Predicted Replenishment

Consider this journey when order reports for a repeat-use product show reorder intervals falling well behind the product's consumption rhythm, or customers coming back long after the product must have run out. You need a depletion estimate per product computed from past purchase intervals together with a reliability level, repurchase and subscription-start events tied to the same customer, stock status for the product or a compatible alternative, and push, email and WhatsApp consent. If the estimate rests on a single purchase, strengthen the prediction rule first; a reminder sent on the wrong date annoys someone whose product is not empty yet.

Incremental replenishment rate vs control

What the reminder adds is how much higher the share repurchasing or subscribing is in the journey group than in the control group; a regular buyer would buy anyway, so purchase counts alone are not enough.

How it is counted: Every customer who passed the checks and was assigned to either group goes into the denominator; those who repurchase the product or a compatible alternative, or start a subscription, within the same follow-up window after assignment go into the numerator. Subtract one share from the other to get the journey's effect; push delivery or an email click does not change the denominator.

Also track

• Subscription or automatic replenishment startsIn the regular-repurchaser branch, the share of subscriptions that actually start after the subscription email; a click on the offer is not a subscription, and a lasting conversion is worth more than a single sale.
• Contribution per touchRead purchases after the push, the email and the WhatsApp message one by one; if the final reminder adds no purchases, consider dropping the WhatsApp step.
• Prediction accuracyCompare when the control group repurchases on its own with the predicted date; if purchases come long after the prediction or not at all, revisit the prediction rule and the reliability threshold.
• Opt-outs and complaintsTrack per channel how many push opt-outs, email unsubscribes and WhatsApp complaints follow the reminders; a reminder for a product that has not run out irritates people, and it shows up here first.

• Trigger and entryThe journey fires when the predicted depletion date approaches; the flow opens only if the prediction is reliable, the product has not been repurchased and the same product or a compatible alternative is available. Customers who pass these three checks are split at random: journey group with reminders, control group without.
• Who must not enterA customer whose prediction is not reliable never enters; the journey closes before it starts. Each repeat-purchase product is its own scope: a customer who has already repurchased does not enter, and attention-based Browse Abandonment and Back-in-Stock Alert, ownership-matured Cross-Sell / Complementary Offer and the not-yet-patterned First-to-Second Purchase are separate journeys.
• Timing and channelsA push first says the product may be running low; wait 2 days. Without a purchase, an email offers regular repurchasers a subscription or automatic replenishment and everyone else repurchase and pack options; wait 3 days. Still no purchase and WhatsApp permitted: a final reminder with a direct purchase link, then 2 more days.
• Exit and handoffRepurchase ends the journey at once and cancels queued reminders; a subscription or automatic replenishment that starts closes it too. An unreliable prediction and no product to suggest are exits as well; without WhatsApp consent, or with no purchase 2 days after the final reminder, no further reminders are sent, and there is no handoff.
• Frequency and consentNo more than three reminders go out per product: a push, an email and WhatsApp only where permitted; push and email need their own consent as well. When the prediction is not reliable no channel sends anything, and the wording says the product may be running low, never that it has run out.

1. 1Unless the journey group replenishes more often than the control group, the reminder merely pulled a purchase forward by a few days for people who would have bought anyway, and more clicks or add-to-carts are no substitute. Read the replenishment gap together with the consent you lost.
2. 2Everything you read here is tied to this channel order, the 2-day and 3-day waits and the reliability threshold you used. If the same customer is held for the same product in Browse Abandonment, Back-in-Stock Alert, Cross-Sell / Complementary Offer or First-to-Second Purchase, credit the purchase to one journey only; with no handoff, entry rules are what keep them from overlapping.
3. 3For a small gap, enter each group's customers and replenishment counts into the A/B test significance calculator; if it is not significant, do not call the journey a winner, extend the follow-up window to the product's consumption cycle and run the same read again.
4. 4Report customers who never started because the prediction was unreliable as their own line; if that share is large, the problem is the prediction rather than the journey, and the fix is better data, not a looser reliability threshold.