By nDir 
In a hypothetical scenario, a B2B software vendor promotes demos of an inventory management platform to manufacturers researching this type of software. Its marketing manager wants to know whether the campaign generates additional qualified demo requests or mainly reaches accounts that would have contacted the company anyway.
That is the question incrementality testing helps answer before the team increases its budget. A useful result combines the estimated change in demo requests with a measure of uncertainty and a clear view of what would make the campaign worth expanding.
Atlantic Tech, a data intelligence company founded by Peter Kazan, describes a model in which results feed back into the system that interpreted the signals and defined the audience. For a marketing team, a randomized comparison adds evidence about how much a campaign changed the qualified demo request rate.
What Attribution Actually Counts
For marketing incrementality, attribution and experimentation serve different purposes. Attribution assigns credit to earlier interactions according to the attribution model and conversion window. A randomized comparison estimates the additional effect of the campaign on the demo request rate.
Google Ads documentation distinguishes attributed conversions from those measured through Conversion Lift. The vendor can use attribution to understand the interactions preceding a request, then compare the randomized groups to estimate additional qualified demo requests. The experimental comparison counts all qualifying requests in each group, including those arriving through other channels.
The Control Has to Answer the Right Question
The vendor first defines which accounts qualify for its audience targeting and engagement campaign. It then randomly assigns 20,000 eligible accounts to two groups of 10,000. One group is eligible for the demo campaign; the control group, or holdout, is excluded from it.
Google’s Experiments Playbook describes this split before exposure. Random assignment makes the groups comparable on average, reducing selection bias at the design stage. The vendor analyzes accounts in their originally assigned groups rather than regrouping them according to whether they later saw an ad.
All contacts from the same company stay in the same group. This reduces the risk that colleagues share campaign messages across treatment and control. Other marketing activities follow the same rules in both groups, so the test estimates what the demo campaign adds on top of those activities.
Before launch, the team records five decisions:
- Which accounts qualify and which campaign the holdout will miss.
- What makes a demo request qualified, with the same criteria applied to both groups.
- How to count each account once if it submits at least one qualifying request.
- How long to observe both groups, allowing time for requests to arrive.
- The smallest increase worth detecting and the sample size needed to assess it.
That last decision connects the experiment to the budget. A team expecting a small change from a low baseline conversion rate may need a larger audience than it can reach. Checking feasibility before launch helps the manager decide whether this test can answer the spending question.
A 20% Increase Needs Context
At the end of the vendor’s observation period, 100 of the 10,000 control accounts have submitted a qualified demo request, compared with 120 of the 10,000 treatment accounts. The rates are 1% and 1.2%. The estimated difference is 0.2 percentage points, equivalent to a relative increase of 20%.
For the treatment group, that amounts to an estimated 20 additional accounts submitting a qualified request. With unequal group sizes, the analyst compares conversion rates and scales the difference to the treatment group’s size.
The next step is to calculate the uncertainty. Using a normal approximation for two independent account-level conversion rates, the estimated difference has a 95% confidence interval of approximately −0.09 to +0.49 percentage points. This is an illustrative calculation for the hypothetical experiment.
Because the interval includes zero, the result is inconclusive about whether the campaign increased qualified demo requests. The observed 20% relative lift is a point estimate, not a demonstrated improvement.
B2B Outcomes Take Time
For the software vendor, the observation window should reflect how long eligible accounts typically take to request a demo. Historical conversion delays help the team choose that window before launch. Both groups need the same follow-up period and qualification criteria.
Google Ads setup guidance allows user-based studies as short as seven days but generally recommends more than 14 days, particularly when conversions take longer. Those platform guidelines do not define a B2B sales cycle.
Demo requests and completed sales also require separate measurement. If the budget decision depends on revenue, the team should track sales over an appropriate period as well. Qualified requests can provide an earlier indicator, while sales outcomes take time to develop.
What the Result Means for the Budget
The vendor’s interval runs from a decline of about 0.09 percentage points to an increase of about 0.49 percentage points. The manager should report that range alongside the estimated 20% relative lift and classify the result as inconclusive. A headline announcing a proven 20% improvement would overstate the evidence.
Before commissioning another test, the team should assess whether a larger feasible sample could detect the minimum improvement needed to justify campaign costs. The team should plan the sample size, observation window, and analysis in advance. Repeatedly extending a test until a favorable result appears undermines that discipline.
Platform reports also require attention to their own methods. Google’s documentation on certainty of lift explains how it presents results, but if it showed a result like this one, a 20% lift with a range that crosses zero, the right reading would be the same: not yet proven.
In Atlantic Tech’sB2B data intelligence model, feedback connects data and execution. Kazan also emphasizes helping partners use the information they receive. For the vendor in this example, useful feedback means recording the uncertainty as clearly as the observed increase.
An incrementality testing report should give the manager the conversion rate in each group, the estimated absolute and relative lift, and its uncertainty. The budget decision then weighs that evidence against campaign cost and the smallest improvement the business needs.