Grocery apps treat every session like a first visit, even though 73% of items in a given cart are repeats, so the weekly restock, which should take two minutes, was being run through full catalogue browsing. In 2023, as Product Design Engineer, I analysed 5,000 orders, identified three distinct shopping modes (replenishment, exploration and urgent) and designed a multi-channel experience that recognises which one is happening.
74%
Less repeat purchase time
71%from 34%
Substitutions accepted
01The problem
73% of items in a given cart are repeats, so the weekly restock should take two minutes. Habitual shoppers were instead browsing like a first-time visitor.
02How we thought about it
The apps were modelled on retail discovery, which assumes novelty. My analysis of 5,000 orders showed three distinct modes: replenishment, exploration and urgent. I treated them as separate design problems rather than one averaged experience.
03What we changed
The session opens around the likely intent, whether restock, browse or one urgent item. Someone who came to replenish no longer starts with a search.
Replenishment leads with what this household actually buys, and every line says why it is there. The shopper confirms rather than searches.
Stock-outs are where grocery trust is won or lost. Alternatives appear at the moment of unavailability, as a decision the shopper makes with context.
The trade-off
Prediction only proposes a draft basket and never places an order, because nothing should be bought on a guess.
04What it did
Reorder flow reduced repeat purchase time by 74%. Substitution acceptance rate increased from 34% to 71% with inline alternative suggestions.
How we knewEvidence
Analytics
Quantitative
5,000
Orders analysed. The fact the product is built around: 73% of items in a cart are repeats.
Design judgement
Qualitative
3
Shopping modes identified from the orders: replenishment, exploration and urgent.
Analytics
Quantitative
34% → 71%
Substitution acceptance, with 74% less repeat purchase time.
Not yet tested
None
No usability test of the replenishment flow is recorded on this page.
DecisionsAnd why
01
02
User testingWhat was and was not tested
Evidence is quantitative (orders and outcomes). A moderated usability test is not recorded.
What I personally owned on this project.