Based on Gopalakrishnan & Park (2021) · n = 12,959 customers · 5 experimental conditions
| Metric | T1L (Base Coupon) | CGL (Control) | Difference | p-value | Sig. |
|---|---|---|---|---|---|
| Purchase Rate (%) | 2.57 | 1.00 | +1.57 pp | 6.15e-05 | *** |
| Redeem Rate (%) | 0.52 | 0.00 | +0.52 pp | 0.0026 | ** |
| ARPC ($) | 0.71 | 0.26 | +$0.46 | 2.02e-07 | *** |
| ARPB ($) | 27.82 | 25.89 | +$1.93 | 0.146 | n.s. |
*** p < 0.001 ** p < 0.01 * p < 0.05 n.s. = not significant
Finding: The base coupon led to a significant increase in purchase rate (+1.57pp, a relative lift of 157%) and ARPC (+$0.46). Yet the redemption rate was only 0.52%. In other words, only about 20% of purchasers actually used the coupon. Most of the lift came from customers who purchased without redeeming. ARPB was not statistically different between the two groups, which means that spending per buyer did not change under the coupon.
| Metric | T1H (Base) | T2H (Better) | CGH (Control) | T1H−CGH (p) | T2H−CGH (p) | T2H−T1H (p) |
|---|---|---|---|---|---|---|
| Purchase (%) | 6.64 | 9.34 | 4.24 | +2.40pp (0.020)* | +5.10pp (<0.001)*** | +2.70pp (<0.001)*** |
| Redeem (%) | 0.83 | 2.06 | 0.00 | +0.83pp (0.021)* | +2.06pp (<0.001)*** | +1.23pp (<0.001)*** |
| ARPC ($) | 2.50 | 2.98 | 1.48 | +$1.02 (0.002)** | +$1.50 (<0.001)*** | +$0.48 (0.076) n.s. |
| ARPB ($) | 37.69 | 31.90 | 34.96 | +$2.74 (0.004)** | −$3.06 (<0.001)*** | −$5.79 (<0.001)*** |
*** p < 0.001 ** p < 0.01 * p < 0.05 n.s. = not significant
Finding: Both coupons significantly lifted purchase rate and ARPC for the high-value segment. The better coupon generated a higher purchase rate than the base coupon (+2.70pp). However, the ARPC difference between the two coupons ($0.48) was not statistically significant (p=0.076). We also find that the better coupon resulted in a significantly lower ARPB ($31.90 vs $37.69). This suggests that the deeper discount brought in more buyers, but those buyers spent less per transaction on average.
| Metric | T1L | CGL | Difference | p-value | Sig. |
|---|---|---|---|---|---|
| Visit Rate (%), unconditional | 34.64 | 10.61 | +24.03 pp | <2.2e-16 | *** |
| Pageviews | Visit | 10.08 | 13.14 | −3.06 | <2.2e-16 | *** |
| Purchase Rate | Visit (%) | 7.42 | 9.43 | −2.01 pp | 0.367 | n.s. |
| Discount | Purchase (%) | 30.87 | 35.38 | −4.51 pp | <0.001 | *** |
Key Insight: The main effect of the base coupon on the low-value segment was a large increase in visit rate (+24pp, a 226% relative lift). Visitors in T1L viewed fewer pages than those in the control group (10.08 vs 13.14). This may suggest that the coupon attracted some less engaged customers. However, purchase conversion among visitors was not significantly different between the two groups (7.42% vs 9.43%, p=0.367). Because conversion held steady while more customers visited, the overall purchase rate increased.
| Metric | T1H | T2H | CGH | T1H−CGH (p) | T2H−CGH (p) | T2H−T1H (p) |
|---|---|---|---|---|---|---|
| Visit Rate (%) | 44.63 | 44.17 | 22.72 | +21.91pp (***) | +21.45pp (***) | −0.46pp (0.766) |
| Pageviews | Visit | 14.92 | 15.99 | 16.18 | −1.25 (***) | −0.18 (0.554) | +1.07 (***) |
| Purchase | Visit (%) | 14.88 | 21.16 | 18.68 | −3.80pp (0.241) | +2.48pp (0.497) | +6.27pp (***) |
| Discount | Purchase (%) | 31.46 | 33.64 | 30.69 | +0.77pp (0.400) | +2.96pp (***) | +2.19pp (0.003)** |
Key Insight: Both coupons increased visit rate by about 22pp for the high-value segment. This lift was similar in size to the low-value segment. The better coupon (T2H) did not generate more visits than the base coupon (p=0.766). Instead, the better coupon led to a higher purchase conversion rate among visitors (21.16% vs 14.88%, p<0.001). This finding suggests that the deeper discount did not bring in more traffic. Rather, it helped convert visitors into buyers at a higher rate.
