The platform can record a conversion without creating the customer
A customer sees an ad, later searches the company name and buys. The advertising platform credits the sale. That may be a valid description of the observed path, but it does not answer the owner's harder question: would the customer have bought anyway?
This difference is called incrementality. It means the additional outcome caused by advertising compared with the outcome expected without it. Attribution assigns credit under a reporting rule. Incrementality asks what changed because the advertising existed.
My verdict is practical: use attribution to operate campaigns and incrementality to govern material investment decisions. Do not demand a perfect experiment for every small adjustment. Do not let an attractive attributed return justify a major budget increase when the business cannot separate new demand from captured brand searches, existing customers or sales already in progress.
If the immediate problem is conflicting totals across advertising, analytics and store systems, use the revenue reconciliation guide. If the question is whether to pay for searches on your own company name, use the brand-search decision guide. For the complete multi-channel view, use the Marketing Proof Stack.
Separate three kinds of paid-media value
Advertising does not have to create a customer from nothing to be useful. The business should name the job before judging the result.
Reach or persuade customers who were unlikely to buy in the relevant period without advertising.
Defend a high-intent route from competitors, confusion or unavailable organic coverage.
Shorten a qualified journey or improve customer mix without claiming every conversion as new.
“Create” is the clearest incremental claim. “Protect” and “accelerate” can also justify spend, but they need their own evidence. A defensive campaign should show that valuable demand is genuinely exposed. An acceleration claim should show earlier or better commercial progression—not simply a shorter platform attribution window.
Research illustrates why the distinction matters. A large eBay paid-search field experiment found that non-experimental estimates overstated returns in that setting because search activity and purchasing intent were correlated. The result is not a universal verdict on paid search; it is evidence that observed conversions and caused conversions can differ materially.
The Additional Customer Proof Gate
I use six checks before allowing attributed performance to support a material paid-media decision.
Name the business result
Choose purchases, suitable enquiries, accepted opportunities or another outcome the business can verify.
Estimate the no-ad path
Record normal demand, seasonality, promotions, sales activity, customer mix and conversion delay.
Remove easy credit
Distinguish brand, non-brand, prospecting, remarketing and existing-customer activity.
Create a credible difference
Use a lift study, holdout, matched geography or stable matched period appropriate to scale and risk.
Value what was added
Subtract media, discounts, refunds, fulfilment, sales effort and capacity cost from incremental value.
Decide before seeing the result
Set the confidence, loss boundary and expand, hold, repair or reduce threshold in advance.
Google describes Conversion Lift as a controlled experiment that compares people or regions exposed to ads with a control group that is not exposed. It reports metrics such as incremental conversions, incremental cost per action and incremental return on ad spend. Google also states that the product is not available to every account and uses feasibility checks because volume, budget, duration and split affect the chance of a conclusive result.
That limitation matters. A small business should not imitate an enterprise test with too little data and call noise a finding. When a formal lift study is infeasible, improve the baseline, isolate demand types and make the next budget decision smaller and reversible.
Read the whole customer outcome, not the campaign total
| Matched condition | Platform-attributed customers | Total verified customers | Verified contribution | Advertising cost |
|---|---|---|---|---|
| Advertising active | 48 | 112 | 28,000 units | 6,000 units |
| Credible baseline | 0 | 94 | 23,500 units | 0 units |
The platform attributes 48 customers, while the matched business comparison observes 18 additional verified customers and 4,500 units of additional contribution before advertising cost. On these synthetic inputs, the advertising has not yet recovered its full cost. The comparison still needs checks for seasonality, customer mix, other marketing and sales changes before any causal claim.
The calculation is deliberately uncomfortable. It prevents two opposite errors: treating all 48 attributed customers as new, or concluding that the ads created nothing because the lift estimate is uncertain. The right response may be to improve the campaign, narrow the audience, change the offer, extend the evidence window or reduce exposure.
