Retention starts with the second order.
Explore the interactive demo →
A decision brief for sales and finance using UCI Online Retail. Purchases through November 2011 are analyzed; December is excluded because it is incomplete.
What to do next
- Investigate the second-order journey. Only 47.4% of customers with a full 90-day observation window placed a second order in that time. Test an onboarding offer against a randomized holdout; measure incremental 90-day repeat and margin, not just coupon use.
- Review account concentration. The top 10% of observed purchasers account for 59.6% of identified purchaser net revenue. Examine large wholesale accounts and account-level refund exposure before setting targets.
- Check seasonality and acquisition mix. Monthly active purchasers range from 741 to 1,664. Compare like-for-like seasons and new versus existing accounts before interpreting changes as churn.
These are descriptive associations, not causal effects. The data contain no campaign exposure, acquisition cost, margin, or reason for inactivity.
Repeat purchasing
1,538 of 4,297 purchasers have only one observed order. Among customers with a second observed order, the median time to it is 50 days; this conditional median excludes one-time customers and is not a population time-to-event estimate. Customers without a second order may return after the data end.
Active customers by month
Distinct identified customers with at least one qualifying purchase. The first and last bars cover complete calendar months, but activity is not evidence of permanent churn.
Show monthly figures
| active | orders | gross | |
|---|---|---|---|
| Month | |||
| 2010-12 | 885 | 1400 | £572,714 |
| 2011-01 | 741 | 987 | £569,445 |
| 2011-02 | 758 | 997 | £447,137 |
| 2011-03 | 974 | 1321 | £595,501 |
| 2011-04 | 856 | 1149 | £469,200 |
| 2011-05 | 1056 | 1555 | £678,595 |
| 2011-06 | 991 | 1393 | £661,214 |
| 2011-07 | 949 | 1331 | £600,091 |
| 2011-08 | 935 | 1280 | £645,344 |
| 2011-09 | 1266 | 1755 | £952,838 |
| 2011-10 | 1364 | 1929 | £1,039,319 |
| 2011-11 | 1664 | 2657 | £1,161,817 |
Cohort purchase activity
Percent of each first-observed-month cohort making at least one purchase in the indicated calendar month after that first month. M0 is 100%; activity is not necessarily consecutive. Em dash means the month has not fully elapsed. The first cohort is left-censored and may contain existing customers.
| M0 | M1 | M2 | M3 | M4 | M5 | M6 | |
|---|---|---|---|---|---|---|---|
| First observed month | |||||||
| 2010-12 | 100.0 | 36.6 | 32.3 | 38.4 | 36.3 | 39.8 | 36.3 |
| 2011-01 | 100.0 | 22.1 | 26.6 | 23.0 | 32.1 | 28.8 | 24.7 |
| 2011-02 | 100.0 | 18.7 | 18.7 | 28.4 | 27.1 | 24.7 | 25.3 |
| 2011-03 | 100.0 | 15.0 | 25.2 | 19.9 | 22.3 | 16.8 | 26.8 |
| 2011-04 | 100.0 | 21.3 | 20.3 | 21.0 | 19.7 | 22.7 | 21.7 |
| 2011-05 | 100.0 | 19.0 | 17.3 | 17.3 | 20.8 | 23.2 | 26.4 |
| 2011-06 | 100.0 | 17.4 | 15.7 | 26.4 | 23.1 | 33.5 | — |
| 2011-07 | 100.0 | 18.1 | 20.7 | 22.3 | 27.1 | — | — |
| 2011-08 | 100.0 | 20.7 | 24.9 | 24.3 | — | — | — |
| 2011-09 | 100.0 | 23.4 | 30.1 | — | — | — | — |
| 2011-10 | 100.0 | 24.0 | — | — | — | — | — |
| 2011-11 | 100.0 | — | — | — | — | — | — |
Interpretation and boundaries
Transactions are invoice lines for a UK-based online retailer, including wholesale customers, from 1 December 2010 to 9 December 2011. This analysis stops before 1 December 2011. Customer IDs are missing on 127,216 of 516,384 lines in the analysis window; those lines cannot be linked to a customer and are excluded from customer metrics. First observed purchase is not necessarily acquisition. The source is historical and cannot establish current performance or represent all retailers.
Orders are distinct customer/invoice combinations with positive quantity and price, excluding cancellation invoices. Revenue concentration uses positive purchase value plus signed cancellation lines for the same identified purchasing customers, recognized on the transaction date. Refunds are not matched to original invoices; refunds after cutoff and nonstandard adjustments are missed. Identified cancellation lines total £-435,656; refund-only customers contribute £-10,408 and are excluded from purchaser concentration. This is net transaction value, not profit or accounting revenue.
Source: Chen, D. (2015), Online Retail, UCI Machine Learning Repository, CC BY 4.0. See README for definitions, exclusions, reproducibility, and data quality checks.