Shopping cart coupon statistics from Alibaba’s randomized promotion study By Said Maybe, Said Maybe. Sources checked 2026-11-11. Compare immediate promoted-product sales with later browsing, purchase incidence and pricing in a large randomized Alibaba experiment. ## Quick answer In a randomized Alibaba mobile shopping experiment, the analyzed treatment observations had a 2.11% promoted-product purchase rate versus 0.98% in the coupon holdout. During a separate four-week follow-up, treatment assignment was associated with a 0.32% relative rise in daily purchase incidence across sellers, while daily expenditure showed no statistically significant difference. The immediate product result and the later platform-wide result use different outcomes and denominators. Zhang, Dai and Dong, How Do Price Promotions Affect Customer Behavior on Retailing Platforms? (https://anderson-review.ucla.edu/wp-content/uploads/2021/03/Dai_et-al_AlibabaCoupons_SSRN-id3029707.pdf) ## Key findings and sources ### Immediate purchase incidence on promoted products In Zhang, Dai and Dong’s 2016 Alibaba mobile experiment, 2.11% of eligible customer-product-day observations in the coupon condition included a purchase of the promoted product versus 0.98% in control. The short-period analysis covered 1,693,239 such observations across arms. Zhang, Dai and Dong, How Do Price Promotions Affect Customer Behavior on Retailing Platforms? (https://anderson-review.ucla.edu/wp-content/uploads/2021/03/Dai_et-al_AlibabaCoupons_SSRN-id3029707.pdf) ### Promoted-product spending in the coupon window During March 12 to April 11, 2016, the Alibaba experiment reported average promoted-product expenditure of 1.93 RMB in coupon treatment and 1.01 RMB in control across eligible observations. The measure was expenditure per eligible observation. Zhang, Dai and Dong, How Do Price Promotions Affect Customer Behavior on Retailing Platforms? (https://anderson-review.ucla.edu/wp-content/uploads/2021/03/Dai_et-al_AlibabaCoupons_SSRN-id3029707.pdf) ### What the later shopping results measured Zhang, Dai and Dong found a 0.53% relative rise in daily product views and a 0.32% relative rise in daily purchase incidence across sellers during the April 12 to May 9 follow-up. Their 25,137,243-customer-day analysis found no statistically significant difference in daily expenditure. Zhang, Dai and Dong, How Do Price Promotions Affect Customer Behavior on Retailing Platforms? (https://anderson-review.ucla.edu/wp-content/uploads/2021/03/Dai_et-al_AlibabaCoupons_SSRN-id3029707.pdf) ### Later shopping carts after coupon exposure The same Alibaba study reported a 0.41% relative rise in post-treatment cart-to-view ratio for assigned treatment across all sellers, in 15,059,021 customer-days with a view. It measures cart activity per product viewed rather than completed purchases. Zhang, Dai and Dong, How Do Price Promotions Affect Customer Behavior on Retailing Platforms? (https://anderson-review.ucla.edu/wp-content/uploads/2021/03/Dai_et-al_AlibabaCoupons_SSRN-id3029707.pdf) ### Coupon assignment and coupon viewing estimates In the 2016 Alibaba analysis, adjusted relative promoted-product purchase incidence rose by 116% for assignment and by an estimated 173% for coupon viewing. The latter uses an instrumental-variable design and depends on its identification assumptions. Zhang, Dai and Dong, How Do Price Promotions Affect Customer Behavior on Retailing Platforms? (https://anderson-review.ucla.edu/wp-content/uploads/2021/03/Dai_et-al_AlibabaCoupons_SSRN-id3029707.pdf) ### Draw, analyzed sample and coupon viewing measure separate stages The Alibaba promotion study began with a one-million-user draw and retained 931,009 after activity exclusions, a difference of 68,991 users. The authors also report 116% relative purchase-incidence lift from assignment and 173% from estimated viewing; the two results concern different estimands. Zhang, Dai and Dong, How Do Price Promotions Affect Customer Behavior on Retailing Platforms? (https://anderson-review.ucla.edu/wp-content/uploads/2021/03/Dai_et-al_AlibabaCoupons_SSRN-id3029707.pdf) ### A 17% coupon can require more than 60% additional orders in a worked case Using the Alibaba study’s reported 17% mean coupon as an illustrative input, a $100 order with $55 variable cost moves from $45 to $28 of contribution. The same contribution requires 60.71% more orders. Zhang, Dai and Dong, How Do Price Promotions Affect Customer Behavior on Retailing Platforms? (https://anderson-review.ucla.edu/wp-content/uploads/2021/03/Dai_et-al_AlibabaCoupons_SSRN-id3029707.pdf) ### Spending per purchasing observation was lower in the coupon arm Using Zhang, Dai and Dong’s rounded Table 2A means, expenditure per purchasing customer–product observation was about 91.47 RMB in the coupon arm and 103.06 RMB in control. This is a conditional descriptive comparison among purchase observations. Zhang, Dai and Dong, How Do Price Promotions Affect Customer Behavior on Retailing Platforms? (https://anderson-review.ucla.edu/wp-content/uploads/2021/03/Dai_et-al_AlibabaCoupons_SSRN-id3029707.pdf) Targeted