Match the budget decision to the evidence
| Evidence pattern | Owner verdict | Next action | Avoid |
|---|---|---|---|
| Incremental suitable customers create acceptable contribution | Expand carefully | Release the next budget band and preserve the measurement design | Assuming the same lift continues at every spend level |
| Attributed results are strong; total business outcomes barely move | Credit is overstated | Separate brand, remarketing and existing-customer demand; reduce or retest | Scaling from platform return alone |
| Total outcomes improve, but contribution does not | Volume without value | Repair customer mix, price, margin or fulfilment before spending more | Calling customer count profitable growth |
| Evidence is directionally positive but inconclusive | Hold with limits | Keep a capped budget, improve power or extend the observation window | Turning uncertainty into a confident positive claim |
| The business cannot establish a stable baseline | Measurement not decision-ready | Stabilise tracking and material commercial changes first | Running a test while price, offer, site and sales process all change |
| Advertising protects a documented high-value route | Maintain targeted defence | Cap coverage and recheck when competitor or organic conditions change | Reporting defended demand as wholly new demand |
Use the Next-Dollar Scale Test when the question becomes how much more to invest. Use conversion tracking to connect advertising with verified customer stages, but keep finance and CRM evidence in the final decision. Google Ads and Meta Ads support should be judged by the commercial job and evidence quality—not by access to a larger attribution number.
A four-step owner process for the next material decision
Choose one disputed claim
State the budget decision, customer outcome, market and evidence window.
Separate demand types
Make brand, prospecting, remarketing and existing-customer activity visible.
Use the strongest feasible design
Protect customer experience and lawful data use while creating a credible control.
Decide with business value
Join media cost to verified outcomes, contribution, capacity and uncertainty.
A 2026 NBER working paper on predicted incrementality used 2,226 Meta ad experiments and found that experiment-informed estimates aligned with randomized-test decisions more closely than seven-day last-click attribution in that research setting. It is a working paper, not a universal benchmark, and several authors disclose current or prior Meta relationships. Its useful owner lesson is narrower: repeated experiments can improve decision systems, while a familiar attribution rule should not be treated as causal truth.
Review approved case evidence, ThomPerformance evidence standards and Thomas Ho's operating model before choosing support. The conversion goal here is a scoped advertising evidence diagnostic—not a promise that every account can produce a precise lift estimate.
Sources and evidence notes
Sources were checked on 7 November 2026. The Additional Customer Proof Gate, three-value model, owner decision matrix and four-step process are original ThomPerformance analysis. The worked numbers are explicitly synthetic. No client result, universal lift threshold or guaranteed testing outcome is claimed.
- Google Ads Help: About Conversion Lift
- Google Ads Help: Conversion Lift based on geography
- Google Ads Help: user-based Conversion Lift eligibility and feasibility
- Meta Blueprint: A/B tests, creative tests and lift studies
- Blake, Nosko and Tadelis: paid-search field experiment
- Gordon, Moakler and Zettelmeyer: Predicted Incrementality by Experimentation
Frequently asked questions
What does advertising incrementality mean?
Advertising incrementality is the additional customer action caused by advertising compared with what would have happened without it. The action may be a purchase, qualified enquiry, accepted opportunity or another defined business outcome. Platform-attributed conversions are not automatically incremental because some customers would have arrived through organic search, direct visits, sales activity or existing demand.
Does a high return on ad spend prove paid ads are working?
No. Return on ad spend can show the value a platform attributes to advertising under its rules. It does not automatically show profit, customer quality or causation. Reconcile the reported value with business records, then use a credible comparison or experiment where the decision is material enough to require causal evidence.
How can a small business measure incremental advertising value?
Start with a stable business baseline, separate brand from non-brand demand, compare qualified outcomes across matched periods and change one material variable at a time. A formal lift study may be infeasible at low volume. In that case, use the strongest available comparison, expose its limits and make smaller, reversible budget decisions.
Should a business stop advertising if incrementality cannot be proven?
Not automatically. Advertising may protect demand, accelerate a sale, improve customer mix or support a launch even when a precise lift estimate is unavailable. Treat uncertainty as a reason to cap exposure, improve measurement and define a review rule—not as proof that advertising has zero value or deserves unlimited budget.
When is an advertising lift test worth running?
Run one when the budget decision is material, the outcome and control condition are clear, sufficient volume exists, customer data can be handled lawfully and the business can tolerate withholding some exposure. Do not run an underpowered test merely to create a precise-looking number; Google explicitly checks feasibility because inconclusive studies are possible.
Pay for the customer difference, not the attribution story
Attributed conversions help teams operate. Additional customers and contribution help owners allocate capital. Define the advertising job, estimate what would happen without it, create the strongest feasible comparison and make the next decision proportionate to the evidence.
Which part of your current paid-media return would survive if the attribution window disappeared?
