Internal Report
Over two years MUD\WTR shifted its Meta delivery optimization from the standard Website Purchase event to a custom Purchase – New Customer event. This report measures that shift at the ad-set level — month by month — and tests whether it correlates with CPM, CTR, CAC, sessions, and purchases. Short version: the shift is dramatic and real; its effect on efficiency is mostly hidden by seasonality, with one yellow flag worth watching.
What changed. New Customer optimization went from 0% of spend (it didn't exist) to 96% in Q2 2026. Standard Website Purchase fell from ~98% to ~1%. The switch wasn't a clean cutover: it ramped in Q4 2025, pulled back in Q1 2026, then went all-in from April 2026.
What moved with it. In the window where the mix actually varies (Jun 2025 on), heavier New Customer optimization tracks higher CAC (r = +0.67) and fewer sessions (r = −0.64). Taken alone, that looks bad.
Why we're not sounding the alarm. The all-in period is Q2 — historically your weakest quarter. Seasonally controlled, Q2'26 CAC ($122) sits between Q2'24 ($125) and Q2'25 ($110) — not an outlier. The one genuine flag: the Q1→Q2 CAC rise was +31% in 2026 vs +22% (2024) and +8% (2025), a steeper-than-normal deterioration that lines up with the April all-in. Suggestive, not proof.
Share of monthly Meta spend by the delivery optimization event each ad set was set to. New Customer optimization didn't exist before mid-2025; by spring 2026 it's nearly the entire account. The choppy path — ramp, pullback, all-in — suggests the team was testing rather than flipping a switch.
Each panel is a single metric over the full window. The purple band behind every chart is New Customer % of spend — so you can eyeball whether a metric bends when the optimization mix bends. Read these as "does the line track the band?", not as proof of cause.
CTR and CPM were already trending up through 2024–25 (cheaper, Reels-heavy clicks) before New Customer optimization existed — so their full-period correlation with the shift is largely coincidental timing, not causation. CAC and sessions are the panels that bend most around the 2026 all-in.
The honest test. Because New Customer % is ~0 for the whole first half of the timeline, it doubles as a "are we in 2026 yet?" variable — anything seasonal will correlate with it by accident. Comparing the same quarter across years strips season out. If New Customer optimization were inflating CAC, Q2'26 (96% NC) should stand out against Q2'24 and Q2'25 (both 0% NC). It doesn't.
Q1→Q2 CAC change: +22% (2024) · +8% (2025) · +31% (2026). Q2 is always softer than Q1, but 2026's drop-off is the steepest — the only metric signature that points, weakly, at the all-in New Customer switch adding pressure on top of normal seasonality.
Pearson correlation between New Customer % and each metric. The full-period column spans all 30 months; the active-period column uses only the 13 months where the mix actually varies (Jun 2025 on). They disagree sharply — and the active-period column is the more trustworthy one, because the full period is dominated by unrelated two-year trends.
|r| < 0.3 ≈ negligible · 0.3–0.5 ≈ weak/moderate · > 0.5 ≈ moderate/strong. None of these establish causation — they're 13–30 monthly points with heavy confounding (seasonality, creative refreshes, budget scaling, and immature attribution on the most recent weeks).
Run a geo or budget-split holdout: a matched set of campaigns on Website Purchase vs Purchase – New Customer for 4–6 weeks, judged on blended new-customer CAC (Daily Spend Tracker ÷ Shopify new customers), not Meta-reported purchases. That's the only way to separate the optimization choice from season, creative, and attribution noise. I can spec it.