| Group | N | Visit Rate | × Conversion | Visit | = Purchase Rate | ARPC ($) | Redeem Rate |
|---|---|---|---|---|---|---|
| CGL (Control, Low) | 1,998 | 10.61% | 9.43% | 1.00% | 0.26 | 0.00% |
| T1L (Base, Low) | 4,787 | 34.64% | 7.42% | 2.57% | 0.71 | 0.52% |
| CGH (Control, High) | 801 | 22.72% | 18.68% | 4.24% | 1.48 | 0.00% |
| T1H (Base, High) | 1,927 | 44.63% | 14.88% | 6.64% | 2.50 | 0.83% |
| T2H (Better, High) | 3,446 | 44.17% | 21.16% | 9.34% | 2.98 | 2.06% |
| Coefficient | Estimate | Std. Error | t-value | p-value | Interpretation |
|---|---|---|---|---|---|
| (Intercept) | 1.484 | 0.206 | 7.20 | <0.001*** | Baseline ARPC for high-value control |
| treatment: base_coupon | 1.020 | 0.245 | 4.16 | <0.001*** | Coupon effect for high-value customers |
| segment: low | −1.225 | 0.244 | −5.02 | <0.001*** | Low-value customers spend $1.22 less at baseline |
| treatment × segment (interaction) | −0.564 | 0.290 | −1.95 | 0.052 | Coupon effect is marginally weaker for low-value |
Interpretation: The interaction term (p=0.052) is close to the conventional significance threshold. This result suggests that the base coupon may have a somewhat smaller effect on ARPC for low-value customers ($0.46) than for high-value customers ($1.02). However, we cannot say this difference is statistically meaningful at the 5% level. The funnel analysis helps explain why: the high-value segment has a higher baseline conversion rate. So the same increase in visits produces a larger revenue lift for the high-value segment. The coupon itself does not appear to work differently across segments.
Both coupon types significantly increased ARPC across all segments. Only about 20% of purchasers redeemed the coupon. The rest of the revenue lift came from customers who bought without using the coupon. This pattern is consistent with coupons serving as an advertising tool rather than a price discrimination mechanism. The coupon brought customers to the website, and many of them purchased using other storewide promotions instead.
Visit rates increased by 22 to 24 percentage points across all segments. This represents a relative increase of over 200%. At the same time, purchase conversion among visitors did not significantly decline. This means the coupon attracted visitors who were about as likely to buy as those who visited without a coupon. More customers came to the website, and the quality of those visitors held steady.
The better coupon ($10 off) and the base coupon ($7 off) produced nearly identical visit rates for the high-value segment (p=0.766). The difference between the two coupons was in purchase conversion among visitors (21.16% vs 14.88%). The extra $3 in discount depth did not generate additional traffic. It did, however, help more visitors complete a purchase.
Under the base coupon, discount rates were not significantly different from the control group. This means there was little margin erosion. Under the better coupon, the discount rate was modestly higher (+2.96pp vs control, p<0.001) and ARPB was lower ($31.90 vs $34.96). The firm gains higher volume but accepts slightly lower margins per transaction.
The base coupon produced a significant lift in ARPC with little margin erosion. The effect was consistent across both segments. Because most buyers did not redeem the coupon, the cost to the firm is low relative to the revenue gained.
The better coupon increased conversion but also raised the redemption rate and lowered ARPB. The incremental ARPC gain over the base coupon ($0.48) was not statistically significant (p=0.076). A cost-benefit analysis is needed to determine whether the extra discount cost per redeemer is justified. Testing intermediate discount depths ($8 to $9) may also be worth considering.