UX metrics that matter
Many organisations have dashboards full of figures on visits, clicks, conversions, time spent or…
A campaign can end with many conversions, a positive ROAS, an apparently efficient cost per acquisition and engagement data that invites optimism. On paper, everything seems to be working. The report shows growth, the charts are rising and the investment appears justified. Yet there is an uncomfortable question that is becoming increasingly difficult to avoid: how much of that result was actually created by the campaign?
In digital marketing, we have become accustomed to measuring almost everything. We know how many people have seen an advert, how many have clicked, how many have reached a page, how many have added a product to their basket and how many have bought. But measuring many things does not always mean better understanding the real impact. Sometimes, a campaign does not generate new demand, but simply appears along the path of users who were already inclined to buy.
That is where the problem begins. If a brand attributes to a campaign sales that would have happened anyway, it may end up making the wrong decisions: increasing budget where there is no genuine growth, reducing investment in channels that do build demand, or celebrating results that merely reflect the efficient capture of pre-existing intent. What looks like performance may, in fact, be a mirage neatly presented in a dashboard.
That is why the conversation around measurement is changing. It is no longer enough to ask how many sales have been attributed to a campaign. The relevant question is different: has the campaign generated additional sales, or has it simply captured demand that already existed? That distinction is the foundation of incrementality, and it marks the shift from marketing that claims results to marketing that seeks to demonstrate real impact.
What incrementality is and why it matters
Incrementality measures the additional impact a campaign generates compared with what would have happened had that campaign not been activated. Put simply, it is not only about knowing how many sales took place during a marketing action, but about identifying how many of those sales exist because of that action.
This distinction matters because in marketing, not every visible result is necessarily a result that has been created. A brand may launch a campaign, see an increase in sales and quickly conclude that the campaign has worked. But that increase may be influenced by many other factors: loyal customers who were already going to buy, a promotion running during the same period, a seasonal rise in demand, the brand awareness built up over time, or even the effect of other channels that do not appear in the main report.
Consider, for example, an online shop that launches a remarketing campaign aimed at people who had already visited a product several times. If many of them end up buying, the dashboard may attribute those conversions to the campaign. But perhaps a significant proportion of those users already had high purchase intent before seeing the advert. In that case, the campaign may have helped, reminded them or accelerated their decision, but it has not necessarily created all the value attributed to it.
The same happens during periods of high demand, such as sales, Black Friday or Christmas. A campaign may show very good results because it coincides with a time when consumers are already more inclined to buy. If that performance is not compared with what would have happened without the campaign, it is easy to mistake a favourable context for incremental impact.
That is why the key idea is simple but demanding: not everything attributed is incremental. Attribution can tell us which advert, channel or platform is associated with a conversion. Incrementality attempts to answer a deeper question: what proportion of that result would not have happened without the campaign? And that difference completely changes how we assess performance, defend investment and make growth decisions.
The mirage of traditional metrics
For years, many digital marketing decisions have relied on metrics that seem to offer a clear reading of performance: ROAS, CPA, CTR, attributed conversions, impressions, clicks, engagement or total sales during the campaign period. They are useful data, easy to compare and highly visible in any report. The problem arises when they are interpreted as definitive proof of value.
A positive ROAS may indicate that a campaign has generated revenue greater than the investment, but it does not necessarily prove that this revenue is new. A low CPA may seem efficient, but perhaps it has been achieved by reaching users who were already close to buying. A high CTR may show that the message has sparked interest, but it does not always mean that this interest translates into real growth. Even an increase in sales during a campaign may have more to do with seasonality, promotion or existing demand than with the advertising action itself.
This does not mean that these metrics are of no use. On the contrary, they remain necessary for managing campaigns, spotting problems, comparing creative assets, optimising budgets and understanding behaviour within each channel. They help answer important operational questions: which advert gets more clicks, which audience converts better, which format costs less or which platform is generating more activity.
But when the conversation moves from operational performance to business impact, these metrics need context. On their own, they do not answer the more strategic question: what would have happened if we had not invested? Without that comparison, marketing risks confusing activity with effectiveness, apparent efficiency with real growth, and attribution with causality.
The mirage emerges precisely there: in dashboards that show positive results but do not explain whether those results are a direct consequence of the investment, or whether the campaign simply featured in a journey that was already going to end in a purchase. That is why modern measurement cannot be limited to accumulating indicators. It needs to combine platform metrics with models, experiments and more demanding questions about the true impact of each decision.
