Incrementality Testing for Accountable Ad Spend

A campaign can report strong conversion numbers and still contribute less new business than the platform suggests. Someone may have found your brand through organic search, been referred by a colleague, or intended to enquire anyway. The advert gets the recorded conversion, but it may not have caused it. Incrementality testing is designed to answer the question that attribution reports cannot fully settle: what additional outcomes did paid media actually create?

For businesses investing in Google Ads, Meta Ads or Microsoft Ads, that distinction matters. It changes how you judge performance, where you place budget and whether an apparently successful campaign is genuinely producing better leads.

What incrementality testing measures

Incrementality is the extra value created because advertising ran. In paid media, this usually means the difference between results from an audience exposed to ads and comparable results from an audience that was not exposed.

The objective is not to prove that an ad appeared before a conversion. Platform attribution already does that, using its own rules around clicks, views and conversion windows. The objective is to estimate the causal effect of advertising on leads, sales, revenue or another commercially meaningful outcome.

For example, a B2B firm might pause a branded search campaign in selected regions while keeping it active elsewhere. If enquiries fall materially in the paused regions, after accounting for normal differences between locations, the campaign is likely generating incremental demand. If enquiries barely move, much of the reported conversion volume may have happened without the ads.

This does not mean branded activity is automatically wasteful. It may protect visibility from competitors, improve the route to conversion or support peak demand periods. It does mean the business should understand what it is paying for, rather than treating every attributed conversion as new value.

Why platform reporting is not enough

Google, Meta and Microsoft report within their own environments. Each platform can legitimately claim credit under its configured attribution model, but none has a complete view of every marketing touchpoint or the customer’s underlying intent.

This creates familiar problems. A prospect may click a Google ad after already seeing a Meta campaign. A returning customer may search directly for your company, click a paid brand ad and submit a form. A lead may be counted by both an ad platform and your CRM, yet later prove unsuitable for the sales team.

None of this makes platform data useless. It remains valuable for campaign management: identifying search terms, audiences, creative, bids and placements that appear more efficient than others. The issue comes when reported conversions become the sole basis for budget decisions.

Incrementality testing adds a more demanding question: if this spend disappeared or changed, would the business see fewer qualified leads or less revenue? That is a more useful measure of accountable growth.

Common approaches to incrementality testing

The right method depends on traffic levels, sales cycles, geography and how easily audiences can be separated. A controlled experiment is preferable, but the level of control must be realistic for the business.

Geographic holdouts

A geographic test runs advertising in one set of matched areas and withholds it from another. It is often practical for businesses with regional coverage, multiple locations or sufficient national demand. The test needs enough conversion volume and locations with broadly similar historic performance.

A London-focused business can use this method carefully, though local market differences can be significant. Demand, competition, demographics and sales capacity vary between boroughs and surrounding counties. Simply switching off ads in one postcode area and comparing it with another can produce a misleading result.

Audience split tests

Where platform tools permit it, a randomly selected control group does not see the campaign while a treatment group does. Randomisation makes this one of the cleaner methods because it reduces the risk that the two groups differ in ways that affect conversion likelihood.

The trade-off is that platform-based experiments are still shaped by the platform’s own measurement environment. They are useful evidence, but businesses should compare results with CRM outcomes, offline conversions and lead quality rather than relying only on reported lift.

Time-based tests

A time-based test compares performance during an advertising period with a period when activity is reduced or paused. This is easier to run but less reliable because demand changes over time. Seasonality, competitor activity, promotions, bank holidays and changes to the website can all affect the result.

It can still be useful for smaller accounts where a formal holdout is not viable. The key is to avoid treating a short pause as proof. Look at a sufficiently long period, use historic benchmarks and document any other changes that could explain the movement.

Set the test around a business outcome

The first decision is not which campaign to test. It is which outcome matters enough to test properly.

For lead generation, form fills alone may be too weak. A more meaningful outcome could be qualified leads accepted by sales, booked consultations, opportunities created or pipeline value. For ecommerce, the focus may be new-customer revenue, contribution margin or repeat purchase behaviour rather than total orders.

This is where clearer tracking becomes essential. If source data does not reliably pass into a CRM, or if sales teams do not consistently record lead status, incrementality analysis will only be as reliable as the underlying data. A campaign may generate a measured lift in enquiries but no lift in sales-qualified leads. That is not a minor reporting issue. It changes the budget decision.

Before running the test, agree the primary metric, the date range, the control method and the decision that will follow each possible outcome. Without this discipline, it becomes too easy to reinterpret results after the fact.

How to run a useful test

Start with a campaign where the business has genuine uncertainty. Brand search, retargeting, broad prospecting and partner-network activity are common candidates because each can receive considerable attribution credit while serving users who may already be close to converting.

Keep other major variables stable. Avoid changing landing pages, offers, conversion definitions, targeting and sales processes during the experiment if possible. If changes are necessary, record them. A test is only useful when you can separate the impact of advertising from other activity.

Allow enough time and volume. A campaign producing a handful of conversions per week is unlikely to support a confident conclusion from a short test. Extending the period, grouping similar campaigns or using a higher-volume metric can help, but there is a limit. Some accounts simply do not have enough data for a formal experiment yet.

Finally, assess the result against commercial measures. If spend falls by £5,000 and reported conversions fall by 80, but qualified opportunities and revenue remain unchanged, the apparent efficiency of the campaign needs scrutiny. Conversely, if a controlled reduction produces a meaningful decline in sales pipeline, the campaign may deserve more investment even when its platform cost per lead looks relatively high.

Reading results without forcing certainty

Incrementality testing is not a promise of perfect attribution. It produces an estimate, usually with a degree of uncertainty. Small effects can be hard to distinguish from normal variation, particularly in low-volume B2B accounts with longer buying cycles.

A sensible interpretation considers direction, scale and confidence. Did the treatment group perform better? Is the difference large enough to matter commercially? Is there a credible alternative explanation? A 3% uplift may be statistically interesting but not justify additional complexity or spend. A 25% uplift in qualified pipeline may be decisive, even if the test is not academically flawless.

It is also worth testing both the campaigns that look weak and the ones that look excellent. High reported return can indicate an efficient campaign, but it can also indicate that the platform is claiming users who were already likely to convert. The greatest opportunity is often not cutting a channel entirely. It is reallocating budget towards activity that reaches genuinely new demand.

Use the findings to improve account decisions

The practical value of incrementality testing is what happens next. A positive result may support increased spend, wider geographic coverage or more investment in the creative and audiences that drove the lift. A weak result may point to reduced bids on brand terms, narrower retargeting windows, exclusions for existing customers or budget moved into prospecting.

It can also reveal problems beyond media buying. If ads drive more traffic but not more qualified leads, review landing-page friction, form quality, follow-up speed and targeting criteria. Paid media should not be judged in isolation from the conversion process it feeds.

For accounts with unclear reporting, a PPC audit is often the sensible first step. It can establish whether conversion tracking, campaign structure and CRM feedback are strong enough to make a meaningful test possible, while identifying what is wasting budget now.

The most valuable outcome is not a neat percentage on a report. It is the confidence to make the next budget decision with clearer evidence of what advertising is genuinely adding to the business.