Display Advertising: What It Buys and How to Know Whether It Worked

Display advertising buys the opportunity to put a visual message in front of people while they use websites, apps, and other digital properties. It can reach potential buyers before they search for a solution or bring previous visitors back into consideration. It does not, by itself, buy attention, leads, or revenue.

That distinction is the basis of a useful display plan. Decide which business outcome the campaign must change, define the delivery that would make that change plausible, and separate platform-attributed results from outcomes the advertising actually caused.

What counts as display advertising?#

Display ads combine visual elements and copy, usually with a call to action linked to a landing page. They may appear around or within the content someone is consuming. A banner is one familiar display format, but it is not the whole channel. Responsive display systems can assemble headlines, descriptions, images, logos, and video into ads that adjust to available inventory. Amazon Ads describes the basic format, while Google documents how responsive display ads are assembled.

Two distinctions prevent common planning errors:

• Display versus search: Search advertising appears in response to a query. Display advertising appears while someone is using other content. Display can therefore reach people before they express demand in search, but the context usually provides weaker evidence of immediate intent.
• Display versus programmatic: Display names an advertising format and inventory context. Programmatic names an automated method of buying and selling inventory. The IAB definition of programmatic is broader than open-auction display; display inventory can also be bought directly from a publisher.

The practical implication is simple: “We are buying programmatic” does not identify the campaign’s format, audience, inventory, or measurement standard. Those decisions still need to be made.

Use display for a defined reach or re-engagement job#

Display is most defensible when the business needs to reach an eligible market beyond active search demand, or to re-engage people with a relevant prior relationship. Those are different jobs and should usually be measured separately.

A prospecting campaign may use page context, topics, named placements, or inferred audience characteristics. A re-engagement campaign may use prior site activity or customer data. Google’s documentation distinguishes content targeting such as topics, placements, and keywords from audiences such as affinity, in-market, remarketing, demographic, and Customer Match segments.

These controls determine who or what is eligible for an impression. They do not establish that the person noticed the ad or that the impression influenced a decision. This matters especially in retargeting: previous visitors are already selected for demonstrated interest, so their high conversion rate may partly reflect what they would have done without another ad.

Before launch, define both sides of targeting:

• Who may receive the ad: the account, role, market, prior interaction, geography, device, and recency conditions that make an exposure relevant.
• Where the ad may appear: acceptable content, publishers, apps, placements, and exclusions, plus any automated expansion the platform may apply.

Then inspect actual delivery against those definitions. A precise audience description is not evidence that the campaign stayed inside it.

Measure delivery before interpreting performance#

A display dashboard compresses several different events. Reading them as one funnel creates false confidence.

Unique reach is not simply impressions with duplicates removed. Google says its Unique Reach metrics estimate people across devices, formats, sites, apps, and networks. Treat the result as modeled reach and keep its reporting scope intact when calculating frequency.

Viewability is a narrower delivery test. Under Google Active View’s standard for display, at least 50% of an ad’s pixels must be in view for at least one continuous second. Google documents the threshold, and the IAB/MRC attention guidelines make the boundary explicit: viewability represents an opportunity to see, whether or not a person actually saw the ad. A viewable impression is better delivery evidence than an unqualified served impression, but it is not an attention metric.

Frequency also needs more than one average. If a campaign serves 600,000 impressions to an estimated 200,000 people, average frequency is three. That could mean fairly even repetition, or many one-time exposures combined with a small group seeing the ad repeatedly. Review frequency buckets and placement concentration before concluding that more impressions created more useful reach.

Match the cost denominator to the decision#

The standard arithmetic is straightforward:

CPM = media cost / impressions × 1,000
Average CPC = click cost / clicks
Average CPA = conversion cost / attributed conversions
Incremental ROAS = incremental conversion value / media spend

Google’s definitions confirm the denominators for CPM, average CPC, and average CPA. The formulas are not interchangeable verdicts.

For an awareness job, CPM can normalize delivery cost, but it says nothing about unique reach or viewability. Viewable CPM narrows the denominator to measured viewable impressions, but still stops short of attention. For an acquisition job, CPA is useful only when “conversion” names an outcome the business values. A form submission, accepted sales opportunity, and closed sale should not share one undifferentiated CPA.

There is therefore no platform-neutral “good display CPM” or CPA. A benchmark becomes decision-useful only after inventory, audience, geography, billable event, fees, conversion definition, and time window are comparable. Cheap impressions are not efficient when they reach the wrong market; a low attributed CPA is not efficient when the credited actions were likely to happen anyway.

Attribution gives credit; incrementality estimates causation#

Display reporting often credits conversions that follow an impression even without a click. Google describes a view-through conversion as an action completed after someone viewed but did not interact with an ad; for Display Network ads, the last viewable impression receives the credit. The configured window determines how long an impression remains eligible. Google’s documentation explains both the credit rule and view-through reporting.

That record answers, “Did a defined conversion occur after an eligible impression?” It does not answer, “Would the conversion have occurred without that impression?” A longer window can capture a longer buying cycle, but it can also admit more conversions that merely followed exposure.

Incrementality addresses the missing counterfactual. A controlled lift study compares a treatment group that can receive the advertising with a control group that cannot. The difference estimates the outcomes caused by the tested campaign. Google’s overview of lift studies describes this treatment-control logic; its user-based Conversion Lift reporting distinguishes incremental ROAS from ordinary attributed ROAS.

The strongest feasible design depends on the operating environment:

• A randomized user holdout can estimate campaign lift when assignment and outcome matching are available.
• A randomized or carefully matched geography test can capture broader outcomes when user-level assignment is unavailable.
• A modeled counterfactual can estimate a baseline when a clean experiment is infeasible, but its conclusion depends more heavily on assumptions and data quality.

The IAB and IAB Europe incrementality guidelines rate explicit test-versus-control approaches as causally strong while noting their cost, time, contamination, and scope limits. They also warn that modeled counterfactuals are vulnerable to selection, omitted-variable, and data-quality bias.

A lift result is not a permanent return for “display advertising.” It belongs to the tested audience, creative, inventory, outcome, budget, and period. Read its uncertainty interval, check whether control users encountered overlapping campaigns, and ensure the measurement window covers the buying cycle. Geography tests add their own constraints: small numbers of dissimilar regions and shared budget effects can complicate causal estimates, as Google Research’s work on randomized geo experiments explains.

Write the display decision before choosing settings#

A usable campaign brief needs six decisions:

Scale only when the campaign reached the intended market, the resulting outcome passed the business quality threshold, and credible counterfactual evidence supports lift. If a causal test is not feasible, state the narrower claim honestly: the spend purchased controlled reach and learning, while reported conversions remain attributed rather than proven incremental.