How to measure marketing ROI: a guide for leadership teams

Attribution, marketing mix modelling and experiments each answer a different question. Here is when to use which.

By Lucienne Okafor, Fellow, Growth & Marketing Analytics · 26 September 2026 · 4 min read

Few questions cause more friction between a chief executive and a chief marketing officer than a simple one: what are we getting for our marketing spend? The finance team wants a number. Marketing offers several, each from a different tool, and they rarely agree.

The disagreement is understandable. Measuring marketing return is genuinely hard. Customers see many messages before they buy, the effects of brand building play out over years, and much of what drives sales, such as price, distribution and the economy, has nothing to do with marketing at all. But hard is not impossible, and the leadership teams that get this right make materially better decisions about where to spend.

The question behind the question

Marketing ROI is usually defined as the incremental profit generated by marketing, divided by what the marketing cost. The key word is incremental. The question is not how many sales followed a campaign, but how many sales would not have happened without it.

That distinction matters because much of what marketing appears to generate would have happened anyway. Customers who search for your brand name and click an advert were probably going to buy. Loyal customers who receive a discount may have bought at full price. A measure that ignores this will reward the channels that are best at being present at the moment of purchase, not the ones that create demand.

Three methods, three different jobs

Multi-touch attribution follows individual customers across digital channels and assigns credit to each interaction. It is useful for day-to-day optimisation of digital campaigns. But it cannot see television, outdoor or word of mouth, it is increasingly limited by privacy changes, and it measures correlation rather than cause.

Marketing mix modelling takes the opposite approach. It uses statistical models on several years of aggregate data to estimate how much each channel, along with price, seasonality and other factors, contributed to sales. It is well suited to annual budget decisions and to measuring brand activity, and it has regained popularity as tracking has become harder. Its weakness is speed and granularity.

Incrementality experiments are the most direct test. Turn a channel off in some regions and not others, or hold back a random group of customers from a campaign, and compare the results. Experiments are the closest thing marketing has to proof, and they are the best way to check whether the other two methods are telling the truth.

Returns diminish, so averages mislead

One of the most useful outputs of good measurement is the shape of the response curve. The first pound spent in a channel usually works harder than the last. A channel with a strong average return may already be saturated, while a channel with a modest average may have plenty of room to grow.

Budget decisions should therefore be made on marginal returns: what would the next unit of spend deliver, in each channel? Reallocating even a modest share of budget from saturated to unsaturated channels often lifts overall returns without any increase in spend.

The question is never whether marketing works. It is which marketing, at what level of spend, and for how long.

Count the long-term effects

Short-term measures tend to favour promotions and performance channels, because their effects are immediate and easy to see. Brand investment works more slowly, supporting pricing power, lowering acquisition costs and sustaining demand over years. A measurement system that only sees the next quarter will steadily shift budget away from the activities that build the business. Good models explicitly estimate these longer-term effects, even if the estimates are uncertain.

What leadership teams should ask for

Ask marketing and finance to agree on one definition of return and one set of numbers. Ask which figures come from experiments and which from models. Ask for the response curve of each major channel, not just its average return. And ask what the business will test next, because a steady programme of experiments is what turns marketing measurement from an argument into a discipline.

RavenArc tests every decision against six questions. See the RavenArc Decision Method.