Attribution is not proof of impact
One of the major challenges of current measurement is distinguishing between assigning credit and demonstrating impact. Attribution attempts to distribute credit for a conversion among channels, campaigns or touchpoints. It may indicate which advert received the last click, which platform took part in the journey or which combination of exposures appears before a purchase. It is a useful tool for organising information, but it does not always tell us whether the campaign was the real cause of the outcome.
Incrementality, by contrast, begins with a different question. It is not satisfied with knowing which channel was present before conversion, but instead tries to measure what would have happened if that channel had not intervened. In other words, it seeks to get closer to causality. It aims to identify whether the campaign generated an additional effect or simply took credit for a sale that would probably have happened anyway.
This difference is especially important in environments where advertising platforms have a partial view of the user journey and, at the same time, incentives to demonstrate that their investment works. A platform may attribute sales to a campaign because it detects that an exposed user ultimately bought. But that reading does not, on its own, prove that the exposure created the sale. Perhaps the user already knew the brand, had compared prices, was waiting for a promotion or had very high purchase intent before receiving the advertising exposure.
That is why an attributed conversion should not automatically be interpreted as an incremental conversion. Attribution looks at the path a conversion followed; incrementality asks whether that conversion would have existed without the campaign. This distinction changes how performance is analysed, because it shifts the conversation from “which channel appears in the report” to “which investment has generated a genuine change in market behaviour”.
When brands do not make this distinction, they may end up overvaluing channels that capture existing demand and undervaluing actions that build demand in the medium term. They may also optimise their campaigns towards the users who are easiest to convert, even if those users do not represent genuine growth. In that scenario, marketing appears more efficient, but it may be reducing its ability to generate new value.
Measuring impact therefore requires going beyond the logic of credit allocation. Attribution helps us understand journeys; incrementality helps us question how much additional value has been created. Both perspectives can coexist, but they should not be confused. A campaign can be very well attributed and still contribute less growth than it appears to.
How incrementality is measured in practice
Measuring incrementality does not mean turning every campaign into a complex statistical project, but it does require changing the way measurement is approached. The question is no longer only how many conversions have been recorded, but what difference exists between a scenario with a campaign and a comparable scenario without one. That comparison is the basis for getting closer to real impact.
One of the most common ways of doing this is by working with test and control groups. The test group receives the advertising exposure and the control group does not, or receives it under different conditions. If both groups are comparable, the difference in behaviour between them can help estimate what proportion of the outcome is genuinely due to the campaign. It is not simply about observing who buys, but about comparing how the likelihood of purchase changes when advertising exposure exists.
Experiments with exposed and unexposed audiences can also be used. For example, a brand may reserve part of its audience so as not to target it during a particular period, and compare its performance with that of those who have seen the campaign. This type of test helps detect whether advertising is creating a significant difference or simply reaching people who already intended to buy.
Another option is geo-testing, which is especially useful when a company operates in different markets, cities or regions. In this case, the campaign is activated in some areas and restricted or excluded in others. The results are then compared, taking into account factors such as sales history, seasonality, previous investment or behavioural differences between territories. It is not a perfect method, but it can be very useful when it is not easy to separate audiences at an individual level.
Holdout groups work according to a similar logic: part of the target audience is kept out of the campaign in order to measure what happens without exposure. Although this can sometimes seem an uncomfortable decision, because it means not targeting potentially valuable users, that small sacrifice can provide a much clearer view of the investment’s real effect.
In more advanced settings, counterfactual models can also be used. These attempt to estimate what would have happened without the campaign on the basis of historical data, behavioural patterns and external variables. Complementarily, marketing mix modelling can help us understand the effect of different channels on aggregate sales, especially when it is calibrated with real experiments and not used as a black box disconnected from practice.
The important thing is to understand that not every company needs to start with the most sophisticated method. An organisation can take its first steps with simple tests, well-designed comparisons and better starting hypotheses. The essential thing is to stop measuring only what the platform attributes and begin to build more rigorous questions: what do we want to demonstrate, which group will we compare it with, how long will we measure it for and what decision will we make according to the outcome?
Incrementality does not eliminate all uncertainty, but it forces more disciplined measurement. And that change is already significant progress: moving from accepting results as they appear in the dashboard to asking what proportion of those results represents genuine growth.