customer–product observations in the March to April 2016 Alibaba promotion period. Divide mean RMB expenditure, including zeros for nonpurchase, by the share of observations with a purchase in each arm. ### Conditional spend and spending per eligible observation moved in different directions The Alibaba Table 2A values imply 11.59 RMB lower spend per purchasing observation in treatment while average expenditure across eligible observations was 0.92 RMB higher. The two measures use different denominators. Zhang, Dai and Dong, How Do Price Promotions Affect Customer Behavior on Retailing Platforms? (https://anderson-review.ucla.edu/wp-content/uploads/2021/03/Dai_et-al_AlibabaCoupons_SSRN-id3029707.pdf) Same randomized promotion, with descriptive conditioning on the post-assignment purchase outcome. Compare 1.93 minus 1.01 across eligible observations and 1.93 divided by 0.0211 versus 1.01 divided by 0.0098 among purchasers. ## Compare the results Figure description: Promoted-product purchases during the coupon window. Exact values are also in the study data. Promoted-product purchases during the coupon window. ## Begin with the eligible shoppers and the exact offer The authors observed a retailer coupon program on Alibaba’s mobile app. They randomly drew one million shoppers already targeted by sellers for a promotion. Half were assigned to be eligible to receive the offer under the program rules and half were withheld from it. Activity exclusions applied before the reported baseline panel of 931,009 customers. In this sample the average discount was 17%, with a range of 8% to 99%. Treat that spread as a description of the offers in the study. The experiment did not randomly assign discount size. During March 12 to April 11, 2016, the 1,693,239 eligible customer-product-day observations showed 2.11% purchase incidence for promoted products in treatment and 0.98% in control. The raw gap was 1.13 percentage points. A store testing its own offer can keep the same discipline by defining eligibility before assignment, including customers who never view a coupon in the primary comparison and stating which product and window count as a purchase. ## Separate a purchase of the promoted product from what the customer spent In the short window, expenditure on the promoted product averaged 1.93 RMB in treatment and 1.01 RMB in control. Those are averages across eligible observations. The study does not supply a profit result for the typical seller. Before widening a promotion, compare contribution per eligible shopper using the discount actually paid, fulfillment costs, refunds and the mix of products ordered. A promotion can sell more units and still change the contribution on orders that would have arrived at the earlier price. The authors also estimate the effect of viewing the coupon by using random assignment as an instrument. They report a 173% relative increase in promoted-product purchase incidence for that estimate, compared with a 116% adjusted increase for assignment. This distinction matters because about 66% of treated users actually viewed coupons. Looking only at viewers would normally select shoppers who had returned and could therefore buy. The instrumental estimate relies on additional assumptions and answers a narrower question. ## Add a follow-up for shopping behavior after the coupon expires In the separate April 12 to May 9 window, assigned treatment users viewed 0.53% more products per day and had a 0.32% relative increase in daily purchase incidence across sellers. The paper did not find a significant difference in daily expenditure over this window. The underlying control averaged 24.47 views per day. A 0.53% increase at that baseline is about 0.13 extra views per customer-day, which makes the scale easier to read. If you evaluate a merchant coupon in one store, the immediate product sale and later store-wide spend are separate lines on the report. Keep the timing explicit and the original group assignments intact. A larger short-run movement and a smaller post-offer movement can coexist. They describe different outcomes and should lead to different decisions about replenishment, offer frequency and whether the incentive earns its cost. ## Check whether shoppers change how they use the cart In the follow-up, the cart-to-view ratio increased by 0.41% relative to control in the all-seller analysis. The result comes from customer-days with at least one view. The study also detected a small −0.02% relative change in price paid under its product and date controls. Taken together, these results make it reasonable to measure whether future cart behavior shifts alongside purchases. They do not show that a particular shopper learned to wait for coupons. For a recurring recovery program, report the share of assigned shoppers with an order after the discount expires and examine how often comparable items sell at their regular price. If a retailer always offers the same reward shortly after abandonment, its own trial can test whether that rule adds profitable sales