Cohorts from a million receipts

A cohort analysis starts with one honest table: group customers by their first-purchase month, count who comes back. I ran it on 1,067,371 real transactions from a UK online retailer (UCI Online Retail II, 2009–2011) and shipped the result as a live dashboard.

What did the cohorts actually say?

Median next-month retention is 20.5% — most customers buy once. The business runs on the ones who don't leave: in this wholesale-heavy customer base, 96.8% of revenue comes from repeat buyers, and the top customer decile alone brings 63.9%.

What did the data hide?

15.5% of invoices were cancellations that would silently inflate revenue if left in, plus guest checkouts with no customer id. The pipeline filters both — and says so, because an analysis you can't audit is an opinion.

Why ship it as a dashboard?

Findings die in notebooks. One live view — six KPIs, a revenue trend, the cohort heatmap — with every number computed from the public pipeline, nothing hand-typed, hosted for $0.

Key takeaways

• 96.8% of revenue from repeat buyers — retention is the business.
• Top decile = 63.9% of revenue — concentration is normal; know yours.
• Filter cancellations first (15.5% here) or your revenue is fiction.
• Ship the one view — a dashboard someone opens beats a notebook nobody does.

Keep reading: What does a startup KPI dashboard cost? and the full case study.