Towards a more rigorous testing culture
Introducing incrementality into marketing does not mean that every decision must become a complex, slow or inaccessible experiment. The key is not to fill the organisation with statistical models, but to build a more rigorous discipline of learning. That is, moving from measuring campaigns solely to justify results to measuring them also to understand what works, under what conditions and why.
The first step is to define hypotheses before activating a campaign. It is not enough to say that the aim is to sell more, generate more leads or improve ROAS. It is worth specifying what is expected to happen: acquire new customers, increase purchase frequency, win back inactive users, drive a particular category or improve conversion within a specific segment. The clearer the hypothesis, the easier it will be to decide later whether the campaign has added real value or has merely produced measurable activity.
It is also important to agree from the outset what is to be demonstrated. Measurement problems often arise because the objective is redefined once the results have been seen. If a campaign does not improve sales, clicks are highlighted. If it does not improve ROAS, the conversation turns to awareness. If it does not generate new customers, engagement is emphasised. All these metrics can have value, but they should not be used to change the criterion for success after the fact. A testing culture requires honesty before, during and after the campaign.
To do this, it is useful to separate diagnostic metrics from business metrics. The former help us understand whether the campaign has been executed correctly: reach, frequency, CTR, views, cost per click or interaction. The latter make it possible to assess whether the action has contributed to a more meaningful objective: incremental sales, new customers, margin, customer lifetime value, market share or category growth. Confusing the two levels can lead to celebrating campaigns that are very active but have little transformative effect.
Another essential practice is documenting learning. An isolated test can provide information, but its real value emerges when the organisation accumulates knowledge. Which audiences responded better, which messages generated more impact, which channels contributed incremental growth, which promotions distorted the results or which periods were not suitable for comparison. Without documentation, every campaign starts almost from scratch. With documentation, measurement becomes strategic memory.
Comparing results between campaigns also helps avoid hasty interpretations. A single experiment may be shaped by context, seasonality, promotional pressure or competitor behaviour. But when patterns are observed over time, decisions begin to rest on more consistent evidence. A testing culture does not seek an absolute truth in every test, but a progressive improvement in the quality of decisions.
That is why the “test and learn” approach is becoming increasingly relevant. It is not only about checking whether a campaign has worked, but about learning which conditions make it work better. This learning can coexist with attribution models, marketing mix modelling and incrementality testing, provided each tool is used to answer the appropriate question. Attribution helps us understand journeys, MMM provides an aggregate view of channel effects and experiments make it possible to get closer to causality.
The ultimate aim is not to measure for measurement’s sake, nor to use tests to confirm decisions that have already been made. A rigorous testing culture helps us make better decisions: where to invest, what to stop doing, what to scale, what to review and what learning to incorporate into the next campaign. In that sense, incrementality is not merely a measurement technique. It is a way of demanding a more honest relationship between marketing and its own results.
Conclusion
Incrementality does not eliminate uncertainty from marketing. No method does so completely. There will always be external variables, behaviours that are difficult to isolate, long-term brand effects and purchasing decisions that cannot be explained solely through a dashboard. But it does introduce a more honest way of looking at results: not only asking what a campaign can claim, but what it has actually created.
This shift matters because it compels us to go beyond the appearance of performance. A campaign may accumulate clicks, attributed conversions, positive ROAS and favourable charts without necessarily generating new growth. It may also have a relevant effect that does not appear clearly in an immediate platform reading. That is why measuring better is not simply about adding more indicators, but about asking better questions.
The central question is not whether a campaign appears associated with a sale, but whether that sale would have happened without the campaign. From there, marketing stops relying solely on surface-level metrics and begins to build a more serious relationship with causality, learning and evidence-based decision-making.
In that sense, incrementality should not be understood as an isolated metric or a methodological trend. It is a sign of maturity. It helps distinguish between capturing existing demand and generating additional value, between justifying investment and learning from it, and between optimising campaigns and building real growth.
The next step is to accept that no measurement tool resolves this challenge on its own. Attribution, marketing mix modelling and experiments all have strengths and limitations. The key lies in knowing how to combine them within an imperfect ecosystem, where measuring well does not mean having a single answer, but building a more complete view of impact. That is where the next conversation begins: how to live with different models without losing sight of the essential business question.
How many of your investment decisions would remain the same if you had to demonstrate incremental impact?
Sources:
Comments