How should we split budget across channels using measurement results?
Our answer
Split budget on each channel's return on the next dollar, from your own tests or a test-calibrated model, not platform ROAS. Move money in steps from channels below breakeven to channels above it, and retest after each move.
- Confidence
- Medium
- Strongest proof
- 2 official rules
- 2 datasets, 11 practitioners, 6 opinions or claims
- Sources
- 13 sources
- 21 statements
G
New here? Take the 1-minute tour
Why this is the answer
3 reasons and our judgmentAverages hide what the last dollar buys. In Haus's example, Meta cost $337 per incremental customer on average but over $1,000 on its last $80,000, against $510 on YouTube. WorkMagic argues a 4 times average ROAS can hide a marginal ROAS below 1 times.
Breakeven is 1 divided by contribution margin, so a 40% margin means 2.5 times (Stella, an incrementality testing software maker). Stella sizes each budget step so the loss at the low end of the test's range stays within an agreed limit.
One test shouldn't move your whole budget. Common Thread Collective (CTC) found Facebook acquisition tests range from 0.5 to 2.4 times what the platform reports, so it moves only partway on one test. Google's Meridian caps channel spend so its optimizer can't pick one channel.
Our judgment
Rank channels against breakeven and move in capped steps rather than all at once. That combines Stella's step sizing, CTC's practice of moving only partway on one result and Meridian's channel spend limits. Retest each channel at its new spend before the next move, because the return on the next dollar changes as spend moves, as the Haus and WorkMagic examples show.
Our judgment, built from Stella, Common Thread Collective, Google, Haus and WorkMagic: our reading of these sources, not a finding from any one of them.
What the sources say
21 statements from 13 sourcesGoogle's Meridian docs say a fixed-budget scenario finds the split that maximizes ROI, while a flexible one finds the most you can spend before hitting a target ROI or marginal ROI.
Lendi Group's mix model showed free-to-air TV was underweighted, and its growth head says TV kept increasing its return. The article gives no return or budget figures.
“swallowed it up and kept increasing ROI”
Boathouse's strategy chief says that for one client an off-the-shelf mix model credited paid media with about 75% of new customers, while a custom model on the same data said under 1%. Check models with lift tests.
“A result below 1% deserves as much scrutiny as a result of 75%.”
Meta'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”
Stella, which sells testing, says a budget call should state direction, size and a check date, with breakeven at 1 divided by margin (a 40% margin means 2.5x). Its step-size examples are illustrative.
“Cutting a channel because the point estimate looked soft is how brands kill channels that were working.”
TechWyse, an agency, says cheaper Facebook clicks left cost per lead flat, and cost per lead shows nothing about how many leads become customers; judge Meta on cost per qualified lead and completed sales.
“A lower social CPC does not establish that the same money would produce more customers if removed from search.”
Silverback Strategies says lead gen brands should fund search to saturation first, then spend the next dollar on demand creation proven with lift tests of four to eight weeks, which small sellers often cannot run.
“Once demand capture is saturated, additional budget there produces diminishing returns, and the next dollar is better spent creating demand.”
WorkMagic, which sells mix modeling, says David Protein found its Meta attribution inflated by 36%, and after reallocating on incrementality revenue grew 34% and profit 37%. No method or period is given.
“David Protein found their Meta attribution inflated by 36%”
Sellforte advises moving budget by marginal incremental ROAS from response curves, within constraints such as peak weeks. Its demo data, not real results, shows platform and calibrated ROAS far apart.
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.”
CTC, which sells geo tests, says across hundreds of its tests Facebook acquisition's median incrementality factor is 1.14, with single tests ranging from 0.5x to 2.4x, so one test is not enough.
“single test results are insufficient”
CTC says even a well-calibrated incremental ROAS should answer to revenue and contribution margin; if iROAS improves while margin is flat or falling, recheck the measurement.
“Even a well-calibrated incremental return on ad spend should always be subordinate to the realities of revenue and contribution margin”
Across 225 geo tests on its platform, Stella, which sells geo testing, found Google branded search lowest, at a median 0.70x incremental ROAS (typical range 0.32x to 0.89x). Its brands are mostly US Shopify stores.
“Reduce to defensive floor (prevent competitor conquest), redirect savings to higher-iROAS channels.”
Google's Meridian budget optimizer splits a fixed or flexible budget across channels, with channel spend limits to avoid results such as putting all spend in one channel.
“Channel-level spend constraints are available for both fixed and flexible budget optimization to prevent unreasonable optimization outcomes, such as putting all spend into one single channel.”
LiftLab, which sells mix models and lift tests, says geo holdouts answer keep-or-cut questions, while budget planning needs the return on the next dollar from spend-level tests. No data.
“An incrementality test without a pre-committed decision is measurement theater.”
WorkMagic, which sells geo tests and mix models, says platform ROAS includes baseline demand, misplaced halo and double counting, and that budgets should follow marginal ROAS, which can be below 1x at a 4x average.
“A channel with a healthy average ROAS of 4x can have a marginal ROAS below 1x if it's past the point of peak efficiency.”
Haus says to compare a new channel with Meta's marginal cost, not its average. In its example Meta cost $337 per incremental customer on average but over $1,000 on the last $80,000, versus $510 on YouTube.
“Yes, $80,000 netted you 76 new customers, which is a CPIA above $1,000 and well above that $510 you measured on YouTube.”
Haus suggests measuring the return on the last 25% of Meta budget with back-to-back spend-level tests first, as the cheapest option, then 3-cell holdout or stacked-spend tests.
“Dean’s recommendation for most brands: start with sequential tests, because they're the least disruptive and still get you to a meaningful answer.”
Haus says a brand with 90% of budget on Meta carries algorithm and account risk, and suggests committing about 10% of the media mix to a new channel once tests show conviction.
“You might commit to reallocating this spend, or commit to putting 10% of your overall media mix into this channel that we have conviction in.”
Jones Road Beauty found Meta incremental but overspent, so it pulled back Advantage+ shopping spend. Branded search at about $3,000 a day was not incremental. No results after the changes.
“We try to have a live test going at all times.”
Haus says OluKai moved about half its view-through channel spend into Meta and cut total spend about 15% after tests, and CAC fell about 20%. The baseline for the 20% is not defined.
In the full answer
For readers who sign in- What to do this week3 steps
- The numbers behind it3 figures from cited sources
- Where each source standsA chart placing each source on the question
- The strongest disagreementThe best case against the answer, from a cited source
- What the research saysWhat academic research adds, kept apart from the answer
- How sure we areWhat its confidence rests on, and what would change the answer