Key takeaways
- Most recommendation rails are wallpaper: batch-computed lookalikes that ignore what the shopper is doing right now.
- AOV moves when recommendations are live (re-ranked from this session's signals), outfit-aware, and fit-aware.
- Fit-awareness is the fashion-specific edge: suggesting items in sizes the shopper can actually wear, powered by the same profile Smart Sizing builds.
- Following intent as it forms beats predicting it from history, shoppers pivot mid-session, and static rails miss the pivot.
- Stock and margin signals keep the rail commercially honest: promote what is available and worth promoting.
Recommendations are not all created equal
Most stores show recommendations. Few show ones that change behaviour. 'You may also like' rails computed in a nightly batch job, showing the same lookalikes to everyone, are wallpaper, not merchandising, scrolled past, never clicked, quietly renting some of the most valuable real estate on the page.
The recommendations that move average order value react to what the shopper is doing right now, not to a profile from last month. The difference is architectural: batch systems answer 'what do people like this person tend to buy', while live systems answer 'what does this person want in this session', and only the second question grows a basket.
What actually lifts AOV
Smart Recommendations re-rank live from shopper signals: what they viewed, dwelled on, added, and skipped this session. The rail reflects intent as it forms. Skipping is as informative as clicking, five passed-over floral dresses is a preference statement no purchase history contains.
The lift comes from relevance and from completing the look. Show the trousers that go with the jacket they are considering, in a size you know fits them, and the basket grows naturally, an outfit is a reason to add, whereas a lookalike is usually a reason to substitute. Substitution shuffles revenue; completion adds it.
Signals that matter in fashion
- Style affinity inferred from the current session, not just past purchases.
- Outfit completion: pieces that genuinely pair with the item in view.
- Fit-aware suggestions, so recommendations come in sizes the shopper can actually wear.
- Stock and price, so the rail promotes what is available and worth promoting.
Fit-awareness: the fashion-specific edge
Generic recommendation engines treat fashion like books: items and affinities. But a perfect stylistic match in a size the shopper cannot wear is worse than irrelevant, it is a disappointment you engineered. Fit-aware recommendations filter and rank by wearability, using the same body profile that Smart Sizing builds from a 30-second questionnaire or two-photo scan.
This is where a suite beats point tools: one profile powers the size answer on the product page and the wearability filter on the rail. And because sized-in recommendations carry less risk, they feed the same economics as sizing itself, fewer of the returns that run at 30% across the industry, more of the confidence that lifts assisted-session conversion 15 to 25%.
Why live beats static
A shopper's intent shifts within a single visit. They came for a coat and got curious about boots; they started at full price and drifted to the sale. Static recommendations miss the pivot and keep selling the coat. Live re-ranking follows the shopper through it, updating the rail as the session's story unfolds.
Following the shopper, rather than predicting them from history, is what turns a recommendation rail from decoration into a genuine driver of basket size. History tells you who someone was; the session tells you what they want now. AOV lives in the second answer.
How do product recommendations increase average order value?
By completing outfits rather than suggesting substitutes: pieces that pair with the item in view give the shopper a reason to add to the basket instead of swapping one item for another. Live session signals and fit-awareness make those additions relevant and wearable.
What makes fashion recommendations different from other e-commerce?
Fit. A stylistically perfect suggestion in an unavailable or unwearable size is a dead recommendation. Fashion engines need body-profile awareness, outfit logic and stock-per-size signals that generic engines lack.
Why are real-time recommendations better than batch-computed ones?
Because intent forms and pivots within a single session, the shopper who came for a coat and got curious about boots. Batch systems recommend from history; live systems re-rank from what the shopper is viewing, dwelling on and skipping right now.
Should recommendations account for stock levels?
Yes, promoting an item that is out of stock in the shopper's size wastes the rail and erodes trust. Commercially honest recommendations weight availability and margin alongside relevance.
Conclusion
Recommendations move AOV when they are relevant, live, and fit-aware. Re-rank from real-time signals, complete the outfit, and only suggest what the shopper can buy and wear. That is the difference between a rail people scroll past and one they buy from.
