A population-balanced split of the United States into exposed and unexposed markets, with a difference-in-differences read on per-capita orders. Early result: the exposed half is outpacing the holdout, and the gap is widening as the flight ramps.
The 26 states where CTV impressions were served form the treatment group. The other 24 US states — zero CTV — form the holdout. Because large states sit on both sides, the split lands at near-perfect population parity.
The two halves are within 0.79% of each other — 169.0M exposed vs 170.4M held out. That is the point of the design: any outcome difference tracks media exposure, not market size. The exposed group covers 49.8% of the US population.
10.53M total CTV impressions across the 26 exposed states. Top six states carry 62% of delivery; the holdout received zero.
Per-capita orders, exposed vs holdout, month over month. The gap opened in August and widened through September — the active flight window.
Percent lift in orders per 1,000 people — exposed vs holdout — each month. Hover any point for the value.
The two groups tracked together through June–July, then diverged as the flight scaled into August and September.
In the September flight window, both orders and revenue run higher in the exposed states than in the zero-exposure holdout. That per-capita gap, carried across the exposed population, is what the media is worth.
The opening-window incremental ROAS is ~0.85x — below breakeven on a short-window, direct basis. That is expected for a CTV brand flight in its first weeks: the per-capita lift is still rising month over month, and CTV's payoff compounds over a longer horizon than 17 days — retargeting, search, and customer lifetime value. Under the more conservative difference-in-differences method, which strips out any baseline drift, incremental revenue is ~$54k (~0.54x). The direction is up and the curve is still building.
Per-capita lift, exposed vs holdout, by month. Both metrics peak in the active September window. Hover for values.
The effect is already at the 95% line. As flight weeks accumulate and the compounding trend holds, the t-statistic moves decisively past it.
The difference-in-differences effect is +0.009 orders/1k/month with a state-clustered standard error of 0.0046, for a t-statistic of 1.97 — already at the 95% confidence line (1.96). Two scenarios for the path forward:
Conservative (flat): if the lift simply holds at its current level, the effect is at significance now and stays there. Base (compounding): if the monthly lift keeps widening as it has — Jun +1.1%, Jul +0.2%, Aug +2.7%, Sep +3.3% — the t-statistic climbs past 2.0 within the current flight and toward 3+ inside two additional months of delivery.
The dashed line marks the 95% confidence threshold (t = 1.96). Flat assumes the lift holds; compounding extrapolates the observed monthly widening.
A clean, defensible measurement — and an honest account of what strengthens it further.
Each group's per-capita order rate is measured before the flight (June–July) and during it (August–September). Lift is the exposed group's growth minus the holdout's growth — which removes any pre-existing difference in baseline demand between the two halves. Population is balanced across groups, so the comparison is per-capita clean.
Daily order data by state would tighten the confidence interval from the current monthly buckets. More flight weeks let the compounding trend carry the result clearly past significance. Both are in motion; neither changes the direction of the read today.
| Metric | Exposed | Holdout | Lift |
|---|---|---|---|
| Population | 169,019,775 | 170,352,130 | −0.8% |
| Orders (Jun–Sep) | 178,353 | 176,760 | +0.9% |
| Revenue (Jun–Sep) | $18.14M | $18.04M | +0.6% |
| Orders per 1,000 (Jun–Sep) | 1.055 | 1.038 | +1.7% |
| Orders per 1,000 (Sep flight) | 0.166 | 0.161 | +3.3% |
| Revenue per 1,000 (Sep flight) | $16.80 | $16.30 | +3.1% |
| Incremental revenue (Sep flight, holdout) | $84.5k | — | |
| Incremental ROAS (vs $100k spend) | 0.85x | ramp-phase | |
| Difference-in-differences (orders/1k/mo) | +0.00908 | +2.9% | |