Replenishment recommendations: reminding repeat buyers at the moment they actually run out

October 22, 2026

Short answer: A replenishment recommendation succeeds when it feels like the store remembered something useful. The customer bought a ninety day supply eighty five days ago; a quiet reminder with a one tap reorder is a service. The same message sent weekly from the day of purchase is noise, and the difference is entirely in whether the timing model respects the customer's actual consumption rhythm.

Build the timing from real intervals

Start with the product's expected life at normal use, then adjust with the customer's own history. A customer who reorders a sixty day product every fifty days runs through it faster than the label assumes, and their second and third intervals are better evidence than any category average. Models should learn per customer, per product, and widen the window when the data is thin rather than pretend precision.

Pack size and household signals refine the estimate: the large bag lasts longer, the multi pet household empties the small one sooner. Each correction is modest, but timing errors compound in the customer's mind. A reminder that lands within a few days of running out feels attentive; one that lands three weeks early feels like a sales quota wearing a helpful costume.

Choose the channel by urgency and habit

On site recommendations suit the customer who returns on their own rhythm: a reorder shelf on the account page and relevant placements when they browse near the expected date. Email and messages suit the customer who forgets, timed to land just before the predicted run out with the exact product and quantity from last time preloaded.

Subscriptions are the stronger offer for the predictable core, and replenishment recommendations are how customers discover they want one. A well timed reminder that converts a third manual reorder into a subscription has done more for retention than a discount, because it removes the remembering work that caused lapses in the first place.

Handle the misses gracefully

Customers stockpile, receive gifts, buy elsewhere once, or stop using a product. Every reminder needs a painless not yet and no longer path, and the model should treat a skipped reminder as information, pushing the next prediction out rather than repeating the nudge on a fixed schedule. Systems that keep shouting after being told no train customers to ignore the channel entirely.

Stock awareness matters too. Recommending a reorder the store cannot fulfil, or pushing a substitute without saying the original is unavailable, burns the trust the programme exists to build. The reminder should check availability before it speaks, and stay silent when the shelf is empty.

Measure retention, not just clicks

The honest metrics are reorder interval stability, lapse rate, and the share of customers still buying the category at six and twelve months, compared against customers without reminders. Open and click rates flatter programmes that message too often; retention exposes them, because annoyed customers do not always unsubscribe, sometimes they just quietly buy elsewhere.

Finally, cap the programme's total voice per customer across products. Five categories each running a well timed reminder can still add up to a message a day. One coordinated calendar that knows everything the store wants to say is the difference between a helpful shopkeeper and five departments competing for the same inbox.

Keep reading: Recommendations on zero results pages: recovering the lost search, Cold start recommendations: what to show shoppers you know nothing about, and How should recommendation ranking weigh margin and stock levels? cover the nearby cases in this series.