Why incremental ROAS changes the budget decision
Almost every ROAS figure in circulation is missing the control group that would show whether the spend caused anything.
In an illustrative comparison, the same campaign can report 5.0x attributed ROAS versus 1.5x incremental ROAS—a 3.5x gap created by crediting revenue likely to arrive anyway. A controlled test can expose that gap, but it costs coverage, time and organisational discipline.
Attributed ROAS divides revenue credited to advertising by advertising spend. Incremental ROAS divides only the additional revenue caused by advertising by that spend:
Incremental revenue = observed revenue − estimated revenue without advertising
Incremental ROAS = incremental revenue ÷ advertising spend
The distinction changes the decision. If finance requires 2.0x, reported performance of 5.0x says scale. Incremental performance of 1.5x says the campaign has failed the agreed hurdle.
Better attribution hygiene can remove obvious reporting errors. Our guide to finding the real number behind a Google Ads ROAS report covers that clean-up. Incrementality addresses the harder question left behind: how much revenue disappears when the advertising disappears?
The pattern we hit most often is a campaign collecting credit from existing demand. Prospects already know the company, return through a paid result and convert. The ad is observable, so the platform claims the sale. Observation does not establish causation.
Our thesis is falsifiable: a representative audit finding valid control groups behind 50 of 100 published B2B ROAS figures would prove “almost every” wrong.
Control groups turn reported return into a budget decision.
A practical UK geo holdout for incremental ROAS
National user-level randomisation is unavailable to many B2B teams, but regional suppression is usually possible. The test is blunt, which is precisely why founders can understand and challenge it.
Use the following four-part design.
- Apply a go-or-no-go threshold. Start only if the four selected regions historically receive at least £15,000 of monthly campaign spend and each proposed arm produces at least 20 CRM-qualified enquiries during one median sales cycle. Below either threshold, do not run a geo test. Pool more time or use a lead-level randomised holdout.
- Create two matched regional pairs. Keep the campaign active in North West and West Midlands. Hold out Yorkshire and the Humber and East Midlands. Match each pair using spend, qualified enquiries and pipeline movement across the preceding two median sales cycles. The regions are a starting configuration, not a claim that their economies are identical. Keep all other UK regions unchanged and exclude them from the analysis.
- Tie duration to the buying process. The minimum treatment period is 1.5 times the company’s median sales cycle, followed by an observation window of one additional median cycle for late revenue. For an illustrative B2B software firm with a 42-day median cycle:
Minimum treatment period = 42 days × 1.5 = 63 days
A proposed 30-day test versus a 42-day sales cycle must therefore be extended. A fixed eight-week rule is less defensible because it ignores how buyers actually progress.
- Pre-register what counts as readable. Require at least 20 CRM-qualified enquiries in each arm, a 90% interval for incremental revenue that excludes £0, and regional sales activity remaining within 10% of its pre-test level. An illustrative estimate of 1.8x with a 1.3x–2.3x interval is readable; the same 1.8x estimate with a −0.4x–4.0x interval is not.
Target users by physical presence, not interest in the region. Suppress only the campaign being tested. Freeze bidding objectives, offers, landing pages and regional sales incentives during treatment.
CRM postcode, qualified stage and eventual revenue must survive the journey from enquiry to close. Reliable B2B conversion tracking is therefore a prerequisite, not an optional reporting improvement.
Where regional controls and CRM joins need one accountable owner, our B2B Google Ads management work covers the operating layer as well as media buying.
A modest holdout can expose an expensive illusion.
How to calculate incremental ROAS and act on it
Take an illustrative UK consultancy running the regional test above. Use these five labelled inputs:
- Test-region advertising spend: £18,000
- Revenue attributed to the campaign: £90,000
- Observed revenue in exposed regions: £96,000
- Holdout-adjusted counterfactual revenue: £69,000
- Finance-approved minimum incremental ROAS: 2.0x
The counterfactual applies the holdout regions’ pre-test-to-test-period movement to the exposed regions’ baseline. It estimates the revenue expected without the campaign.
Reported ROAS = £90,000 attributed revenue ÷ £18,000 spend = 5.0x
Incremental revenue = £96,000 observed revenue − £69,000 counterfactual revenue = £27,000
Incremental ROAS = £27,000 incremental revenue ÷ £18,000 spend = 1.5x
Measured gap = 5.0x reported ROAS − 1.5x incremental ROAS = 3.5x
The campaign reports 5.0x versus an incremental 1.5x on the same spend. Because 1.5x is below the pre-agreed 2.0x hurdle, the test does not support scaling.