over several purchase cycles. A large marketplace with many participating sellers has dynamics a single DTC store cannot assume it shares. A customer buying across several retailers can shift the next order to a seller that never issued the coupon. The marketplace results include that possibility. A single-store trial can instead observe orders on its own storefront and avoid treating activity at other sellers as measured revenue. ## Keep the analysis sample and the exposure question separate The authors initially drew one million users and retained 931,009 after the stated activity exclusions. That leaves 68,991 outside the analysis sample, about 6.90% of the initial draw. It is an eligibility and analysis difference. It is not an abandonment rate, a count of missing coupons or the share of people who chose to ignore a promotion. Within the experiment, assignment and viewing answer different questions. The paper reports an adjusted 116% relative purchase-incidence lift from assignment and an estimated 173% lift from viewing. The 57-point difference compares two relative lift estimates, not two purchase rates. A person may never see the coupon after being assigned. Comparing people who opened an offer with those who did not can introduce selection because the groups already differ in shopping activity. For a store test, record how many eligible shoppers were assigned before measuring downstream delivery, display or redemption. Use all assigned eligible shoppers for the primary randomized comparison. Report visibility separately to diagnose why results may differ by browser, channel or checkout stage. If a store changes the inclusion rules after seeing who viewed the offer, show which groups changed and hold any causal interpretation until the analysis is reconciled. Keep the assignment comparison even when redemption is the more convenient number to retrieve. ## Put the coupon’s cost against the orders it has to add The Alibaba promotion study reported an average discount depth of 17% in its sampled promotions. A merchant can use that value to see how a promotion affects the contribution needed for break-even. The study’s observed purchase result does not establish the merchant’s margin or the effect of a new campaign. Suppose a $100 order costs $55 to fulfill and leaves $45 of contribution before a coupon. A 17% discount leaves $28. Moving from 100 baseline orders to 160 discounted orders produces $4,480, slightly below $4,500 without a coupon. Under those assumptions, orders need to rise by 60.71% to break even. Run this comparison on expected contribution per eligible shopper. Keep the discount paid on orders that would have happened anyway. Let customers who qualify but never open the message remain in the denominator. If the basket mix changes, estimate the actual contribution of both assigned groups after returns rather than projecting a single order margin. ## More buyers and lower spend per purchasing observation can occur together The Alibaba experiment reports that 2.11% of targeted treatment customer–product observations ended in a purchase, compared with 0.98% in control. Expenditure per targeted observation rose from 1.01 to 1.93 RMB. Because nonpurchase observations contribute zero expenditure, dividing expenditure by purchase incidence yields about 91.47 RMB per purchasing observation in treatment and 103.06 RMB in control. The result is descriptive among people who purchased after assignment. The treatment can change which people and items enter that group. The 11.59 RMB difference cannot be interpreted as a causal reduction in what the same person would spend. It does show why a team reading a coupon report should inspect purchase incidence, spend per eligible observation and the changed mix of buyers separately. ## Make the paid-order and contribution outcomes explicit in your own test Use an eligible customer as the denominator for the campaign’s incremental paid orders and contribution. Keep a diagnostic view for basket price and quantity among purchasers. The first view answers whether the promotion produced more value overall. The second can help explain whether the mix moved to smaller or cheaper orders. If a promotion discounts a specific item, also check whether buyers substitute away from other products that carry higher margin. Track all orders within a fixed follow-up window and record whether an order was refunded. When the treatment increases purchase incidence but the value of each purchased basket falls, the rollout decision depends on contribution per eligible customer, with the coupon cost counted on purchases that would have happened anyway. ## Study data Table: Zhang, Dai and Dong, How Do Price Promotions Affect Customer Behavior on Retailing Platforms? Measure | Value | Context Users in the random draw | 1,000,000 customers | 500,000 in each group before exclusions Customers after daily activity exclusions | 931,009 customers | The reported baseline panel contains this many customers Mean discount depth in sampled promotions | 17% | Discount divided by original