Impressions tell you an ad was served. Completion rate tells you it played to the end. Neither tells you whether the campaign changed anything. For a brand's first CTV campaigns in APAC, the metrics worth agreeing before the flight are fewer than the dashboards suggest, and most of them are about reach, lift and proof.
Reach and frequency, by household
Start with unique households reached, deduplicated across every publisher and app on the plan. Then read the frequency distribution, not the average. An average of 4 can hide a plan where a third of households saw the ad once and a small group saw it 15 times. The shape shows whether the budget built reach or spent itself on the same screens.
Frequency deserves attention in this region. Research published through IAB SEA+India in 2025 found that 2 in 3 consumers in Southeast Asia report fatigue from seeing the same ads repeatedly on the same channels. Put a frequency cap in the plan, and ask each publisher for the distribution by household in its delivery report.
Deduplication is the hard part. Each publisher counts its own households, and stitching them together is limited in most APAC markets. Where you cannot deduplicate, report reach per publisher and say so in the summary, rather than adding the numbers up.
Brand lift
A brand lift study compares survey answers from an exposed group with a control group that did not see the ad: ad recall, awareness, consideration, intent. It is the standard measure for a brand-building flight.
It is also blunt. Manmohan Singh makes the point in a 2026 piece on LinkedIn: a campaign that reached millions of verified households is often judged on a survey of a few hundred people, with a confidence interval wide enough to hide the result. Ask the provider for the sample size per cell and the interval, not only the point lift. On a small test flight, the money is often better spent elsewhere.
Search lift
Branded search during the flight and for a couple of weeks after is cheap to read and often moves first. Compare it with the same period before the campaign and, where you can, with a region that did not run CTV. It correlates with awareness. It does not prove it, because seasonality, PR and other media move search too.
Incrementality, and why view-through is not it
Nobody clicks a television. When a household sees an ad on Tuesday, searches on Thursday and buys, last-click gives the credit to search. View-through attribution was the patch: it counts conversions from households that were exposed. The flaw is that it has no counterfactual. It cannot say whether those households would have bought anyway.
Singh's worked example shows the size of the gap. On one campaign, view-through put the cost per conversion at USD 23 and a randomised holdout put it at USD 112, for the same spend. That ratio is not a rule, but the direction is common. View-through flatters, because the households a campaign reaches are often the ones already close to buying.
The fix is a holdout. Withhold the ad from a random share of eligible households, or from matched regions, and compare outcomes. In much of APAC, household-level matching is limited, so a geo holdout is usually the practical choice: run in some cities or provinces, hold out comparable ones, and compare search, site visits or sales. It costs some reach and needs enough volume to read. For a first campaign, one clean geo test tells you more than any number of view-through reports.
Buyers have reason to ask for this. A 2026 survey by a CTV buying platform, reported in The Drum, found only about a third of marketers fully trust the performance numbers platforms report.
Completion rate is hygiene
On non-skippable CTV inventory, video completion rate is high by design. It confirms the ad played. It says nothing about whether it worked. Watch it for anomalies instead: a near-perfect completion rate across a long list of unfamiliar apps is a reason to ask questions.
That is where fraud sits. CTV bought from known publishers carries little of it. The risk lives in the long tail. In the US, Nielsen data summarised by analyst Michael Beach in 2026 show 9 companies accounting for 69% of TV viewing time. Ad fraud researcher Augustine Fou asks the obvious question: if viewing is that concentrated, who is watching the tens of thousands of other apps that show up in programmatic CTV logs? He also notes that a CTV bid request can be fabricated without any device at all. Ask for delivery by named app and publisher. If a report cannot say where the ad ran, its completion rate means little.
The same applies to CPM. A cheap CPM on inventory you cannot verify, with no measurable effect, costs more than a higher one that moved awareness.
APAC measurement reality
Maturity varies by market. Australia and Singapore have more developed panels and attribution tools; Indonesia, Thailand, Vietnam and the Philippines have less. Cross-publisher deduplication is limited almost everywhere.
Co-viewing is common. Several people often watch one screen, so household impressions understate the people reached. By how much depends on the market, the publisher and the content. Ask each publisher how it estimates co-viewing and treat the result as an estimate, not a multiplier to apply to your reach.
A framework for the first 3 campaigns
Campaign 1. Household reach and frequency distribution per publisher, delivery by named app, completion rate as a hygiene check, and search lift against a pre-period. Write the measurement plan into the insertion order or deal terms before the flight starts.
Campaign 2. Add a geo holdout. Pick matched regions, hold one or more out, and read search and site visits or sales in both.
Campaign 3. Add brand lift if the budget supports a sample large enough to read, and set its result next to what the holdout showed.
Throughout, read delivery against the plan line by line, shortfall first. Measurement only means something if you know what actually ran.
Where agents fit
Measurement starts with knowing what was bought. When a buying agent such as Skopa agrees a plan with each publisher's selling agent, every line names the publisher, the placement and the flight, and delivery is read against that plan on a shared record. That makes the questions above easier to ask. The trader still decides what to measure and what the answer means.