More coupon buyers. Was the subsidy buying repeat business?
I connected package use, repeat purchase, and subsidy economics to help an on-demand grocery product evaluate its pricing and product choices.
- My role
- Evaluation, diagnosis & iteration support
- Original tools
- SQL · Excel · BI
- Deliverable
- Evaluation framework & pricing insights
The problem
A low-priced package of six vouchers attracted buyers. The business needed to know whether it encouraged repeat orders at an acceptable subsidy cost.
What I did
- Reworked the value evaluation to examine buyer selection and the order that triggered purchase.
- Connected voucher use and repeat purchase to pricing, checkout guidance, reminders, and subsidy efficiency.
Package buyers were not automatically an appropriate comparison group.
Buyers versus non-buyers
People who choose a package may already differ in purchase intent and product familiarity. Their higher spending need not be value created by the package.
Before versus after purchase
The order that triggers package purchase can inflate a simple before/after comparison. New users can also have incomplete purchase history.
My approach: Refine the eligible sample and order comparison, then keep the resulting value estimates separate from a controlled causal effect.
The cheapest tier had a large one-use audience.
In the 1.5 CU tier, 56% used one voucher and 12% used none. More frequent use was associated with repeat package purchase. Explore both views in the matrix.
Package buyers by price and voucher use
Vouchers used during the seven-day package
(CU)0123456 1.54.58.5
44,800 of 80,000 buyers used one voucher (56%). 13% bought another package within 14 days after expiry.
View cohort counts & definitions
| Price (CU) | Vouchers used | Buyers | Tier share | Repeat buyers | Repeat rate |
|---|---|---|---|---|---|
| 1.5 | 0 | 9,600 | 12.0% | 576 | 6.0% |
| 1.5 | 1 | 44,800 | 56.0% | 5,824 | 13.0% |
| 1.5 | 2 | 14,400 | 18.0% | 3,888 | 27.0% |
| 1.5 | 3 | 5,600 | 7.0% | 1,680 | 30.0% |
| 1.5 | 4 | 3,200 | 4.0% | 1,088 | 34.0% |
| 1.5 | 5 | 1,600 | 2.0% | 608 | 38.0% |
| 1.5 | 6 | 800 | 1.0% | 336 | 42.0% |
| 4.5 | 0 | 4,000 | 8.0% | 320 | 8.0% |
| 4.5 | 1 | 17,000 | 34.0% | 3,740 | 22.0% |
| 4.5 | 2 | 14,000 | 28.0% | 5,180 | 37.0% |
| 4.5 | 3 | 7,000 | 14.0% | 2,870 | 41.0% |
| 4.5 | 4 | 4,000 | 8.0% | 1,800 | 45.0% |
| 4.5 | 5 | 2,500 | 5.0% | 1,200 | 48.0% |
| 4.5 | 6 | 1,500 | 3.0% | 780 | 52.0% |
| 8.5 | 0 | 800 | 4.0% | 80 | 10.0% |
| 8.5 | 1 | 3,600 | 18.0% | 900 | 25.0% |
| 8.5 | 2 | 5,600 | 28.0% | 2,240 | 40.0% |
| 8.5 | 3 | 4,000 | 20.0% | 1,840 | 46.0% |
| 8.5 | 4 | 2,800 | 14.0% | 1,428 | 51.0% |
| 8.5 | 5 | 2,000 | 10.0% | 1,120 | 56.0% |
| 8.5 | 6 | 1,200 | 6.0% | 720 | 60.0% |
Decision: Evaluate useful voucher redemption and repeat purchase alongside package sales. Test reminders and benefits at the moments users might return.
More purchases had hidden weaker subsidy efficiency.
July brought the highest purchase volume but a weaker modeled return per subsidy unit. In August, purchases and platform subsidy fell while the estimated extra-GMV-to-subsidy ratio improved.
Review spending and value together
View data & definitions
The value numerator is a scoped model estimate, not randomized incremental revenue. Monthly package purchases include repeat purchases and are not unique users. The matrix uses separate buyer cohorts and must not be added to these monthly counts. Period changes do not isolate the effect of a price adjustment.
| Month | Package purchases | Platform subsidy (CU) | Estimated extra GMV (CU) | Estimated extra GMV / subsidy |
|---|---|---|---|---|
| May | 210,000 | 840,000 | 21,000,000 | 25.0× |
| June | 420,000 | 1,800,000 | 54,000,000 | 30.0× |
| July | 750,000 | 4,200,000 | 63,000,000 | 15.0× |
| August | 700,000 | 3,500,000 | 77,000,000 | 22.0× |
| September | 680,000 | 3,200,000 | 80,000,000 | 25.0× |
Decision: Judge pricing with purchase volume, repeated use, and platform cost together; investigate which user groups produce useful repeat business.
Make the package useful to the customer and economical for the platform.
What I delivered: An evaluation framework and usage insights informing checkout guidance, usage reminders, and pricing iterations. The decks document product and minimum-price changes, alongside further work on customer-specific pricing.
Cohorts, modeled value & source notes
Based on my JD Daojia Q2/Q3 work reports and subsequent promotion review. Original tools: SQL, Excel, and BI. The English presentation, Python charts, and interactive matrix are newly produced.
- A package contains six vouchers valid for seven days. Each price cohort has a full validity period plus 14 days after expiry to observe another package purchase. Prices are illustrative CU.
- Buyer distribution sums to 100% within each price tier. Repeat-purchase rates use the buyers in that exact price-and-usage cell. Users select into purchase price and usage patterns; the association does not establish a causal two-voucher threshold.
- Estimated extra GMV divided by platform subsidy is a modeled merchandise-value efficiency ratio. It excludes margins, merchant costs, and full business profit.
- The historical evaluation addressed buyer-selection and before/after problems through scoped samples and adjusted order comparisons. It does not by itself establish a randomized counterfactual.
- Documented checkout and pricing iterations are distinct from later pricing plans; this case does not assert that every proposed segmentation or pricing system was deployed.
Original decks, internal dashboards, pricing logic, SQL, and customer records remain private.
Download package cohorts and economics