Why connected TV advertising measurement needs a plan before launch

What to take away

Why measurement gets argued about

Two people can look at the same connected TV campaign and disagree about whether it worked. One is reading platform delivery data. The other is reading a brand tracker that moved two points. Neither is wrong. They are answering different questions with different instruments.

Connected TV advertising measurement is the practice of deciding in advance which question matters and which evidence will settle it. That sounds administrative. In practice it prevents most arguments, because the argument is usually about definitions rather than results.

It also protects the money. Streaming inventory is bought across several platforms, each with its own reporting conventions. Totals will not reconcile to the last decimal. A measurement plan tells you which discrepancy is tolerable and which one means something broke.

Definitions are the usual flashpoint. A completed view, a reach figure and a conversion window can all mean different things on different platforms.

One page is enough. Cover the decision, the primary metric, the data sources and the review date. That page will do more for trust than another dashboard.

Decide the question first

Start with one sentence: what decision will this campaign inform? A launch that needs reach wants different evidence from a retargeting push that needs conversions. Write the sentence into the brief so it survives the handover.

Then choose the metric that answers it. Reach and frequency for the first. Cost per acquisition for the second. Everything else becomes supporting evidence rather than the headline.

Set the review date at the same time. If the decision is made in a quarterly planning meeting, measurement that lands three weeks later is decoration.

Agree the secondary metric and the tripwire. Say in advance what result would make you move budget. A plan with no tripwire gives nobody anything to act on mid-flight.

Here is an illustrative one-page plan, for a six-week CTV test with a £40,000 media budget.

The budget, the target and the sources are illustrative. The shape is the point: one decision, one headline number, named sources, a date and a tripwire.

The metrics that carry weight

Most CTV reporting reduces to a handful of numbers. Delivery metrics show whether the buy ran as agreed. Response metrics show whether anything happened afterwards. Brand metrics sit between the two.

Which group leads depends on category and budget. A fuller breakdown of the measures UK advertisers track sits in this guide to connected TV advertising key metrics in England, which is worth reading before you fix your dashboard layout.

One caution. Do not put delivery and response on the same chart. They move on different clocks, and the comparison invites false conclusions about creative quality.

Panel data and platform data

Platform dashboards count what the platform served. They are accurate about delivery and weak on who was watching. Panel data works the other way: a recruited sample, weighted to a population, used to estimate viewing behaviour.

The trade-off is sample size against census size. Nielsen's 2023 explanation of what panel data is and why it matters sets out the logic, including why recruited panels are used where a full census of viewing is impractical.

That article is background from 2023, not current evidence on UK viewing levels. Use it for the method rather than the numbers.

In the UK, broadcast viewing estimates come from BARB, whose panel underpins the currency numbers agencies trade on. BARB publishes on a rolling basis, so those figures are current rather than archival. If you plan to quote those figures publicly, BARB's publishing guidelines for journalists explain how viewing data may be cited and what attribution is required.

For CTV, expect to blend three sources: platform delivery logs, an independent audience panel, and your own first-party response data. Each covers the blind spots of the others.

Panels have limits too. A small sample struggles with narrow audiences, and every estimate carries a margin of error. Ask for the confidence interval, not just the number.

Co-viewing and the household problem

A connected TV set is shared. One impression may reach two or three people, while the platform counts a single device. That gap matters when a plan is built on reach against adults.

Panel data can estimate co-viewing, because panellists report who was in the room or metering is tied to known household members. Platform data usually cannot.

Decide whether your reach target is household reach or person reach, and state that choice in the plan. Mixing the two produces plans that look over-delivered on one measure and under-delivered on the other, and the argument restarts.

Where person-level reach matters, ask suppliers what modelling sits behind the figure and how it was checked.

Attribution and its limits

Attribution tries to connect an ad exposure to a later action. On television, that connection is inferred rather than observed, because one household contains several people and several devices.

Household-level matching is the common approach. It tells you that a device in an exposed home visited your site. It does not tell you that the person who saw the ad was the one who visited. Treat the result as a lift estimate, not a receipt.

The practical fix is triangulation. Run a matched-market or holdout test where budget allows, then read the platform numbers alongside it. This comparison of connected TV advertising attribution methods explains where each technique is strong and where it tends to overstate.

