L1 › L2 › L3 Category Deep Dive • Generated 2026-04-14
| Category Path (L1 › L2 › L3) | Total Issued | Redeemed | Not Redeemed | Redeem % | Total $ Issued | Avg $ | Redeemed $ | Unredeemed $ | Rate |
|---|
Does a bigger promo actually get used more? Breakdown by individual promo code dollar amount.
PS_CES.test4) began on 2025-04-28.
This section includes only promo codes issued on or after 2025-04-28 so every promo had an equal chance for its redemption status to be captured.
Including earlier promos would artificially inflate “not redeemed” counts where we simply had no signal.
Universe for this section: promo codes ().
Clear trend: higher-value promos get redeemed at higher rates. ⚠ $100 is n=17 — statistically unreliable.
$10 dominates volume. Stacked bars show the redeemed/unredeemed split per promo code amount.
📋 Methodology Note
Why are the group sizes unequal? The No Promo group is naturally much larger than the Promo group — most customers who contact CS do not receive a promo code. We intentionally keep both groups at their natural sizes rather than downsampling to match. Forcing equal group sizes would discard statistically valid observations and introduce unnecessary sampling bias. Since all comparisons use per-customer averages (avg GMV lift %, avg pre/post spend per customer), the size imbalance does not skew the results — averages are scale-invariant by definition.
Observational analysis, not a randomized experiment. Customers were not randomly assigned to promo or no-promo groups. Promo recipients may differ systematically from non-recipients — for example, they may be higher-engagement customers or have made larger recent purchases. This means the retention difference is a correlation, not a proven causal effect of the promo itself. An A/B test with random assignment would be the gold standard for isolating causality, and is recommended as a next step.
Every L1 › L2 › L3 path scored on ROI and GMV retention advantage vs the No Promo baseline.
Compares post-contact spending for customers who received ANY promo code vs those who did not.
Same No Promo baseline — treatment group narrowed to customers who redeemed their promo code. ⚠ Redeemed data available Q2–Q4 only (starts WK13 FY26).
📋 Methodology Note
Why are the group sizes unequal? The No Promo group is naturally much larger than the Promo group — most customers who contact CS do not receive a promo code. We intentionally keep both groups at their natural sizes rather than downsampling to match. Forcing equal group sizes would discard statistically valid observations and introduce unnecessary sampling bias. Since all comparisons use per-customer averages (avg GMV lift % = (Post-30d − Pre-30d) ÷ Pre-30d per customer), the size imbalance does not skew the results — averages are scale-invariant by definition.
Observational analysis, not a randomized experiment. Customers were not randomly assigned to promo or no-promo groups. Promo recipients may differ systematically from non-recipients — for example, they may be higher-engagement customers or have made larger recent purchases. This means the retention difference is a correlation, not a proven causal effect of the promo itself. An A/B test with random assignment would be the gold standard for isolating causality, and is recommended as a next step.
Categories scored on GMV retention advantage vs No Promo baseline (30d symmetric windows).
Compares post-contact spending for customers who received ANY promo code vs those who did not. Uses symmetric 30-day windows: 30d before vs 30d after contact date.
Same No Promo baseline — treatment group narrowed to customers who redeemed their promo code. ⚠ Redeemed data available Q2–Q4 only (starts WK13 FY26).
Avg GMV per customer in each non-overlapping time bucket (e.g. “1–14d Post” = sum of all GMV in days 1–14 after contact, avg'd per customer). Contact date = day 0 (red dashed). Faded = pre-contact; solid = post-contact. Chart 1 Promo Issued vs No Promo. Chart 2 Redeemed vs No Promo. Both answer: why does the 30-day analysis show more decline than 14-day?
Blue = all promo recipients (redeemed + not redeemed, weighted avg). Gray = No Promo.
Amber = customers who actually redeemed their promo code. Gray = No Promo (same baseline).
What share of promo-receiving customers fall into each promo-to-AOV ratio bucket? Full FY26 — no redemption filter.
Step 1 — Identify Missing Order Cases: Filter FY26 incidents on 8 L3 reason codes related to missing or damaged items: Never Arrived (Lost/Stolen), Missing Items, Damaged Item, Wrong Item(s) Received, Items/Order Damaged, Bundle/Set/Missing Box X of Y, Order Missing at Store for Pickup, Wrong Item/Order Dispensed.
Step 2 — Calculate Total Adjustment: For each order (SALES_ORDER_NUM), sum the adjustment amount across all line items from CES Appeasements where the appeasement type is one of: Item Refunds, Item Replacement, Item Adjustments, Manual Refund.
Step 3 — Get Order Value: Pull the original order value from COIM as SUM(GMV + Tax) for each order number.
Step 4 — Classify: If Total Adjustment ≥ Order Value (GMV + Tax) → Full Missing Order. Otherwise → Partial Missing Order.
Step 5 — Promo Overlay: Check if the same incident (REF_NUM) also received a Promo Code appeasement, and whether it was redeemed.
How many missing-item incidents resulted in a full order adjustment vs partial?
Among missing-order cases, how many also received a promo code alongside the refund/replacement?