Verdant Outdoor: Platform ROAS 4.2, Analytics ROAS 1.9, and a CFO With a Knife
Verdant Outdoor sells hiking apparel online. Last quarter it spent $150,000 on Meta. The platform reported a 4.2 ROAS; the analytics tool attributed $285,000 of revenue to paid social, a 1.9 ROAS. Total revenue across all channels was $1.9M against $420,000 of total marketing spend. Over the past six weeks CPM rose 40%, frequency reached 6.8, and click-through fell from 1.4% to 0.8%. Contribution margin after product and shipping is 52%. The paid social lead wants to double the budget on the two best-performing ad sets, citing the 4.2. The CFO wants to cut Meta to zero and put everything into branded search and email, which show very high attributed ROAS.
1. Reconcile the two ROAS figures and say what the 52% margin implies about the correct decision.
Read the worked answer
The two numbers measure different things and neither is incrementality. Platform ROAS credits view-through and modeled conversions inside a short window, including purchases that would have happened anyway; it claims 150,000 x 4.2 = $630,000. The analytics figure of $285,000 is last-click or last-non-direct, which systematically undercounts upper-funnel prospecting that later converts through email or branded search. Both are estimates with opposite biases.
Set the breakeven bar. At 52% contribution margin, breakeven ROAS is 1/0.52 = 1.92. Meta at an attributed 1.9 is already at or marginally below breakeven. And because attributed revenue usually exceeds incremental revenue (it absorbs some organic cannibalization), the defensible prior is that true incremental ROAS is at or below 1.9, not 4.2. The burden of proof sits on the channel.
That does not justify cutting to zero. Cutting abruptly resets ad-set learning, forfeits the halo that shows up in direct and branded search, and the CFO's high-ROAS alternatives are precisely the channels most inflated by attribution - branded search in particular harvests demand someone else created.
Read the fatigue signals as the real finding. Frequency 6.8, CPM +40%, CTR halved from 1.4% to 0.8% describe a channel at saturation with creative that has stopped working, not a channel that never worked. The paid lead's plan to double budget on the two best ad sets is exactly backwards: those are the most fatigued assets.
Correct move this quarter. Refresh creative, cut spend 20-30% to a floor while holding a measurement cell, and run a geo test. Set the decision rule before data arrives: keep or scale if incremental ROAS clears 1.92; cut toward zero if it does not.
2. Design the incrementality test. What would make it uninterpretable, and what would you do with the result?
Read the worked answer
Design. Build 40 matched DMA pairs on 12 months of baseline revenue, seasonality and channel mix. Randomize within pairs - one DMA of each pair goes dark on Meta for four weeks, the other continues as normal. Hold search, email and pricing constant in both cells; any change there contaminates the read. Primary metric is total revenue across all channels per DMA, not Meta-attributed revenue, because the point is to capture the halo. Pre-register the analysis, the metric, and the decision thresholds before launch.
What makes it uninterpretable. Too few or poorly matched DMAs, so baseline variance swamps the effect; changing creative or budget in the control cell mid-test; running during a promotional period that differs by market; measuring sales rather than revenue so discounting masks the lift; stopping the test early when the weekly numbers look good. Be explicit about the minimum detectable effect - with retail revenue volatility, expect to detect roughly a 4-5% revenue shift. If the true effect is 2%, this test will not see it and you must not call a null result proof of zero.
Interpreting the result. Compute incremental revenue as the dark-out shortfall scaled to the spend removed, and compare that iROAS to 1.92.
- iROAS clearly below 1.92: the CFO is right on economics. Cut to a maintenance floor and reinvest in retention.
- iROAS between 1.92 and 3: hold budget, fix creative, stop reporting platform ROAS internally.
- iROAS above 3: the channel is under-invested; scale in 20% increments while monitoring frequency, since the fatigue evidence says headroom is finite.
Whichever way it lands, replace the internal ROAS dashboard with marginal iROAS plus blended MER (1.9M/420k = 4.5) so future decisions use one consistent currency.