Incremental ROAS vs Attributed ROAS: How to Find Wasted Marketing Spend

ROAS is one of the most widely used metrics in marketing because it appears to give businesses exactly what they want: a direct relationship between money spent and revenue generated. A campaign spends £100,000, reports £500,000 of revenue and therefore produces a ROAS of 5. The number is simple, comparable and easy to use in a budget conversation.

The problem is that the revenue in that calculation is usually attributed revenue. It is revenue that a measurement system has assigned to the campaign according to a set of attribution rules. That is not necessarily the same as the revenue the campaign caused.

This distinction becomes particularly important as performance marketing budgets increase. A channel can continue reporting an attractive attributed ROAS even after much of the incremental opportunity has been exhausted. The campaign is still present before conversions, so it continues receiving credit, but the final portion of spend may increasingly be reaching customers who would have purchased anyway.

That is why a high attributed ROAS should not automatically be interpreted as evidence that a channel deserves more budget. In some cases, an exceptionally high attributed return is a reason to investigate the channel more closely.

Incremental ROAS asks a different question. Rather than determining how much revenue should be credited to marketing, it attempts to estimate the additional revenue created because the marketing intervention occurred. For capital allocation, that is the distinction that ultimately matters.

Attributed ROAS and incremental ROAS measure different things

Attribution systems are designed to distribute conversion credit across observable marketing interactions. The exact approach varies. A company may use last-click attribution, data-driven attribution, a platform-specific model or another method for allocating revenue between touchpoints.

These systems can be operationally useful. They help marketers understand customer paths, inspect campaign activity and connect observable interactions with subsequent conversions.

The difficulty comes when attribution is treated as evidence of causality.

Consider a customer who has already decided to buy from a brand. They search the company name on Google, click a paid brand advertisement and purchase. The advertising platform can correctly record that the paid-search click occurred immediately before the conversion and assign revenue to it.

What attribution cannot tell us from that observation alone is whether removing the advert would have removed the sale. The customer may simply have clicked the organic result instead.

Incremental measurement is interested in that counterfactual, but even that description needs some care. The objective is not merely to classify revenue as either incremental or non-incremental. A useful incrementality analysis needs to understand the size of the effect, the conditions under which it occurred, the treatment that was actually delivered, the uncertainty around the estimate and whether the result can support the wider budget decision.

If £100,000 of advertising receives credit for £500,000 of revenue but only £150,000 of additional revenue was created because of the intervention, the attributed ROAS is 5 while the incremental ROAS is 1.5. Both numbers can be calculated correctly while describing very different economic realities.

The problem therefore is not that attribution is inherently wrong. It is that attribution and incrementality answer different questions, and attributed revenue is often used to make decisions that require causal evidence.

Why demand capture tends to look exceptionally efficient

The gap between attributed and incremental performance is particularly important for marketing activity close to the purchase.

Brand search is an obvious example. A customer who searches directly for a company's name is already demonstrating unusually high purchase intent. That intent may have been created by years of brand building, a recommendation, previous experience, offline advertising, a store visit or another influence the search platform cannot observe.

Because paid search sits immediately before the purchase, it is easy to measure and easy to credit. The factors that created the customer's preference are usually much harder to observe.

Retargeting has a similar structural advantage. Retargeting audiences are created specifically because users have already demonstrated interest. They have visited a website, viewed a product, started a basket or performed another action associated with a higher probability of purchase.

If those customers subsequently convert, the campaign can report excellent attributed performance. The fact that the audience was already more likely to purchase is not necessarily reflected in the headline ROAS.

This does not mean brand search or retargeting has no incremental value. Both can create genuine additional revenue. Competitors may bid on brand terms, paid search can affect visibility, and retargeting can bring customers back who otherwise would not have returned.

The important point is that a high attributed ROAS is not sufficient evidence of that incremental effect. The closer marketing gets to existing demand, the more important it becomes to distinguish between creating a sale and being present when the sale happens.

Why attributed ROAS can remain healthy after marginal returns deteriorate

The difference becomes larger as spend scales.

Performance channels generally do not produce the same incremental return at every level of investment. Early spend may reach highly responsive customers who genuinely require the marketing intervention to convert. As more money enters the channel, those opportunities are progressively exhausted.

Platforms then have to find somewhere to deploy the additional budget. They may broaden audiences, increase frequency, bid more aggressively for existing demand or move towards customers who are already increasingly likely to convert.

The average performance of the channel can still look attractive because the highly productive early spend and the less productive marginal spend are blended together.

This is the problem with relying on average ROAS for capital allocation.

