Are Banner Ads Working? Measure Reach, Conversions, and Incremental Results - Yenra

Read the measurement chain in order

On a small screen, scroll the table sideways to read all columns.

Google's Active View documentation counts a standard display impression as viewable when at least half the ad is visible for at least one second. Its large-display exception uses a 30 percent area threshold for ads of at least 242,500 pixels. Check the applicable format and report definition instead of applying the standard-display rule to every placement.

Also inspect measurability. A viewable rate calculated from measurable impressions can look good while a substantial share of delivery cannot be measured. Ask for both the viewable count and the measurable count. No viewability threshold proves that a person paid attention, understood the message, or was a suitable customer.

Find out what received credit

Choose a business outcome with a precise definition: a completed first order after cancellations, an attended demonstration, or an eligible application. Keep page visits and button clicks as diagnostic events. If the platform optimizes for a shallow event, it can improve that event while the outcome you care about remains flat.

A view-through conversion records an eligible later action after an ad impression without a qualifying click under the platform's rules. It can describe a useful path, but the same person may already have intended to buy. Google's conversion-window documentation distinguishes click-through, engaged-view, and view-through windows. Record those settings alongside results; changing a window changes what is eligible to be counted.

Inspect prospecting and existing-customer activity separately. Showing banners to frequent buyers may collect substantial conversion credit even if the advertising adds little. Likewise, a buyer may interact with several channels before ordering. Adding each platform's attributed purchases can count the same sale more than once.

Reconcile to your order or customer records using authorized data and a consistent period. Document reporting delays, returns, canceled orders, time zones, and whether the report uses interaction date or conversion date. A mismatch is something to explain, not automatically evidence that one system is broken.

Use a comparison that can answer the causal question

A randomized holdout assigns eligible people to treatment and control before exposure. The treatment group is eligible for the campaign; the control group is withheld from it. Compare outcomes across the assigned groups, including people who never received an impression. Comparing only people who happened to see or click ads with everyone else introduces selection differences.

Google's Conversion Lift overview describes user-based and geographic approaches using treatment and control groups. The product is not available to every account. A geographic test also needs enough comparable regions and a design that handles regional variation; it is not simply a before-and-after sales chart.

Download the holdout calculation example (CSV), which includes the inputs, computed results, and limits. A real analysis should report an uncertainty interval and use a sample-size and duration plan suited to the decision. If the interval includes no lift, the evidence may be inconclusive even when the point estimate is positive.

Avoid changing the offer halfway through, stopping as soon as results look favorable, or allowing other campaign changes to differ systematically between groups. Record cross-device leakage and incomplete outcome measurement. If a credible experiment is unavailable, present attribution and trend evidence with that limitation rather than labeling it incremental impact.

Troubleshoot before blaming the creative

Delivery is high but viewability is low: inspect placement and device reports, format compatibility, and measurable inventory. Ask whether the campaign is buying the intended opportunity to see.

Clicks are high but useful visits are scarce: test page speed and the landing-page promise, then investigate accidental clicks, irrelevant placements, and invalid traffic. Google's invalid-traffic explanation includes activity that does not represent genuine user interest, such as automated or accidental clicks. A poor conversion rate alone does not prove fraud.

Attribution is strong but overall sales are flat: examine existing-customer concentration, view-through settings, channel overlap, and the purchase cycle. A period of flat sales can have many causes. Use these findings to design the next comparison rather than assuming either success or failure.

Write the decision with its evidence

Keep a short record of the objective, audience, placement, creative version, spend, outcome definition, attribution windows, and study design. Record what changed since the previous review. Renew or expand only when the evidence supports the business objective at a tolerable cost.

A useful conclusion might be: “Delivery is measurable, qualified visits increased, and the holdout estimate is positive but too uncertain to support a budget increase. Keep spending capped while completing the planned test.” That tells the next reviewer more than a green dashboard badge.