Key takeaways
- Marketing efficiency metrics like return on ad spend (ROAS) and cost per acquisition (CPA) are useful starting points — but they can mislead you if you're comparing them without accounting for diminishing returns.
- Platform-reported attribution numbers routinely overstate performance, which means the efficiency reads you're relying on may not reflect what your marketing is actually doing.
- Average efficiency and marginal efficiency aren't the same thing. Confusing the two can cause you to pull budget from channels that are actually working.
- Incrementality testing and causal marketing mix modeling (MMM) give you a more reliable foundation for reading efficiency across channels.
- The goal isn't to track more metrics — it's to track the *right* ones, with the right methodology behind them.
If you've ever looked at your channel performance dashboard and felt something between mild skepticism and deep mistrust, you're not alone.
Marketing efficiency metrics are supposed to tell you which channels are working and which aren't. But when your ad platforms report one thing, your analytics tool reports another, and your CFO is asking for a clear picture of return on investment (ROI), it's easy to wonder if any of these numbers are telling the full story.
The short answer: sometimes they are, and sometimes they aren't. The longer answer — and the more useful one — is that efficiency metrics are only as reliable as the methodology behind them. This guide breaks down the most common marketing efficiency metrics, where they tend to go wrong, and how to build a measurement approach that actually holds up.
The efficiency metrics you're probably already using
Most marketing teams are tracking some combination of these:
- Return on ad spend (ROAS): Revenue generated per dollar spent on advertising.
- Cost per acquisition (CPA): How much you spend to acquire one customer.
- Cost per click (CPC): What you pay for each click on an ad.
- Customer acquisition cost (CAC): Total cost of acquiring a new customer, often broader than CPA.
- Return on investment (ROI): Net profit relative to total investment.
These metrics aren't wrong — they're just incomplete on their own. The problem isn't the metrics themselves. It's what they can't tell you: whether your marketing *caused* those results.
As we outline in Marketing measurement: What to measure and why, just because conversions roll in after a campaign launches doesn't mean the campaign caused them. Customers who would have converted anyway get counted. Platform algorithms take credit for purchases that were already in motion. And suddenly your efficiency numbers look a lot rosier than reality.
Why platform-reported efficiency diverges from actual performance
Every ad platform has an incentive to show you strong performance — and their attribution models reflect that. Last-click, view-through, and other platform-native attribution models tend to claim credit for conversions that would have happened regardless of the ad.
It helps to understand the incrementality fundamentals here. Imagine you're about to buy a pair of boots your mom specifically asked for as a birthday gift. You see an Instagram ad for those exact boots right before you check out. Did the ad cause the purchase? No — but Instagram's attribution model will count it as a conversion. As we describe in Incrementality: The fundamentals, these are exactly the moments that inflate platform efficiency reads and lead marketers to over-invest in channels that look efficient but aren't actually moving the needle.
Health and wellness brand Ritual ran into this head-on. They suspected their TikTok numbers were inflated, so they ran a controlled experiment using the Haus platform. They found that TikTok's platform reporting was overstating performance by about 10% — a finding that's since saved them millions in annual spend. That's what happens when you replace platform-reported efficiency with experiment-backed efficiency.
Average efficiency vs. marginal efficiency — and why the difference matters
Here's a subtler problem that even experienced marketing teams run into: comparing the *average* efficiency of one channel to the *average* efficiency of another.
Say you're spending heavily on Meta and considering diversifying into YouTube. You run an incrementality test on YouTube and measure a cost per incremental acquisition (CPIA) of $510. You compare that to your Meta CPIA of $337. Meta wins — so you pull back YouTube before it gets a real chance.
But as we explain in Diversifying the right way: A framework for calculating the marginal efficiency of your marketing channels, that comparison is flawed from the start. You shouldn't be comparing YouTube's efficiency to Meta's *average* efficiency — you should be comparing it to Meta's *marginal* efficiency.
Marginal efficiency measures the return on your last unit of spend, not your total investment. So what does that look like in practice?
In the original example, you were spending $320,000 on Meta at a $337 CPIA. When you scale Meta up by 25%, your overall CPIA rises to $392 — still within your efficiency guardrail of $400. Looks fine, right?
Look closer. That additional $80,000 generated only 76 new customers. That's a marginal CPIA above $1,000 — well above the $510 you saw on YouTube. So in reality, YouTube *was* the better investment. You just couldn't see it because you were comparing it to the wrong number.
"If you're not able to pinpoint the marginal return, there's no way to really methodically compare your diversified new channels against your primary one," says Haus Measurement Strategist Dean Gordon. "It doesn't really matter if YouTube or Snapchat or Reddit has a higher CPIA than Meta in totality — what matters is how it compares to the last dollars you're spending on Meta."
Tools like Haus Diminishing Marginal Return Testing are specifically designed to help you draw that marginal return curve — so you can find the point of diminishing returns before you waste budget on it.
