**To:** Aisha Rahman | **From:** Ingrid Solberg | **Re:** Loyalty economics — revised verdict

**Headline: the "members underperform" story was a composition problem twice over. The real dividing line is engagement, not enrollment.**

**1. Engaged members (those who've ever redeemed a point) genuinely outperform similar non-members — this is the verdict you asked for.** Within Apparel, tenure-controlled: at 24+ months, "ever-redeemed" members average 10.0 orders and €1,873 revenue vs. 5.3 orders and €1,013 for non-members — with refund rates essentially tied (39.1% vs 39.5%). The drag comes entirely from the 83% of members who've **never** redeemed a single point — they trail both engaged members and non-members at every tenure band (12–24mo: €468 revenue vs. €750 for non-members). *(Reports: "Apparel: never-redeemed vs ever-redeemed vs non-members, by tenure"; confirmed platform-wide in the same-named report without the vertical filter — identical pattern, e.g. 24+mo: 9.7 orders/€1,844 engaged vs. 7.9/€1,544 non-member vs. 3.3/€511 disengaged.)*

**2. No evidence of gaming — I consider this closed.** Nobody redeems within 30 days of enrolling (that bucket was empty), and the ~30% of engaged members who never order again after their first redemption is identical whether they redeemed fast or slow (29.5% vs 29.9%) — normal churn, not reward exploitation. *(Report: "Post-redemption drop-off by time-to-first-redemption.")* This closes out last week's unresolved question.

**3. The programme's real cost is concentrated, and richer than I reported last week.** The engaged 17% (225 members) have redeemed €177.8k against €434k of their own revenue — a **40.95%** effective discount rate, well above the 19.4% blended figure. The other 83% cost nothing in cash but carry €323.6k in paper liability that, per cohort-aging data, mostly never gets claimed (redemption rates plateau at 32–37% even for 3-year-old cohorts). *(Reports: "Loyalty unit economics: ever-redeemed vs never-redeemed"; "Loyalty points redemption rate by enrollment cohort age.")*

**4. The €491k liability is both a timing lag and a structural ceiling.** Redemption rate climbs from 4% (2-month cohorts) to 32–37% (2+ year cohorts), then flattens — so the liability keeps growing in euros as we enroll more people, but each cohort's eventual payout rate looks capped around a third of what's issued, not runaway.

**5. I chased a scary lead and it didn't hold up.** Order frequency fell sharply after a member's first redemption (6.64→2.79 orders/year), which looked like redemption killing engagement. But a neutral baseline (splitting each customer's own order history at its midpoint, not anchored to redemption) shows frequency **rising** for every group, including redeemers (6.40→7.17). The "drop" looks like a statistical artifact — people redeem right after a buying burst, then revert to normal — not a real cost of giving rewards. *(Report: "Order frequency deceleration by customer group.")*

**Confidence:** High that engagement is the real dividing line, that gaming isn't occurring, and the liability has a natural per-cohort ceiling. Lower on causality: we still can't prove redeeming *makes* someone a better customer vs. good customers simply being the ones who earn enough to redeem — everything here is observational.

**Caveats:** Platform-wide refund comparisons still mix verticals — engaged members are disproportionately Apparel shoppers, so trust the within-vertical numbers (near parity) over the raw platform ones (which show a gap again). Some platform-wide cells (n=6–12) are too thin to quote confidently.

**Next steps:** Run a real intervention — nudge a sample of the never-redeemed 83% toward their first redemption and see if they start behaving like organically-engaged members or don't. That's the only way to move from correlation to causation here, and it would directly test whether "drive engagement" is a fixable lever or just selection we're observing.