# How should recommendation ranking weigh margin and stock levels? | KilterMuse Source: https://kiltermuse.com/blog/2026-10-20-margin-inventory-aware-recommendations/ KilterMuse answers October 20, 2026 **Short answer:** The honest model is two stage: relevance decides which products deserve to be considered for this shopper, and business factors order that shortlist. Stores that let margin lead from the start end up recommending profitable irrelevance, which shoppers learn to ignore within a few sessions. ## Why stores reach for the margin lever Every recommendation slot is valuable shelf space, and giving it to a low margin, low stock product when an equally relevant alternative earns more and can actually ship feels like waste. Seasonal stock adds pressure: recommending a product with a warehouse of depth behind it, over one about to sell out, can be the difference between a clean season and a clearance problem. None of that is sinister. Physical shops have always placed profitable, well stocked lines at eye level. The digital version becomes a problem only when the weighting is hidden, heavy handed, or applied where it distorts a promise, such as in search results presented as pure relevance. ## Guardrails that keep trust intact Cap the business influence: a product should need a relevance score within a sensible distance of the top before margin or stock can lift it into view. Never let stock weighting recommend an item that is effectively unavailable in the shopper's size or region, because a recommendation that ends at a dead end teaches shoppers the rail is advertising, not help. Keep sponsored and boosted placements labelled where the law or platform rules expect it, and keep the logic out of surfaces presented as editorial. A best match sort claim is a promise; a recommendations rail is a suggestion, and the two surfaces deserve different rules. ## Inventory signals that help, and one that backfires Depth aware ranking helps most with sizes and variants: recommending the colour with full size availability over the colour broken to two sizes saves shoppers a wasted click and saves the store a frustrated session. End of line stock with genuine depth can be surfaced to relevant shoppers before it becomes a clearance problem, at full price, which is the best outcome a markdown budget never sees. The backfire is scarcity feedback: down weighting low stock items can starve a bestseller in its last units, while up weighting deep stock can flood rails with the products the store most wants to move rather than the ones shoppers most want, and both errors are visible in click data within days if anyone looks. ## Margin data quality decides whether this works Ranking is only as good as the margin feed, and margin built on stale landed costs will promote the wrong products with great confidence. The feed needs the same freshness discipline as price itself: current costs, returns adjusted where the data exists, and a known update rhythm the ranking team can state. Start with a small weight and measure like a scientist: recommendation click and conversion rates, margin per session, and return rates for the test group against control. If margin per session rises while conversion holds, the weighting is earning its place; if conversion slips, the store is taxing relevance and will pay for it in trust. ## Say what the rail is doing, at least internally Document the ranking policy in one page the whole team can read: what factors, in what order, with what caps, and which surfaces they apply to. Undocumented weighting accumulates as individual tweaks until nobody can explain why the store recommends what it recommends, at which point every bad recommendation becomes an argument. Review the policy when the range or season turns over. Weights tuned for a clearance heavy quarter, left running into a new season launch, will quietly bury the new products the business most wants seen. Keep reading: [How does visual search change how shoppers find products?](https://kiltermuse.com/blog/2026-10-20-visual-search-shoppers-photos), [How can recommendations protect discovery of new and niche products?](https://kiltermuse.com/blog/2026-10-20-recommendation-diversity-new-brands-discovery), and [How should product search handle shoppers who describe what they want instead of naming it?](https://kiltermuse.com/blog/2026-10-18-search-shoppers-describe-not-name) cover the nearby cases in this series. ## Frequently asked questions ### Should profit margin influence product recommendations? It can, legitimately, after relevance has set the shortlist. Margin ordering among products a shopper is genuinely likely to want is normal merchandising; letting margin override relevance turns recommendations into advertising shoppers learn to ignore. ### How can stock levels be used in ranking without backfiring? Use depth to favour variants with real availability in the shopper's size and region, and to surface deep stock to relevant shoppers early. Avoid starving strong sellers in their last units or flooding rails with overstock nobody wants. ### What data quality does margin aware ranking need? A margin feed as fresh and honest as the price feed: current landed costs and, where possible, returns adjusted figures. Ranking on stale costs promotes the wrong products confidently, which is worse than not using margin at all. ### How should a store test business weighted ranking? Start with a small capped weight, then compare click, conversion, margin per session, and returns against a control group. Keep the weighting only if margin rises without conversion or trust measures slipping.