sold with product A, divided by units of A. It is a more honest measure of cross-sell than the headline AOV lift, because it shows which pairings actually hold rather than how much a discount inflated the basket. Rework puts the benchmark for mature upsell and cross-sell programs at 15–30% conversion on orders, with an achievable AOV lift of 10–25%. In fashion specifically, "shop the look" presentations report 18–25% AOV gains, while accessories — lightweight, high-margin, and designed to be worn together — contribute roughly 29% of total fashion sales. The most common Black Friday tier was 20% off, used by about a third of stores, which is a reminder that most retailers reach for a discount before they reach for a better suggestion.

SignalObserved figure
Global items per order (May 2024–Apr 2025)4.52
Digital fashion accessory attach~0.4 items per order
"Shop the look" AOV lift18–25%
Cross-sell program conversion15–30% of orders
Post-purchase offer conversion13.6%

The interesting evidence is about timing rather than depth. A 2025 Rebuy case study found post-purchase offers accounted for 23% of additional revenue at a 13.6% conversion rate, without heavy discounting. The buyer had already committed to the primary item and was not in a comparison mindset, so a well-matched add-on felt like finishing a decision rather than starting a new one. Occasion-framed suggestions — "complete the look," "brunch ready" — outperform algorithm-only "customers also bought" modules because they match the context the shopper is already imagining. Relevance beats a coupon: one more item at full price is worth more than two items at 20% off.

Failure modes are consistent and easy to list. Interrupting checkout with a modal adds friction at the worst moment. Irrelevant recommendations — printer ink next to a sofa — erode trust in every future suggestion, so relevance thresholds matter more than coverage. More than three or four options overwhelm and depress conversion on all of them. And most cross-sell modules are designed for desktop while 60–70% of traffic is mobile, where a poorly fitted carousel is invisible. A fifth caution that rarely appears in growth decks: if the return rate on upsold products runs well above baseline, the engine is overselling things buyers do not actually want, and the AOV gain is borrowed from the returns budget.

This is exactly the kind of surface problem an embedded runtime is built for. A mini-program has the state that a static page lacks: what was viewed, what was abandoned, what was ordered last time, and what the current basket contains. Attach suggestions can be computed from that context at the moment of decision, rendered inline rather than as an interruption, and carried into a post-purchase step that keeps the same session alive. On mobile — where nearly all the traffic is — the runtime gives a consistent component instead of a desktop module that breaks at small widths. Frequency caps and suppression of already-owned items stop the recommendation from becoming noise, and the same logic can enforce a three-item limit per suggestion block.

The merchant-side lesson is to instrument attach rate as a first-class metric rather than a footnote under AOV. Track it by pair, by device, and by placement, then watch the return rate on the attached item alongside it. A pair that raises AOV and return rate together is destroying margin while looking good on a growth chart. A pair that attaches at full price and stays attached is the real compounding asset.

For a host platform, the strategic value is that attach behavior is portable. A pairing that works for one merchant in the catalog can be offered as a primitive to others in the same category, and the runtime supplies the contextual data that makes it relevant instead of random. The host earns a share of a fuller basket while the merchant keeps the price — which is the arithmetic that makes recommendation-led growth preferable to promotion-led growth when the goal is profit rather than volume.

Sources