The counterfactual: proving marketing caused the sale
Almost every marketing claim hides one missing number: what would have happened if you’d done nothing. That number is the counterfactual, and once you start asking for it, “the number went up” stops sounding like proof and starts sounding like a question.
By William Walczak — CEO, Hiilite Creative Group Inc.What a counterfactual actually is
A counterfactual is the estimate of what would have happened if you had run no marketing — the do-nothing baseline. It is the world that didn’t happen: the sales you’d have made anyway from repeat buyers, word of mouth, seasonality, and people who were already going to find you.
The reason it matters is simple. You can only see one version of reality — the one where you ran the campaign. To know whether the campaign did anything, you have to compare that reality against the one where you didn’t. Lift is the gap between the two. No baseline, no lift — just a number floating in space.
This is the whole game behind honest measurement. When we talk about incrementality or read a result off the Measurement Ladder, we’re really asking one question over and over: compared to what?
Why “sales went up after we advertised” isn’t proof
Picture a hardware store that runs ads in May. Sales climb 30%. The owner credits the ads. But May is also when everyone starts their garden, fixes the fence, and reopens the cottage. Some of that 30% was coming regardless. Maybe all of it.
This is the difference between correlation (two things moving together) and causation (one thing making the other happen). The ads ran, sales rose — that’s correlation. Whether the ads caused the extra sales is a separate claim that needs evidence the raw before-and-after number can’t give you.
The trap is everywhere because it feels intuitive. We see B follow A and our brain fills in the cause. A counterfactual forces the harder question: how much of B would have shown up with no A at all? For a deeper look at why the same tactic produces different answers in different conditions, see why the same play works differently.
Four practical ways to estimate the baseline
You can’t observe the do-nothing world directly, so you estimate it. These are the workhorse methods, from cleanest to most circumstantial.
Pick the one that fits your budget and risk tolerance. A randomized holdout is the gold standard; the others are useful when a clean experiment isn’t practical.
- Randomized holdouts — split your audience at random: most see the marketing, a held-out group sees nothing. Because the split is random, the holdout group is your counterfactual. Whatever they buy is the baseline; the difference is your lift.
- Geo-tests — run the campaign in some regions (say, three cities) and deliberately go dark in similar ones. The dark regions stand in for “what would have happened.” Good when you can’t split individuals but can split a map.
- Synthetic control — when you don’t have clean comparison regions, build a fake one by blending several untreated areas into a weighted “synthetic” twin that tracked your market before the campaign. After launch, the gap between you and your twin is the estimated lift.
- Difference-in-differences — compare the change in a treated group to the change in an untreated group over the same period. By subtracting the untreated group’s drift, you strip out shared trends like seasonality and leave the part the campaign added.
How to read a counterfactual claim without being fooled
Most of marketing measurement is a fight over how good the baseline is, not whether one exists. A randomized holdout gives a baseline you can trust because nothing else differs between the groups. A synthetic control gives an educated estimate that’s only as good as the pre-period fit. Knowing which you’re looking at tells you how much weight the “lift” number can bear.
Watch for the common dodges. “Conversions doubled” with no comparison group is a before-and-after story, not a counterfactual. Attribution that hands 100% of credit to the last click assumes the customer would never have bought without that ad — an implied counterfactual that’s almost always wrong. Model-based estimates such as marketing-mix modelling infer a baseline statistically rather than testing it, which is reasonable but weaker than a live holdout.
None of this means you need a PhD to act. It means you ask one question of every result: what is this being compared against, and how was that comparison built? If nobody can answer, treat the number as a hypothesis, not a fact.
What this means for a small business
You don’t need to run experiments on everything. You need to insist on a baseline before you believe a win — especially before you scale spend on the strength of it. “The number went up” is the start of an investigation, not the end of one.
Start small and cheap. Hold out 10% of an email list. Pause ads in one quiet region for a month. Run a new offer in two locations and not the other two. Even a rough counterfactual beats none, and it’s usually far cheaper than the wasted budget it saves. Our incrementality calculator and measurement-method picker help you size a fair test for your traffic.
This habit fits inside a larger loop — sense what’s happening, seize a tested bet, transform what works into a repeatable capability. If you want the full version, read Sense–Seize–Transform for small business and the Growth Mapping framework.
Keep reading
Questions, answered
What is a counterfactual in marketing, in plain terms?
It’s the estimate of what would have happened if you ran no marketing at all — the do-nothing baseline. Your real results only count as “lift” when they beat that baseline. Without it, you’re just reporting that sales happened, not that your marketing caused them.
If sales rose right after my campaign, doesn’t that prove the campaign worked?
Not on its own. That’s correlation — two things moving together. Sales could have risen from seasonality, repeat customers, or word of mouth you’d have earned anyway. To show causation, you compare against a baseline (a holdout, a dark region, a synthetic twin) that estimates what sales would have been with no campaign.
What’s the simplest way to estimate a counterfactual on a small budget?
A randomized holdout. Keep a random slice of your audience — say 10% of an email list or one comparable region — out of the campaign, then compare. Because the split is random, the held-out group is a fair stand-in for “what would have happened,” and it costs almost nothing to set up.
More from this series
Seven plain-English pieces on measuring marketing and turning tests into durable capability:
Want to know whether your marketing is the reason?
One conversation to map what to measure and how we’d prove it.