Building a more reliable efficiency framework
So what does a better measurement approach actually look like? A few principles to anchor on:
Start with causality, not correlation. Efficiency metrics derived from causal experiments are more trustworthy than those pulled from platform dashboards alone. Incrementality tests — where you expose one group to your marketing and withhold it from another — let you isolate the true impact of a campaign. This gives your efficiency reads a causal backbone they otherwise lack.
Layer in Causal MMM. Incrementality testing is powerful, but you can't run experiments on everything all the time. That's where Causal MMM comes in. Unlike traditional marketing mix modeling, Causal MMM incorporates experiment results as ground truth to calibrate the model — giving you a more accurate view of how each channel contributes to business outcomes, even when you're not actively running a test on it. If you want to see how this plays out in practice, How to turn your MMM into a decision engine is worth a read.
Measure marginal efficiency, not just average efficiency. As the Meta/YouTube example shows, average efficiency can mask what's really happening at the margin. Before you reallocate budget, make sure you understand the incremental return on your last dollars spent — not just your blended return.
Build a measurement plan before you need one. The worst time to figure out how you'll measure a campaign is after it's already launched. Assembling a marketing measurement plan walks through how to design a measurement framework — with incrementality at the center — before you go to market.
Reconcile, don't ignore. Platform attribution and experiment-based measurement will often disagree. Rather than defaulting to whichever number is more convenient, use the gap between them as a signal. A large discrepancy between platform-reported ROAS and your incrementality-backed ROAS is telling you something important about where your measurement methodology might be inflating results.
The bottom line
Marketing efficiency metrics are only as useful as the method behind them. ROAS, CPA, and CPIA can all be meaningful signals — but they need to be grounded in causal measurement, interpreted at the right level of granularity (marginal, not just average), and cross-checked against experimental data before you use them to make budget decisions.
The good news: better measurement is more accessible than it used to be. Experiment-backed efficiency reads, causal MMM, and tools that surface marginal return curves are no longer reserved for brands with massive data science teams. They're table stakes for any growth marketer who wants to spend smarter.
Frequently asked questions
What's the difference between ROAS and incremental ROAS?
ROAS measures total revenue attributed to your ads divided by what you spent. Incremental ROAS measures only the revenue that *would not have occurred* without your advertising. Platform-reported ROAS counts conversions that were already in motion — people who would have bought regardless. Incremental ROAS, derived from controlled experiments, strips those out and gives you a truer read on what your spend is actually causing.
Why might ad platforms overstate performance?
Ad platforms use attribution models — last-click, view-through, and others — that are designed to claim credit for conversions, not to isolate causality. Because platforms have an incentive to show strong results, their models tend to count customers who would have converted anyway. The gap between platform-reported performance and actual incremental performance is often significant enough to change budget decisions.
How do I know if my marketing spend has hit the point of diminishing returns?
The signal is usually a rising average CPIA paired with a much higher marginal CPIA on additional spend. If scaling a channel by 25% moves your blended cost per acquisition modestly but the incremental customers generated from that extra budget are far more expensive than your guardrail, you've likely passed the efficient portion of your spend curve. Haus's Diminishing Marginal Return Testing is built specifically to map this curve so you can identify the inflection point before over-investing.
Can I run incrementality tests on every channel at once?
Not practically — and not without risking interference between experiments. Incrementality testing requires holding back budget or exposure from a control group, which creates real trade-offs. That's one reason Causal MMM is a useful complement: it uses experiment results as ground truth to calibrate a model that covers channels and time periods where you aren't actively running tests, giving you broader coverage without requiring a continuous experiment on every channel simultaneously.
When should I compare a new channel's efficiency to my primary channel's marginal efficiency rather than its average?
Always — but especially when you're deciding whether to invest further in an existing channel or diversify into a new one. Average efficiency on your primary channel reflects the blended return across all your spend, including the early, efficient dollars. Marginal efficiency reflects only what your next dollar would return. If a new channel's CPIA is lower than the marginal CPIA on your primary channel, it's likely the better investment — even if it looks worse on an average basis.
What is Causal MMM and how does it differ from traditional marketing mix modeling?
Traditional MMM uses statistical regression to estimate channel contributions based on historical spend and outcome data. The limitation is that correlation can masquerade as causation — a channel that happens to spend more during high-sales periods may appear more effective than it is. Causal MMM incorporates results from controlled incrementality experiments as calibration inputs, anchoring the model to known causal effects rather than pattern-matching alone. The result is a more reliable view of channel contribution that holds up when used to guide real budget decisions.
How should I handle the discrepancy between platform-reported numbers and experiment-backed results?
Treat the gap as a signal, not noise. A large discrepancy between platform ROAS and your incrementality-backed ROAS typically means the platform's attribution model is taking credit for organic or cross-channel conversions. Rather than averaging the two or defaulting to whichever number is more convenient, use experiment results as your anchor and work backward to understand what's driving the inflation. Over time, tracking this gap channel by channel helps you build a more calibrated picture of where platform reporting can be trusted — and where it needs to be discounted.
Related reading
