Programmatic KPIs only help when they start from your business objective, not from whatever the platform dashboard shows. The dashboard provides available data, but it should not decide what success means. This guide shows how to measure programmatic advertising on three levels of evidence, how to assess incremental reach, and how attribution vs incrementality changes what you can claim.

Key takeaways

• Work from the business objective to the campaign objective, then to the KPI.
• Keep delivery, campaign and commercial evidence apart, because each answers a different question.
• Incremental reach depends on what you can observe across activities, so report how it was estimated.
• Attribution vs incrementality: attribution assigns credit, while incrementality asks what the advertising caused.
• Plan controls and holdouts before the campaign, because you cannot create them afterwards.

How do you choose the right programmatic KPIs?

Choose programmatic KPIs by working from the business objective to the campaign objective, and only then to the KPI. In other words, the metric follows the question, not the other way round.

For example, a business wants more qualified enquiries in a region. A campaign might support that by introducing the brand to relevant future buyers. Its immediate measures could include suitable audience delivery and consideration. Meanwhile, qualified enquiries remain the commercial outcome to investigate. To keep those questions separate, use three levels of evidence.

How to measure programmatic advertising on three levels of evidence

Each level answers its own question and has its own main limitation.

Commercial outcomes

Commercial outcomes answer the question: did the business result change? Examples include confirmed sales, revenue, qualified leads and new customers. However, other activity and market conditions also affect the result.

Campaign outcomes

Campaign outcomes ask whether the activity fulfilled its audience or advertising job. Examples are relevant reach, awareness, recall, consideration or brand lift. The limitation is that some of these measures require research, estimates or a comparison group.

Delivery and channel metrics

Delivery and channel metrics ask whether the campaign ran as intended. Here you look at impressions, CPM (cost per thousand impressions), viewability, completion, traffic quality and pacing. Treat them as your programmatic KPIs for delivery. Still, good delivery alone does not establish a business effect.

These levels work together. Poor delivery helps explain a weak audience result. In contrast, good delivery with no useful response suggests a different question about the audience, message or role. Some metrics, such as completion, can help diagnose delivery and also support assessment of the advertising job. Therefore, define how you are using them so their role in your programmatic KPIs is clear.

Which KPIs fit demand creation and demand capture?

The right evidence depends on the advertising job, as our guide to brand building and performance marketing explains.

For demand creation, begin with whether the intended audience received suitable exposure. Reach and frequency describe that delivery. Where available, completion and attention can add information about the experience. Awareness, recall or consideration research can then assess a change in how the audience thinks about the brand. Branded Search and direct traffic can provide supporting demand signals, but they can also change for other reasons.

For demand capture, on the other hand, your programmatic KPIs evaluate the agreed action: qualified visits, enquiries, sales or another conversion. Connect platform reporting with lead quality, confirmed customers or revenue where possible. After all, a low cost per lead is less useful if the leads do not become customers.

If both jobs are present, keep separate sections in the report. Also compare campaigns only when they share the same job and compatible definitions. A creation campaign’s lower last-click conversion rate does not by itself show failure, because its own objective still needs evaluating.

What can each type of evidence tell you about programmatic results?

Every evidence type behind your programmatic KPIs tells you something useful, but none tells you everything. For instance, platform and verification delivery reports show recorded delivery and quality within the provider’s scope. They cannot show that every impression was noticed or caused a purchase.

Similarly, reach and frequency estimates describe exposure among recognised devices, households or modelled audiences, not exact unique people across every platform. Search, direct traffic and sales trends show what changed over the same period or in the same area, but not that this campaign alone caused the change. Finally, a well-designed controlled test estimates the effect against a valid comparison, yet it is not immune to sampling, spillover or design limitations.

Reports should therefore state the unit, coverage and method behind each number. “Measured”, “estimated” and “inferred” describe different kinds of evidence, so do not use them interchangeably.

How do you assess incremental reach?

Incremental reach asks how many people one activity reached who were not already reached by another. It is a valuable question, but the answer depends on what you can observe across those activities.

Within the DSP buy

Where recognition and reporting support it, a DSP (demand-side platform) can estimate deduplicated reach and overlap across the inventory it bought. Ask whether it counts devices, households or modelled people, and which environments are covered. This helps you assess overlap between parts of the programmatic campaign. It does not, however, automatically reveal overlap with separate Google, Meta or other activity.

