d
INVESTIGATIONS

Weekend revenue drop

INVESTIGATION · SUPERVISING · 5 PARALLEL CUTS

Weekend revenue is down 12.4 percentage points WoW vs the prior 8-week trend. Find what's driving it — and only the things that survive counter-evidence.

47
CUTS TRIED
5
ALIVE NOW
0.92
BEST EXPLAINED
18
PRUNED
YOU 16:02:00
Why is weekend revenue down 12.4pp WoW? Finance wants an answer by Monday.
d
I116:08:55·ORTHOGONAL SWEEP · 47S WALL TIME
confidence 78%

Acquisition channel is the strongest signal — it alone explains 0.78 of the gap.

I ran six orthogonal cuts on the 12.4pp drop in parallel: region, plan tier, channel, surface, cohort, day-of-week. Five came back weak — nothing above 0.34. Channel returned 0.78, and the rest dropped out cleanly.

That's enough to commit to channel as the trunk and prune everything else. Day-of-week is symmetric across Sat/Sun, so this isn't a single-day artefact.

NEXT

Channel is the trunk. Pruning region, surface, cohort, and day-of-week — drilling into paid channel next.

CITED⊞ finance.daily_revenue⊞ paid.campaign_daily+2 more
YOU 16:11:55
Channel — drill in.
d
I216:18:42·DRILLING CHANNEL · 4 PARALLEL CUTS
confidence 71%

Three weekend-creative campaigns account for 89% of the channel drop.

Inside paid, two platforms move the number — Meta and TikTok. Below that, three campaigns: ME-1842, ME-1851, TT-0317. All weekend-creative variants. Their share of paid revenue collapsed roughly 41% on weekends only.

Creative mix is stable, attribution-window change explains some but not all (0.31). The campaigns themselves are the carrier.

NEXT

Don't tell finance yet. Run counter-evidence on the three campaigns — YoY seasonality, pixel firing, bots, and pricing-experiment overlap.

CITED⊞ paid.campaign_daily⊞ meta_ads_export.parquet
YOU 16:24:11
Before we tell finance, check counter-evidence and find the timing.
d
I316:36:50·TIMING TRACE · ALL OF PAID
confidence 85%

The drop is a step, not a slope — it opens cleanly on iso-week 43 (Oct 26).

Plotting weekend revenue against the trailing-8-week mean, the gap appears as a single discontinuity, not a decay. That alone rules out the gradual stories — creative fatigue, brand drift, audience saturation — without any extra evidence.

Whatever happened, it happened on or right before Oct 26. I'm looking at the ops.changes log for that 48-hour window.

NEXT

Looking at ops.changes for the 48-hour window around Oct 25. A single config change is the most likely story.

CITED⊞ finance.daily_revenue⊞ ops.changes
d
I416:48:33·ROOT CAUSE · 4 COUNTER-TESTS REJECTED
confidence 92%

Someone halved the weekend Meta budget on Oct 25. That's the whole story.

ops.changes returns one row in the window: change_id=48211, by e.miller, 16:42 UTC Oct 25, scope='weekend', entity='meta_budget' — value halved. Three Meta/TikTok campaigns running on that cap are exactly the three I isolated earlier.

Four counter-tests rejected (seasonality, pixel firing, bot share, pricing-experiment overlap). Three convergent streams accept (O021 campaign, O030 YoY, O041 timing). Confidence 0.92.

NEXT

The right next step is a one-weekend pilot of the restore. Counterfactual O051 says +9.4pp; breakeven is +6.0pp. Worth doing before any global change.

CITED⊞ ops.changes⊞ finance.daily_revenue+2 more
YOU 17:02:11
I'm off this weekend. Pilot the restore on one weekend before we change anything globally.
d
I509:12:48·PILOT · MID-TEST
confidence 88%

Restore pilot tracking +8.7pp lift vs control. Model said +9.4pp.

Counterfactual O051 said restoring the budget would lift weekend revenue by +9.4pp (80% CI [+7.1, +11.0]). Pilot O060 is running in 12 metros, 4 control, partway through one weekend — currently at +8.7pp, well inside the band.

If it holds at the read in ~38 minutes, the expected business impact is +$1.27M/week in weekend revenue. I'll pin the recommendation when the test closes.

NEXT

Hold for the full pilot read at ~17:30. If it lands at or above +6.0pp, expected business impact is +$1.27M/week. I will send the recommendation to finance the moment it closes.

CITED⊞ finance.daily_revenue⊞ paid.campaign_daily