Imagine that the first £1 million invested in a channel generates a strong incremental return. The next £500,000 still creates additional revenue, but at a lower rate. Another £500,000 produces very little additional growth.

The average return across the full £2 million can remain acceptable even though the final portion of the budget is economically unattractive.

At Growth Dynamics, we refer to this flat part of the response curve as the Death Zone. Spend continues to increase, platforms continue reporting conversions, but the marginal contribution to topline revenue has largely disappeared.

The allocation decision is therefore not primarily about whether the historical channel average was profitable. It is about the expected return on the next unit of capital.

A channel with an average attributed ROAS of 6 can still be a poor place for the next £100,000 if its marginal incremental return has already collapsed.

High attributed ROAS can therefore be a diagnostic signal

Marketing teams naturally spend time investigating channels with low ROAS. That makes sense, but unusually high attributed returns deserve scrutiny as well.

An exceptional ROAS may reflect a genuinely underfunded opportunity. It may also indicate that a channel is heavily concentrated on customers already close to purchase.

The distinction can often be investigated initially without building a complex causal model.

Start by examining how spend and total revenue move together. If investment in a conversion channel has increased substantially while topline revenue has barely changed, that should raise a question regardless of what the attribution dashboard reports.

Then look further up the journey. Is the campaign creating more qualified traffic? Are commercially meaningful behaviours increasing? Is branded demand changing? Is the customer pool expanding, or is the channel simply capturing a larger share of the conversions already present?

This is descriptive evidence, not causal proof, but it is extremely useful for locating where a measurement problem may exist.

A particularly useful pattern is rising attributed revenue alongside relatively flat business outcomes. If a platform claims progressively more conversions while total sales remain broadly unchanged, the most likely explanation is not necessarily that the platform has become dramatically better at creating demand. It may simply have become better at claiming demand generated elsewhere.

That observation becomes a hypothesis worth testing.

Brand search shows the problem clearly

Brand search is one of the easiest places to understand the distinction between attributed and incremental performance because substitution can be relatively visible.

Suppose a business spends heavily bidding on its own brand name. Paid search reports a very high ROAS and appears to be one of the most efficient channels in the marketing mix.

Before increasing that budget, I would want to understand the environment around the branded query.

Are competitors actively bidding on the company's terms? How dominant is the organic result? How much branded organic traffic already exists? What happens to total search conversions when paid coverage changes?

If paid brand investment is reduced and paid clicks fall significantly while organic traffic absorbs much of the movement, the attributed revenue loss will be much larger than the actual business loss.

That is economically important.

The campaign did not necessarily create all of the revenue it previously received credit for. Some of its value was navigational or defensive.

The correct conclusion is not that every company should turn off brand search. The conclusion is that brand search should be funded according to its incremental value, competitive context and substitution behaviour rather than the amount of revenue it can claim under attribution.

The same logic applies to other channels positioned very close to existing demand.

Retargeting creates a similar measurement problem

Retargeting is often evaluated by comparing the conversion behaviour of retargeted users with broader audiences. This can be misleading because the groups are different before advertising even begins.

Someone who visited a product page yesterday already has a higher purchase propensity than somebody who has never engaged with the brand. Showing that the first group converts at a high rate does not establish that the retargeting caused the difference.

A more useful analysis asks what additional outcome the intervention created among otherwise comparable eligible customers or markets.

The answer may still support continued investment. A retargeting campaign with an attributed ROAS of 12 and an incremental ROAS of 3 can still be commercially attractive.

What changes is the basis for the budget decision.

The business should not behave as though every £12 of attributed revenue disappears when £1 of retargeting spend is removed. It should make the decision using the best estimate of the incremental response and the uncertainty around that estimate.

This becomes particularly important when determining how far a campaign should be scaled. Even an incremental channel can eventually move beyond the useful part of its response curve.

How I would investigate whether spend has moved beyond incrementality

I would not start with an experiment simply because a channel has a suspicious ROAS. Experiments are valuable, but they should be used when the decision requires causal evidence rather than as the default answer to every measurement question.

The first step is usually descriptive.

Look at spend against total revenue rather than only attributed revenue. Examine how relevant mid-funnel behaviours move as investment changes. Inspect branded demand, qualified traffic and other signals that help establish whether the underlying pool of demand is expanding.

Then go inside the channel.

A channel-level number can hide major differences between campaigns, audiences, placements and creative. It is possible for the channel to remain valuable overall while a substantial portion of spend is sitting in areas with poor economics.

The objective is to understand whether the apparent problem is a channel problem, an execution problem or a saturation problem.

