When AI Recommends Substitutes for Out of Stock Items
Recommendation engines increasingly suggest a substitute when the wanted item sells out. When that helps, when it annoys, and how to judge the swap.
The out of stock page used to be a dead end with a notify me box. Now an AI layer often steps in with a substitute: a similar size, a different colour, a comparable product. Done well, it rescues a sale. Done badly, it recommends a substitute that shares nothing with the original but a category.
Substitution is a judgement task, and judgement is where these systems vary most. This guide covers what makes a substitute credible, how stores should constrain the engine, and what the results look like when they do not.
Similarity needs more than category matching
A credible substitute matches on the attributes the shopper chose the original for: size, fit, material, capacity, compatibility, or dietary profile, depending on the product. Category level similarity, another shampoo, another lamp, is the weakest form and the one shoppers reject fastest.
Stores should define, per category, which attributes are non negotiable in a substitute and which can flex. An engine constrained by those rules recommends fewer substitutes and far better ones.
Say plainly that it is a substitute
The presentation should be honest: the original is out of stock, this is the closest alternative, and here is how it differs. Burying the substitution, or styling the substitute to look like the original product page result, trades a short term click for long term distrust.
Differences deserve specifics. A different size, a changed material, a lower capacity: name the one or two differences that matter in text beside the recommendation, so the shopper judges with facts instead of discovering them on delivery.
Keep the original routes alive
A substitute offer should sit alongside, never instead of, the classic routes: notify me, check other locations, or view the full category. Shoppers loyal to the exact item exist, and forcing them through a substitute funnel reads as the store not stocking what it advertised.
Track which route shoppers take. A healthy pattern shows substitutes chosen by some, notifications by others. A pattern where substitutes dominate usually means the other routes were hidden, not that the substitutes were loved.
Measure returns on substituted orders
The honest test of substitution quality is the return rate on substituted orders compared with normal orders. A substitute that converts and returns at twice the normal rate is not rescuing revenue, it is borrowing it from the returns budget.
Review the worst substitute pairs monthly and feed them back as exclusions. Recommendation systems improve fastest when the store curates their failures, not only their training data.
Related reading
If this topic is useful, these earlier guides go deeper on connected questions: How AI Search Summaries Describe Your Returns Policy, When AI Chatbots Give Sizing and Fit Advice, Accuracy Gets Tested, and When AI Search Answers Send Shoppers to Out of Stock Products.
Frequently asked questions
What makes an AI substitute recommendation credible?
Matching on the attributes the shopper actually chose the original for, such as size, fit, material, or compatibility, and stating plainly that it is a substitute and how it differs.
Should substitutes replace the notify me option?
No. They should sit alongside notify me, stock checks, and the full category. Some shoppers want the exact item, and hiding those routes damages trust more than a lost sale.
How should stores judge substitute quality?
By return rates on substituted orders versus normal orders, and by acceptance rates per category. High returns mean the similarity rules are too loose, whatever the click rate says.
Who sets the rules for what can be substituted?
The store should, per category, naming which attributes are non negotiable. Leaving similarity entirely to the engine produces technically related and practically useless suggestions.