Media buying analytics in 2026: from CPM to CPA, and why not CTR — the four metrics we actually decide on

An advertiser dashboard still typically blinks 20-30 metrics. Most are noise. After two years of internal audit we kept four as the basis for decisions — the rest is diagnostic. Here are the four, and why CTR is not on the list.

Why not CTR

Let us start with the absentee. CTR (click-through rate) is the classic first row of any adtech report — and in 2026 it is only relevant when the click itself is the campaign objective. In 80% of cases it is not.

Two concrete reasons that kill CTR as a primary metric:

Bot and accidental-click noise: on the numerator side, accidental mobile banner taps account for 18-31% of clicks. Raw CTR is contaminated, and cross-campaign comparison is illusory.

Click ≠ intent: a campaign with 0.8% CTR can produce more high-intent clicks than a 1.4% one if the latter is poorly targeted. CTR-based ranking drives bad allocation decisions.

We observe CTR (because it is an anomaly signal) but do not optimise for it. The four we do optimise:

1. CPA (Cost per Acquisition) — in context

CPA is cost per advertiser-side conversion (purchase, lead, signup). Numerator: total spend. Denominator: attributed conversions.

The 'in context' part is critical: raw CPA is misleading on its own, because 4,800 HUF is good against a 12,000 HUF average order, and a disaster against a 2,400 HUF one. We always tie CPA to the advertiser's CAC target, fixed as a context parameter in the campaign configuration. The dashboard always shows CPA alongside its target ratio.

In 2025 our average publisher-side conversion CPA was 3,200 HUF at a target ratio of 0.78 (i.e. we ran 22% under target on average). This is not 'the metric's max value', it is a strict feedback loop.

2. ROAS (Return on Ad Spend) — Net, not Gross

ROAS is revenue generated per unit of ad spend. Net ROAS = Net revenue / Ad spend, meaning returns, cancellations, and refunds are subtracted.

Net vs. Gross is not cosmetic: in fashion and FMCG campaigns Gross ROAS runs 8-14% higher than Net because returns and unpaid invoices distort it. A Gross ROAS of 4.2 may be a Net ROAS of 3.7 — still positive, but it leads to a meaningfully different optimisation decision.

We measure Net ROAS on a 14-day cohort because that is the typical return window. A shorter window overstates ROAS.

3. Attentive Reach — successor to 'viewable impression'

The 'viewable impression' concept (50% of pixels on screen for ≥1 second) is a 2014 metric and is no longer enough in 2026. A banner can be 'viewable' while the user is scrolling past in half a second.

Attentive Reach has two components: viewable impression plus a minimum 2-second dwell time over the banner. Without eye tracking, 2 seconds is a reliable gaze proxy.

The measurable difference: in an average publisher campaign viewable impressions hit 78%, attentive reach 41%. It halves. Advertisers find this initially uncomfortable, but cross-checking with CPA and ROAS clears it up fast: attentive reach correlates strongly with downstream conversion (Pearson 0.61), viewable impression does not (Pearson 0.18).

4. Incremental Lift — the control-group metric

This is the most expensive and most complicated of the four, which is why it often does not appear on cheap dashboards. Incremental lift measures how many conversions the campaign causally generated above the control-group baseline.

Mechanics: 5% of the audience is held out — they do not see the campaign. The holdout conversion rate is subtracted from the exposed group's. The difference is incremental lift.

It often turns out that a campaign with 'measured ROAS 5.0' has incremental ROAS of 2.3 — because most conversions were attributed to buyers who would have bought without the ad. The campaign was not worth running.

The 5% holdout is expensive (5% of forfeit revenue), but 80% of our campaign-strategy decisions hinge on this measurement. An advertiser not measuring this is optimising blindly against the 'attributed-as-if' number.

The data pipeline behind it

The four metrics are near-real-time — CPA and ROAS at 30-minute lag, Attentive Reach at 5-minute lag, Incremental Lift once daily (because it waits for the 14-day cohort). The pipeline runs on ClickHouse, 18 billion event rows, replicated across two DCs. Not heavier infra than the brand-safety pipeline, but with stricter data-integrity requirements — a lost event row leads to a Net ROAS number error, which is money.

Takeaway

The four metrics — CPA in context, Net ROAS, Attentive Reach, Incremental Lift — are not 'the important four among many', they are the four we decide on. The rest (CTR, viewability, frequency, even-distribution) is diagnostic. If your advertiser dashboard shows 12 numbers with equal weight, the decision-maker does not know where to look — and will not look anywhere.