Let’s start with the “bad” news: the true incremental ROI of any campaign is unknown and unknowable. Marketing performance lives in a world of noise, lags, and missing data. We would all love to be able to report numbers like “3.7x ROI” as if they were the objective truth – but they are not.
Unfortunately, many marketing teams are still reporting on point estimates (a single ROAS or CPA), and this is very dangerous when you’re using them to justify spend decisions worth millions of dollars.
This is also the easiest way to lose trust with your CFO. We’ve seen the pattern play out over and over again:
- Quarter 1 – “We’re driving a 5x ROAS on Meta prospecting. Tons of room to scale.”
- Quarter 2 – Spend scales, but revenue misses forecast.
- Quarter 3 – Another miss. CFO starts asking harder questions.
- Quarter 4 – “Whatever numbers marketing brings me, I multiply by 0.7 and subtract another 20%.”
The problem is that point estimates create a false sense of precision. Every number you present is just an estimate that comes with assumptions and is bound by uncertainty – but you will barely see that nuance on your report decks.
This article explains why forecast ranges matter, how to build them into your MMM, and how to communicate them to finance.
Why Marketing Lift Doesn’t Tell You If a Channel Is Profitable
We had a client recently who ran an incrementality study and came back excited: “we saw a 3% lift and it’s statistically significant!” But once we took a closer look, it turned out that lift translated into a CPA with an uncertainty range from $25 to $750. The reported $75 CPA was just the midpoint.
If your profitability threshold is a $150 CPA, then the channel might be highly efficient… or it might be burning your budget. A midpoint alone doesn’t tell you that. The full range does.
This is counterintuitive because most marketers have been trained to look for statistical significance in A/B tests. That mindset doesn’t map well to measuring marketing performance. The question isn’t “is there any lift?” because most marketing generates at least some lift. The real question is: Is that lift profitable? And to answer that, we need to convert lift into ROI or incremental CPA and, very importantly, know what the range looks like.
How to Build and Use Forecast Ranges in MMM
Any statistical model worth using should represent the uncertainty it picks up. At Recast, we generate ROI curves that include credible intervals by default.
Here’s how it works:
We start with very uninformative priors – say, that a channel’s ROI could fall anywhere between 0x and 20x. Then, when we have experimental results (like a geo-lift test), we use those results as informative priors but only during the time the test was actually run.
If we have an experiment from May 1 to May 15 that says incremental ROI was between 4.5x and 5.5x, we use that informative prior only during the test period. That tells Recast that ROI curves can go up and down, but they must pass through those test results.
This helps us avoid the two big problems we see in other MMMs: they straight up ignore uncertainty intervals, and they assume lift test results are permanent. An ROI estimate from a test window doesn’t necessarily apply weeks or months later when the context has changed.
And if your MMM vendor doesn’t give you access to forecast ranges? Red flag. Ask directly: “Can I see the ROI range, not just the average?” If the answer is no, you’re not getting the full picture (and you should probably run away!).
One final recommendation we tend to give here is to plan against the lower bound of your ROI range. You want to take action based on the data, but you don’t really want to be over-optimistic in case it backfires.
Better Decisions Come From Ranges + Clear Recommendations
The root misconception that we’d like to help overcome is that uncertainty in the data means that you can’t make decisions and that showing a range of outcomes will make you look indecisive.
Great marketing leaders show a range and then suggest an actionable next step to it. They ask:
- What’s the best-case ROI?
- What’s the worst-case scenario we still find acceptable?
- How would the forecast shift if we landed at the low end?
And then they tell the CFO: “the data supports a 2.2x to 3.8x ROI. We’re going to forecast against 2.4x and increase spend by 25% because it’s within our profitable benchmark.”
But if you just report 3.7x, and your CFO plugs that into a financial model, you’re setting up the business (and yourself, and the CFO) for failure.
Your CFO wants clarity – not certainty. Certainty is impossible. Give them the data ranges and a clear recommendation so they’re not blindsided when results vary.
