Big and busy isn't the same as worth investing in.
FoodCorp runs four stores across London, Birmingham, and Nottingham, and the obvious way to pick a marketing target — whichever is biggest or busiest — rewards traffic, not loyalty. A store can pull crowds and still leak every new customer straight back out.
The better question is which store turns marketing spend into repeat, high-value customers. Answering that meant measuring behaviour over time, not a single snapshot.
A customer you keep is worth more than three you acquire and lose.
Four monthly measures of health.
Customer acquisition rate.
The share of genuinely new customers each month — the top of the funnel, and the one most easily mistaken for success.
Repeat purchase rate.
The percentage of customers who transact more than once in a month — the clearest signal of whether people come back.
Revenue per customer & growth rate.
Average monthly revenue per unique customer, and month-on-month growth of the customer base — value and momentum together.
Transactions to dashboards.
Raw transactions were loaded into a database, shaped into KPI views with SQL, computed monthly per store in Databricks, and visualised in Tableau for the store-by-store comparison.
Defining KPIs as SQL views — for example, repeat rate as customers with more than one transaction over total customers in the month — kept every figure reproducible and auditable, which is the whole point of a KPI document.
Leadenhall Street keeps its customers.
London — Leadenhall Street stood out on the measure that compounds: a repeat-purchase rate around 90%, paired with strong, stable revenue per customer and positive growth momentum. Birmingham — Frederick Street had acquisition peaks but weak retention — a leaky bucket — and the other London and Nottingham stores trailed on loyalty.
So the recommendation was Leadenhall Street, precisely because a campaign there is more likely to convert a new customer into a returning, high-value one rather than a one-time visit.
A poster, a report, a recommendation.
The analysis fed both an individual report and a group A1 poster, so the same KPI story had to work as a detailed write-up and as a glanceable summary. Tableau dashboards carried the trends; the recommendation carried the conclusion. Keeping the KPI definitions in a shared document meant everyone was arguing about the same numbers.
Three things, next time.
Add a cohort view.
Tracking each month's new customers as a cohort would show retention decay directly, rather than inferring it from monthly repeat rates.
Model expected ROI, not just KPIs.
Translating the KPI edge into a projected return on a fixed marketing spend would make the recommendation quantitative, not comparative.
Control for store size explicitly.
Normalising KPIs against catchment or footfall would rule out size as a hidden driver and strengthen the retention argument.