How do we find the spend level where Meta's marginal return falls below our target as budget rises?
Our answer
Run incrementality tests at current Meta spend and at a higher level, and judge the added spend on its own. Stop raising spend where its return falls below your target.
- Confidence
- Medium
- Strongest proof
- 1 controlled test
- 1 Meta rule, 3 practitioners, 6 opinions or claims
- Sources
- 10 sources
- 11 statements
- R
HGNRMRD
New here? Take the 1-minute tour
Why this is the answer
3 reasons and our judgmentAverages hide the expensive last dollars. Marginal return usually falls before the average does (PMG). In Wayflyer's illustration, one campaign went from a ROAS of 4.2 times at $2K a day to 2.1 times at $6K a day.
Peaks can push spend past where added spend pays. Ridge's scale-up test showed it was overspending around Black Friday, with worse total sales divided by total ad spend (MER). HexClad runs a scale-up holdout before Black Friday to check whether spending up is incremental.
Mix-model curves need tests behind them. Meta's Robyn documentation cites a finding that uncalibrated models missed lift-test results by 25% on average, and Grüns calibrates its mix model with test results.
Our judgment
Our stop point is the return on the added spend, read from incrementality tests, against your own target, which comes from your margin plan (The business metric that governs spend). We drew it from PMG's stop rule and Grüns's use of tests and marginal return for budget calls. We add re-testing before peaks, from Ridge and HexClad. We treat mix-model curves as a guide to which spend levels to test, not the answer, given the Robyn documentation's calibration finding, CTC's caution on spend levels outside the historical range, the identification result from Dew, Padilla and Shchetkina, and Kaminsky's warning that model runs disagree.
Our judgment, built from PMG, Grüns, Ridge, HexClad, Robyn (Meta open-source MMM), Common Thread Collective, Ryan Dew, Nicolas Padilla and Anya Shchetkina (academic) and Michael Kaminsky (Recast): our reading of these sources, not a finding from any one of them.
What the sources say
11 statements from 10 sourcesMeta's Robyn mix model docs say ad returns diminish, so budgets should follow the next dollar's response, and models should be calibrated with lift or geo tests. A relayed whitepaper puts uncalibrated models 25% off.
“uncalibrated models show 25% average difference to the ground truth”
Gruns' growth head matches the tool to the decision: incrementality tests, a mix model and marginal return for big budget calls, and attribution and platform data for daily moves.
“I'm also not letting attribution decide whether Meta deserves another $500K.”
PMG, an agency, says marginal ROAS, the change in revenue over the change in spend, usually falls before average ROAS does. Stop adding spend once the next dollar no longer pays back after margin.
“Stop increasing a channel when the next dollar no longer pays back.”
Google's Meridian assumes media has diminishing returns. Marginal ROI is the return on the next dollar, and a channel counts as saturated when it gets small, with no threshold set. Experiments can calibrate it.
“is a general term, and no specific threshold is defined.”
Ridge's CMO says a scale-up test showed Ridge was overspending around Black Friday and getting worse MER, so more spend can lower profit even when the totals look good.
“During Black Friday, Cyber Monday, you can spend buckets of non-incremental dollars and still be happy with the outcome … just because the numbers are so big.”
HexClad's Connor Rolain says the team does not know how incremental peak-day spend is, so before BFCM it is running a short-term scale-up holdout test. This is a plan, not a result.
“So before BFCM we're running a short-term scale-up holdout test.”
Connor MacDonald says Ridge starts incrementality tests on its top two channels three weeks before Black Friday to see where the next dollar should go.
Michael Kaminsky, whose firm Recast sells mix models, says most businesses do not need one and it can mislead them; he cites studies where last-touch was about 85 percent as good as experiments in simpler mixes.
“most businesses do not need an MMM. In fact, an MMM will be net negative for them because it will actively mislead them.”
Wayflyer's illustrative examples show ROAS roughly halving when daily budgets tripled, such as 4.5x to 2.4x in two weeks; it advises raising budgets gradually and judging the next $500 a day.
CTC warns that predicting efficiency at spend levels well outside your historical range should be treated with caution.
“predicting efficiency at spend levels well outside the historical range should be treated with caution.”
This academic paper says standard mix data often cannot tell diminishing returns from effects that change over time, which point to different budget splits. It proposes experiments that vary spend.
“nonlinear and time-varying effects are often not identifiable from standard marketing mix data”
In the full answer
For readers who sign in- What to do this week3 steps
- The numbers behind it2 figures from cited sources
- The strongest disagreementThe best case against the answer, from a cited source
- How sure we areWhat its confidence rests on, and what would change the answer