Incrementality testing for big spenders

Quick Answer

Above $10k/day spend, attributed ROAS becomes increasingly misleading. Incrementality testing — measuring the actual lift Facebook ads create vs what would happen without them — is the only reliable way to know what's working. The two methods that work: geo-based hold-outs (turn off ads in one region, compare to others) and Meta's built-in conversion lift studies. Both take 2-4 weeks and disrupt revenue temporarily, which is why most accounts skip them — and pay for it in misallocated budget.

The Framework

1. Why attributed ROAS lies at scale

Attribution credits Facebook with conversions that would have happened anyway. The bigger your spend, the more this gap grows. At $5k/day, attribution might overstate by 20-30%. At $50k/day, it can overstate by 80%+.

2. Geo hold-out tests (the practical method)

Pick a geo with at least $1k/day in spend. Pause Facebook ads in that geo for 4 weeks. Compare total revenue from that geo vs a similar control geo where ads continue. The difference is your true incremental lift.

3. Meta's conversion lift studies (the official method)

Meta runs randomised studies where some users see your ads and some don't, then measures the difference. More statistically rigorous than geo tests but requires Meta rep coordination and won't run for accounts below a certain spend threshold.

4. The minimum test duration

Run for 4 weeks minimum. Below that, weekly variance and seasonality will dominate the signal. 4 weeks is enough to see real lift and ride through normal noise.

5. Set expectations for the cost

Hold-out tests will reduce revenue temporarily. Budget for it. The cost of running the test is much smaller than the cost of misallocating $100k+/month based on false attribution.

6. Re-test annually (or after major changes)

Incrementality drifts. After major creative refreshes, audience expansions, or geo launches, re-test. Past lift doesn't predict future lift if the underlying setup has changed.

Real Numbers from the Field

A health brand spending $18k/day reported a 4.1 blended ROAS via Facebook attribution. We ran a 4-week geo hold-out in their UK market (their second-largest geo). UK Facebook spend went to zero. Revenue from UK dropped 28% — meaning 72% of attributed revenue would have happened anyway. Real incremental ROAS was 1.15, not 4.1. They restructured budget allocation across channels and recovered 40% of their previous Facebook spend back into channels with higher true incrementality.

Frequently Asked Questions

How much revenue will I lose during the test?

10-30% of revenue from the hold-out geo, for 4 weeks. Plan and budget for it — don't try to run hold-outs in revenue-critical periods.

Can I use Meta's built-in lift studies instead of geo hold-outs?

Yes, if your account qualifies. Lift studies are more rigorous but harder to set up. Geo hold-outs work for everyone.

What if I can't afford to lose revenue for 4 weeks?

Use a smaller geo or run the test in a slow period (post-Christmas, mid-summer for some categories). Don't skip the test entirely — the alternative is misallocating much more money long-term.

Will my reported ROAS match incremental ROAS?

Almost never. Reported ROAS is usually 2-5x higher than incremental. The gap is normal — just account for it in budget decisions.

Should I test channel-by-channel or aggregate?

Test Facebook independently first. Then test other channels separately. Combined tests can't distinguish which channel is driving lift.

Stop Guessing What Your Ads Look Like at Scale

When you're spending £5k+ a day, you can't manually screenshot every ad on every device, in every placement, in every locale. Things break silently. Creators ship the wrong aspect ratio. Localised text overflows. Safe zones get clipped by Reels UI.

Pix-Vu lets you preview every ad creative across Facebook, Instagram and Reels placements before launch — desktop, mobile, all aspect ratios, all crop variants. Spot the broken creative before you spend £20k discovering it the hard way.

Try it free at https://pix-vu.com — built for media buyers who can't afford a single wasted impression.