Use three budget rules after the result:
- Lower interval bound clears the finance hurdle: increase budget by 10–20%, preserve the holdout and measure the marginal lift again.
- Point estimate clears the hurdle but the lower bound does not: hold budget until more outcomes mature.
- Upper interval bound remains below the hurdle: reduce the tested spend by at least 20% or redirect it to a new experiment.
Set the hurdle before seeing the result. Decisions involving repeat purchases belong in a separate customer lifetime value return calculation, while differing service margins require a profit-based return rule. Neither adjustment replaces the control group.
A finite experiment is also a sensible use of scoped Google Ads campaign support when an internal team can operate the account but lacks test design capacity.
Scale only when the causal return clears the agreed hurdle.
If the path from ad click to signed revenue still has gaps, Actualyse will trace each hand-off with you — book a call
What does not prove incrementality
Dashboards reward measurement activity even when the underlying evidence remains unchanged.
Four common substitutes fail the causation test.
- Adding more tracking tags. Tags can recover missing events and improve diagnosis. They still observe only exposed users, so they cannot estimate what those users would have done without advertising.
- Comparing the month before with the month after. The comparison mixes advertising with changing demand, sales capacity, pricing, holidays and competitor activity. Time moved alongside the campaign.
- Running an ad-copy A/B test. Variant A versus variant B identifies the stronger advertisement. If both groups receive advertising, neither estimates the no-ad outcome.
- Splitting brand and non-brand reporting. The split describes two forms of captured demand. It does not reveal whether either campaign created additional revenue, especially when brand demand was already moving towards the company.
Platform-reported conversion value also does not become causal because the CRM confirms that a transaction occurred. Matching the transaction proves identity, not incrementality.
Measurement without a counterfactual cannot establish causation.
What an incrementality test cannot rule out
Geography is not random, and regional buyers do not remain neatly inside experimental borders.
The honest limit here is that the result can still be confounded by seasonality, a concurrent sales push and brand demand already in motion. Three confounds need written controls.
- Seasonality: national seasonality affects both arms, but regional events may not. If matched regions’ normalised pre-test trends differ by more than 20%, rematch them or abandon the design.
- A concurrent sales push: outbound calls, partner referrals or account-based activity can inflate one arm. If sales-contact volume shifts by more than 10% between exposed and held-out regions, invalidate the test and restart.
- Brand demand already in motion: switching paid search off may move existing buyers into organic or direct routes. The test can estimate whether the paid campaign added total revenue, but it cannot identify which earlier activity originally created that demand.
Travel, remote working, VPNs and multi-office buying committees also create leakage. Assign outcomes using the customer’s CRM address rather than the click location, and record that rule before launch.
What this cannot tell you is the value of every interaction in the buying journey. The estimate applies to the tested campaign, regions, period and operating conditions. Material changes to the offer or sales process create a new question.
Clean execution narrows uncertainty but never abolishes confounding.
FAQ
Four implementation decisions commonly remain after the core test is designed.
Can a smaller advertiser test incrementality?
Use a lead-level randomised holdout when at least 50 eligible leads can enter both the exposed and control groups. Below 50 per arm, treat the result as directional and do not publish an incremental ROAS figure.
Should branded search be included?
Test branded search separately. If organic and direct routes absorb at least 80% of lost paid visits while qualified opportunities remain within 10% of baseline, cap branded bidding to defensive or strategically important terms.
How often should the test be repeated?
Repeat after a budget change of 25% or more, a material offer or landing-page change, or six months—whichever arrives first. Incrementality is a property of current conditions, not a permanent campaign score.
Should the primary outcome be pipeline or closed revenue?
Pre-register one primary outcome. Use closed revenue when at least 30 wins are expected within the measurement window; otherwise choose sales-accepted pipeline with one frozen stage definition and report revenue later as a secondary outcome.
Incrementality becomes useful when edge cases receive written rules.
Summary
Apply five operating rules:
- Reject causal ROAS claims that lack an exposed-versus-control comparison.
- Start a geo holdout only above £15,000 monthly regional spend and 20 qualified enquiries per arm.
- Run treatment for at least 1.5 times the median sales cycle, then allow one cycle for revenue maturation.
- Scale by 10–20% only when the lower interval bound clears finance’s pre-agreed hurdle.
- Restart the test when regional sales activity shifts by more than 10%.
Controlled evidence should govern the next pound of advertising.
Actualyse builds measurement and attribution setups that tie B2B ad spend to real revenue. Book a call to talk through where yours stands.