Customer or visitor exclusions

Suppression can support a prospecting strategy by excluding recognised customers or recent visitors. Report which lists you used, their dates and their match coverage. Yet it is an imperfect proxy for reaching beyond a known audience. Unmatched customers can remain eligible, and recent visitors represent only part of the people reached by existing advertising. As a result, exclusions do not prove person-level incremental reach across the whole plan.

Research or panels

Audience research and cross-media panels may estimate overlap across covered media, depending on campaign scale, sample, platform coverage and cost. Ask which exposures are directly observed and which are modelled. A modelled cross-platform number may be useful, but present it as an estimate with assumptions rather than a complete count.

Downstream signals

Branded Search, direct visits and qualified enquiries may provide evidence consistent with additional useful exposure. They are not reach measures, though. A suitable comparison region or control can strengthen the interpretation, if the design is valid.

For a first campaign, agree which combination of these approaches is sufficient to guide the next decision. If the central hypothesis requires an exact cross-platform count of incremental reach that the setup cannot provide, change the question or fund an appropriate method.

Attribution vs incrementality: what is the difference?

Attribution assigns credit to observed touchpoints, while incrementality asks what difference the advertising caused. That is the core of attribution vs incrementality, and it decides how strong a claim your programmatic KPIs can support.

Attribution helps describe recorded conversion paths and can support routine optimisation. It does not show what would have happened without the advertising. Incrementality, by contrast, needs a valid comparison, sufficient scale and a measurable outcome. Randomised audience holdouts or carefully designed geographic tests are possible approaches, each with different practical requirements.

Lift describes a measured change in an outcome, and brand lift studies often use surveys to compare groups. Whether they support a causal conclusion depends on how the groups were created, the sample and the study design.

One properly designed test is more useful than several studies too small to answer their questions. If a test is infeasible, report the weaker evidence honestly instead of putting an incrementality label on it. Keeping attribution vs incrementality clear in the report protects you from claiming more than you know.

Plan studies before the campaign

Controls, holdouts, eligibility rules and any required baseline must be in place before the relevant exposure. You cannot recreate a group that was never exposed once everyone has had the opportunity to see the campaign. Analysis can happen later, and survey fieldwork may happen during, near the end of or shortly after the flight. So agree the schedule with the provider to measure recall at an appropriate time.

Attention measurement also needs a method suited to the question. Attention measurement guidance from the IAB (Interactive Advertising Bureau) and the Media Rating Council (MRC) provides a framework, while providers can use different approaches. Ask whether the result is directly observed or modelled, what inventory is covered and which comparisons are valid. Attention adds evidence about exposure, but it does not replace commercial measurement.

Why do programmatic reports disagree?

DSP conversions, verification figures and website analytics may disagree. Instead of averaging the numbers behind your programmatic KPIs or choosing the largest one, investigate the reasons.

Common causes include different attribution windows, click-through versus view-through counting, deduplication, time zones, consent or recognition coverage and implementation errors. For example, your analytics may credit only a click, while a platform credits a sale after a view within its attribution window. The difference is partly definitional. It is not automatically missing sales or a broken tag.

Check tagging and definitions first. Then document which source you will use for each question and how you will handle known differences consistently.

How should you review a first programmatic campaign?

Agree before the review what evidence is sufficient. The limits of a first campaign should narrow the claim, not remove accountability. Decide which evidence is essential to the capability hypothesis you set when you planned your first campaign, which signals are supporting context and which outcomes remain beyond the setup’s coverage. If the available evidence cannot answer the central question, change the question or commission the missing method before spending.

Then review the campaign in four questions:

1. Did the campaign deliver what we planned? Check scope, budget, quality and any material changes.
2. Did the capability we chose work? Assess the hypothesis against the evidence agreed before launch.
3. Did the intended audience respond as expected? Evaluate the relevant campaign outcomes, separately for each job.
4. Did commercial results change, and how confident are we that the activity contributed? Separate controlled evidence from trends and interpretation.

The review should distinguish a delivery failure from an ineffective message, and an observed change from a causal contribution. A campaign can deliver as planned without producing its intended effect. It can also produce useful learning without yet establishing incremental revenue. So record what remains uncertain and make the next decision explicit.

Agree before launch what evidence would justify a second test, continued investment or stopping. The final report should support that decision rather than force every result into a single score. That decision then feeds your wider programmatic roadmap for 2027.

FAQ

Where to go from here

Good programmatic KPIs do not force every result into one score. Instead, they show what was measured, what was estimated and what remains uncertain, so your next decision rests on evidence.

Our playbook Getting Started with Programmatic Advertising in 2027 shows you how to plan, run and review a first campaign.