Why a Performance Marketing Agency Should Measure Incrementality, Not Just Attribution

A campaign reports 120 conversions. The tempting conclusion is that advertising created 120 conversions. It may not have. Some customers would have bought after searching for the brand, returning directly to the website or responding to another channel, even if the campaign had never run. Attribution records which touchpoint received credit; incrementality asks how many outcomes happened because advertising changed behaviour.

That distinction matters when budgets are tight. A campaign can show an attractive return on ad spend while collecting credit for demand that already existed. Conversely, an upper-funnel campaign may appear inefficient because its contribution is dispersed across later searches, visits and sales conversations. A performance marketing agency should therefore treat platform attribution as an operating signal, not a complete answer to the commercial question: what did this spend add?

Why Attributed ROAS Can Mislead a Performance Marketing Agency

Attribution and incrementality answer different questions. Attribution distributes credit among observed touchpoints according to a chosen model and lookback window. Incrementality compares an exposed group with a credible version of what would have happened without the advertising. Google describes conversion lift as a controlled experiment that separates people who saw advertising from people who did not, then measures the difference in downstream conversions. That difference is the lift created by the campaign, rather than the total number attached to it.

Consider a branded search campaign. People who type a company name into Google often already know the business and may be close to making a purchase. The campaign can still protect visibility, clarify the offer or prevent a competitor from taking the click, but attributed sales do not prove that every buyer needed the advert. If the business pauses branded advertising for a carefully selected control group and sales barely change, the campaign may be capturing existing demand more than creating additional demand.

The reverse problem appears with prospecting. A video or social advert may introduce the brand, yet the eventual conversion is credited to search, direct traffic or email. Better tracking reduces missing data, which is why Blue Beetle’s guidance on under-reporting and server-side tracking remains important. But more complete attribution is still attribution. It improves the record of the journey without automatically proving the cause of the outcome.

A Performance Marketing Agency Needs a Counterfactual

The counterfactual is the result that would have occurred without the campaign. It cannot be observed directly for the same customer at the same time, so a useful test creates a comparable control group. If 1,000 exposed customers generate 80 purchases while a comparable unexposed group generates 60, the incremental result is 20 purchases, subject to the test design and statistical uncertainty. The other 60 should not automatically be treated as advertising-created sales.

This is also why a before-and-after comparison is usually weak evidence. Sales may rise after a campaign due to seasonality, pricing, stock availability, sales activity, economic conditions, or a competitor leaving the market. Google’s Meridian guidance stresses that causal measurement must account for variables that influence both media activity and business outcomes. A coincidental movement in the same direction is not the same as causation.

How a Performance Marketing Agency Should Design the Test

Start with one decision, not a general desire to ‘measure better’. The test should resolve a live budget question, such as whether branded search creates additional enquiries, whether paid social prospecting adds new customers, or whether increasing spend in a market produces enough extra revenue to justify the increase. Define the commercial outcome before choosing the platform method. Qualified opportunities, first purchases, gross profit or retained customers are usually more useful than clicks and form fills.

A practical incrementality brief should specify:

• The hypothesis: what behaviour is the campaign expected to change?
• The treatment and control: who receives advertising, and who provides the comparison?
• The primary outcome: which business result will determine the decision?
• The guardrails: what must remain stable, including pricing, promotions, geography and sales activity?
• The decision rule: what level of incremental return would justify maintaining, reducing or increasing spend?

Choose the smallest credible method

User-level lift tests are useful when a platform can randomly assign eligible people to exposed and control groups, and the campaign has sufficient conversion volume. Geo experiments compare matched regions, then change media pressure in selected locations. Google’s forthcoming GeoX framework describes holdback, go-dark and heavy-up designs for publisher-agnostic geographic testing. For larger advertisers with multiple channels and sufficient historical data, marketing mix modelling can estimate channel contribution and diminishing returns. Google positions Meridian as an open-source model that can be calibrated with experiments.

Smaller businesses should not imitate enterprise measurement for appearance’s sake. A clean test in two comparable locations, a controlled pause in one campaign type, or a rotating holdout across matched periods may produce a more defensible answer than a sophisticated model fed with limited data. The method should match conversion volume, sales cycle, geography and the size of the budget decision.

Protect the test from operational noise

Do not launch a new promotion, change prices, redesign the landing page and alter sales follow-up midway through the experiment. Those interventions make the result harder to interpret. Allow enough time for delayed conversions, particularly in B2B or high-value purchases, and analyse the confidence interval rather than treating the central estimate as exact. A result of ‘15% lift, plus or minus 14%’ supports a different decision from ‘15% lift, plus or minus 3%’. Neither should be reduced to a confident headline without its uncertainty.

Turn Incrementality Into Better Budget Decisions

The point of testing is not to produce another dashboard. It is to change how money is allocated. Compare incremental revenue or profit with the full cost of the campaign, then calculate incremental return. If advertising costs AED 100,000 and generates AED 160,000 in additional gross profit, the commercial case is very different from a platform report showing AED 160,000 in attributed revenue. The first figure is causal and profit-based; the second may include sales that would have happened anyway and ignores the economics of fulfilment.

Results should also be interpreted by customer and campaign type. Retargeting often reports excellent attributed performance because it reaches people already close to conversion. That does not make retargeting unhelpful, but it may mean its incremental return is lower than its dashboard return. Prospecting can show the opposite pattern. A strong performance marketing agency uses the gap between attributed and incremental performance to understand the role each campaign plays, rather than forcing every channel into the same last-click logic.

One test should lead to the next decision. If paid social proves incremental at the current spend, test whether a controlled increase continues to add customers or reaches diminishing returns. If branded search shows limited lift, reduce coverage selectively rather than switching it off everywhere. If results vary sharply by region, investigate differences in brand awareness, distribution, or sales capacity before applying a single national budget rule. Measurement becomes valuable when it creates a repeatable learning cycle: test, decide, implement and retest after conditions change.

Data infrastructure still matters. Meta explains that its Conversions API can connect website, app, offline and CRM events to its systems and support lift studies for offline purchases. Clean event definitions, deduplication and CRM outcomes help the test observe what the business actually values. They do not replace experimental design, but poor data can weaken even a well-designed experiment.

Measure What Advertising Added

Attribution is useful for daily optimisation, diagnosing journeys and identifying where conversions are recorded. It becomes dangerous only when credited outcomes are treated as proof of caused outcomes. Incrementality adds the missing comparison and gives leadership a more defensible basis for deciding where to increase, defend or withdraw spending.

Blue Beetle works with businesses that want marketing decisions connected to commercial results, not platform metrics in isolation. If your reporting shows activity but cannot explain what advertising genuinely added, work with a performance marketing agency to build a measurement plan that combines reliable tracking, practical experiments and clear budget decisions.