price Targeted treatment users who actually viewed coupons | 66% | Rounded share viewing at least one coupon, as stated in the paper Promoted-product purchase incidence, coupon treatment | 2.11% | Unadjusted treated mean during March 12–April 11 Promoted-product purchase incidence, holdout | 0.98% | Unadjusted control mean during March 12–April 11 Promoted-product expenditure, coupon treatment | 1.93 RMB | Average expenditure in the treatment group; all eligible observations Promoted-product expenditure, holdout | 1.01 RMB | Average expenditure in the control group; all eligible observations Adjusted purchase incidence lift from assignment | 116% | Relative intent-to-treat effect for the promoted product with product and date effects Adjusted expenditure lift from assignment | 90% | Relative intent-to-treat effect on promoted-product expenditure Estimated purchase incidence lift from viewing the coupon | 173% | Instrumental-variable estimate using randomized assignment to estimate exposure Control products viewed per customer-day in follow-up | 24.47 views | Control mean across all sellers in April 12–May 9 follow-up Follow-up views lift from assignment | 0.53% | Relative change in daily product views across all sellers Follow-up purchase incidence lift from assignment | 0.32% | Relative change in probability of any purchase on a customer-day across all sellers Follow-up cart-to-view ratio lift from assignment | 0.41% | Relative change in products added to cart per product viewed, after the treatment period Follow-up price paid change from assignment | -0.02% | Relative estimated price paid change; the paper uses product and date effects ## What changes when you compare the figures Table: Calculations from the published values above. Comparison | Result | Calculation Promoted-product purchase incidence difference | 1.13 percentage points | = 2.11 - 0.98 Treated to control promoted-product purchase incidence | 2.15 times | = 2.11 / 0.98 Difference in reported expenditure on a promoted product | 0.92 RMB per observation | = 1.93 - 1.01 Daily product views in the treated group implied by reported increase | 24.6 views per customer-day | = 24.47 * (1 + 0.53 / 100) Sample difference after activity exclusions | 68,991 users | = 1000000 - 931009 Excluded share of the initial random draw | 6.9% | = (1000000 - 931009) / 1000000 * 100 Reported relative lift estimates, viewing minus assignment | 57 points of relative lift | = 173 - 116 Worked case, order contribution after a 17% discount | $28.00 | = 100 - 55 - 17 Worked case, required order uplift at the discounted contribution | 60.71% | = (45 / 28 - 1) * 100 Worked case, total contribution for 160 discounted orders | $4,480.00 | = 160 * 28 Coupon arm spend per purchasing observation | 91.469 RMB per purchase observation | = 1.93 / 0.0211 Control arm spend per purchasing observation | 103.061 RMB per purchase observation | = 1.01 / 0.0098 Coupon-minus-control conditional spend | -11.592 RMB per purchase observation | = 1.93 / 0.0211 - 1.01 / 0.0098 ## Methodology and sources We transcribed results from the UCLA-hosted author manuscript. Section 4.2 reports one million randomly drawn targeted customers before activity exclusions, and Table 1 reports 931,009 in its baseline panel. Table 2 is a customer-product-day analysis of the coupon period. Table 3 uses customer-day outcomes in the April 12 to May 9 follow-up. Table 4 defines additional outcome-specific populations. We use the exact table values, preserve relative changes and keep the instrumental-variable result for coupon views separate from the randomized assignment result. Simple differences and ratios use the displayed rounded means. No causal claim here concerns standalone DTC performance. [Download the source data (CSV)](decision-table.csv) [Download the calculations (CSV)](calculations.csv) [Download the citation notes (CSV)](citation-notes.csv) [Download the evidence behind the new findings (CSV)](finding-evidence.csv) ## Related reading Online discount statistics from a 314-trader pricing sweep (https://saidmaybe.com/guides/online-discount-statistics/) Equal-Value Checkout Promotions Can Get Different Responses (https://saidmaybe.com/guides/equal-value-checkout-promotion-statistics/) Countdown timer statistics and how to verify an ecommerce offer (https://saidmaybe.com/guides/ecommerce-countdown-timer-and-scarcity-statistics/) Check the real price before you promise two Shopify discounts (https://saidmaybe.com/resources/shopify-discount-combination-cases/) When to offer a discount in an abandoned checkout conversation (https://saidmaybe.com/guides/offer-a-discount-in-checkout-conversation/) Browse checkout recovery research (https://saidmaybe.com/research/) ## Sources and definitions Zhang, Dai and Dong, How Do Price Promotions Affect Customer Behavior on Retailing Platforms? (https://anderson-review.ucla.edu/wp-content/uploads/2021/03/Dai_et-al_AlibabaCoupons_SSRN-id3029707.pdf) The source data includes the population, period, denominator and limitations for every figure. Said Maybe · saidmaybe.com