Internal Analysis
The fairest possible test: hold the season constant and vary the year. April–June ran on Website Purchase in 2024 and 2025, but on the custom Purchase – New Customer event in 2026. If the event hurts efficiency, 2026's Apr–Jun should look clearly worse than the prior two years. It doesn't. On CPM, CTR, and cost-per-order, the all-New-Customer quarter lands right in the normal range.
Apr–Jun cost-per-order: 2024 (Website Purchase) $125 → 2025 (Website Purchase) $108 → 2026 (New Customer) $122. The New-Customer quarter sits between the two Website-Purchase quarters — essentially tied with 2024.
The CPM step-up predates the event. CPM roughly doubled from $21 (2024) → $39 (2025) — both years running 100% Website Purchase. 2026 ($38) is flat vs 2025. So the expensive-impressions story is auction inflation over time, not the delivery event.
CTR is the best of the three years in 2026 (1.01% vs 0.58% / 0.93%). Three independent methods — the 2-year correlation report, the within-month head-to-head, and now this seasonal isolation — all land in the same place.
The one caveat that keeps your hypothesis alive: this measures cost per order. The New-Customer event's whole job is the new-customer mix, and the original symptom was new-customers/day (600→400). If the custom event finds orders but not genuinely new customers, the damage would live in cost-per-new-customer — invisible to everything here. That's the test that could still convict the event (see bottom).
The cleanest cut: April through June only, compared across 2024, 2025, and 2026. The event regime is the big thing that changed in 2026. Account-level (matches the figures the audit cited); a purchase-campaigns-only cut is within a dollar or two and tells the same story.
Read it: if the custom event were structurally inefficient, 2026 would break away from the pack. Instead its CPO ≈ 2024, its CPM ≈ 2025, and its CTR is the strongest of the three.
X-axis is month of the year; the three lines are 2024, 2025, 2026. The shaded band is Apr–Jun — the event-isolation window (2026 = New Customer; prior years = Website Purchase). Look at whether the purple (2026) line breaks away from the others inside the band. It tracks them.
Earlier the standout signal was that 2026's cost-per-order rose +32% from Q1 to Q2 — steeper than prior years — right as New Customer went all-in. But the seasonal view shows that's driven by an unusually efficient Q1 2026 (the Website-Purchase "pullback" quarter), not an abnormally bad Q2.
Q2 cost-per-order is ~$122 in 2026 vs $125 in 2024 — flat. The big "+32%" is the low Q1 base ($93, the best Q1 of the three years), so it's mean-reversion to a normal Q2, not event damage.
Everything above is cost per order. MUD\WTR's real KPI — and the audit's original symptom — is new customers (600→400/day) and blended new-customer CAC. A custom "new customer" event could keep cost-per-order flat while quietly degrading the share of orders that are genuinely new (e.g., the Elevar signal misfiring). That damage is invisible here.