If the observed evidence suggests a meaningful amount of spend may no longer be creating additional revenue, the causal question becomes commercially important enough to test.

This is where incrementality can provide stronger evidence.

The design needs to match the decision. That could involve a geographic experiment, a controlled reduction in spend or another treatment appropriate to the channel and available data.

The analysis should not stop at whether a statistically significant effect was detected. It should examine the magnitude of the change, treatment delivery, heterogeneity, uncertainty and the extent to which the measured response can reasonably be applied beyond the tested conditions.

In one example from our own work, the spend-response evidence suggested roughly 30% of a conversion budget was sitting beyond the useful part of the curve. That portion of the budget was reduced and topline revenue did not move.

The important lesson is not that 30% is a universal benchmark. It is that attribution alone would not have identified where the incremental return had flattened.

Attributed ROI targets can distort the whole marketing system

The measurement issue becomes more serious when attributed ROI is turned into an organisational target.

Leadership understandably wants marketing to demonstrate financial accountability. An ROAS target appears to provide a straightforward mechanism for doing that.

But people optimise towards the metrics they are given.

If the team is rewarded for increasing attributed ROI, capital naturally moves towards the activities best able to claim conversions. Brand search, retargeting, high-intent audiences and conversion campaigns become increasingly attractive because their relationship with the sale is easier to observe.

Activities that create future demand frequently look less efficient under the same framework because the resulting revenue appears later and may be claimed by another channel closer to the transaction.

The reporting system can therefore improve while the underlying marketing system becomes increasingly dependent on harvesting demand that already exists.

Over time, that can create a structural problem. Less money is invested in growing the pool of customers willing to buy from the company, while more money competes for the existing pool.

Performance marketing then becomes harder because the available demand has not grown sufficiently. Costs rise, marginal returns weaken and the organisation responds by pushing the performance system even harder.

The problem is not irrational marketers. It is rational behaviour inside an incentive system built around a metric that does not answer the capital allocation question.

Attribution remains useful, but it should not decide the budget alone

None of this requires abandoning attribution.

Attribution is useful for understanding observable journeys, diagnosing campaign behaviour, investigating tracking issues and helping marketing teams operate digital channels.

The mistake is asking it to do something it was not designed to do.

Adding more touchpoints to the model does not solve the causal problem. A more detailed record of the observed journey remains a record of the observed journey.

Incrementality provides stronger causal evidence because the intervention is deliberately varied and compared with a counterfactual. But even incrementality does not produce universal truth. The result is evidence about a particular treatment under particular conditions, with uncertainty and limits around how far the effect can be generalised.

Marketing mix modelling answers another question again. It can help decompose larger patterns across the marketing system and estimate response curves, but its conclusions depend on what the historical variation can identify and on the modelling assumptions used.

These methods should therefore operate as complementary lenses rather than competitors.

Observed evidence can reveal where something looks economically wrong. Incrementality can test important causal hypotheses. MMM can help place those findings inside the broader marketing system.

The method should follow the decision.

Incremental ROAS is ultimately about capital allocation

The reason the distinction between attributed and incremental ROAS matters is not methodological purity. It is money.

A business has a finite marketing budget and needs to decide where the next pound or dollar should go.

Sometimes the evidence will show that a high-performing conversion channel still has significant incremental headroom and deserves more investment.

Sometimes the channel will remain valuable but its marginal return will suggest holding spend rather than increasing it.

Sometimes the final portion of the budget will be capturing customers who would largely have purchased anyway, making divestment the better decision.

One caution on reading the aftermath of a cut. When the least productive spend is removed, ROI genuinely improves; the arithmetic is correct and the efficiency gain is real. The risk sits later in time. Part of what broad spend does is refill the pool of future buyers, and a cut that looks costless today can drain that pool quietly, with the revenue consequence arriving quarters later while the ratio still looks excellent. A better ROI on a shrinking base is not a win. After any material cut, watch the demand indicators alongside the ratio: branded search, baseline revenue, new-customer share. If those start sliding while ROI improves, the improvement is being financed by the future.

The important question is not which channel can claim the most revenue.

It is which investment changes the business outcome enough to justify the next unit of capital.

Attributed ROAS can help describe what happened inside the observable marketing journey. It cannot answer that allocation question on its own.

Growth Dynamics works with B2C brands to identify where marketing spend is producing genuine additional growth, where marginal returns have flattened and where budget can be reallocated more productively. We begin with observable business evidence, use designed causal tests where the allocation decision requires them, and place those results inside a broader measurement system.

The objective is not to maximise the ROAS reported by the marketing stack.

It is to know where the next dollar should go, and why.