Two errors recur most often. Crediting CTV with organic demand, and leaving view-through windows open for weeks. Ask what the window is before comparing anything.

Reporting rhythm and ownership

Reporting should have a rhythm, not a panic. A weekly delivery check catches under-delivery while there is still time to shift budget. A monthly performance read is enough for optimisation decisions. One end-of-campaign review closes the loop.

Thinkbox's guidance on how to measure results and review a TV campaign sets out the questions a review should answer, including whether the original objective still holds after the buy has run.

Name an owner for each number. The agency may own delivery, the brand may own response, and the data team may own the match. Unowned numbers drift.

Keep a single source of truth. A connected TV advertising reporting dashboard is only useful if everyone reads the same version.

Log every discrepancy as it appears, with a note on the cause. Over a year that log becomes the most useful document in the account, because it stops the same argument recurring.

Benchmarks and context

A number without a comparator is a mystery. Benchmarks supply the comparator: what a typical completed-view rate or cost per acquisition looks like for your category and buy type.

Typical figures for UK CTV give you a starting range. Completed-view rates are commonly reported above 90%, with most non-skippable CTV and broadcaster VOD campaigns landing in the mid 90s. CPMs for UK CTV and broadcaster VOD typically sit between £15 and £40 per thousand impressions, depending on targeting, inventory type and season.

Cost per completed view follows from those two numbers. A £20 CPM with a 95% completion rate works out at about £21 per thousand completed views. That is arithmetic, not a benchmark, and it shows how sensitive the metric is to the CPM you actually pay.

Cost per acquisition resists a single range, because it depends on category, price point, conversion rate and attribution window. Build it from your own conversion rate and the CPM band above.

Use benchmarks with care. They are averages drawn from other campaigns, with other creatives and other budgets. They tell you whether a result is unusual. They do not tell you whether it is good.

The IPA's effectiveness resources are a sensible starting point for framing what good looks like across channels, including digital, because they deal in published evidence rather than vendor claims.

Our connected TV advertising benchmark research covers how to build a comparator set you can defend in a planning meeting, including how to weight sources.

For example, a brand running a £30,000 four-week test might compare its cost per completed view with two or three campaigns of similar length rather than with a category average built from year-round activity.

First-party data and identity

Your own data is the part of the stack nobody else controls. Site visits, app events, CRM records and sales files all help, provided the identifiers survive the media buy.

Often they do not. App-to-web handoffs lose the identifier. Shared devices blur it. Consent choices remove it for part of the audience.

Plan for partial matching. A match rate covering perhaps half of exposed households still supports directional comparison, as long as you never present it as complete. Record the match rate next to every result.

Owned signals also give you the quickest read on whether anything happened. They are not a substitute for independent audience data, but they arrive sooner.

The programme side of CTV

Much CTV inventory is bought programmatically, which shapes what you can measure. Auction-level data gives granular delivery reporting but limited audience information unless you supply it yourself. Before launch, decide which audience signals you will pass in the bid request and what you expect back in reporting.

IAB UK's Back to Basics guide to programmatic is a useful primer on how the chain works, from supply through to verification, and where data is lost between the two. Read it before you fix your reporting fields, because the losses it describes become gaps in your dashboard.

Practical consequences follow for the plan. Deal IDs and private marketplaces usually give better transparency than the open auction. Supply-path checks matter more when inventory is aggregated across many sellers. Ask your partner which signals travel with the impression, and in what form, because that answer sets the ceiling on what you can report.

Log-level data is the raw material for your own analysis. Ask whether you receive it, at what level of aggregation, and how long it is retained. Those answers shape what you can ever measure.

A pre-launch checklist

Use this before the campaign goes live, not after the first report disappoints.

Planning:

Delivery and reporting:

The last point is not optional. Consent for measurement, and the lawful basis for matching data, sit inside UK GDPR and the advertising codes. Our guide to UK rules and compliance for connected TV advertising sets out the main obligations, and it is worth a read before your legal team asks.

Common questions

How long should a CTV measurement window run?

Long enough to cover the purchase cycle you are measuring. A two-week window suits fast response categories. Considered purchases need longer. Fix the window before launch so nobody extends it after a weak result.

Can I trust platform-reported reach?

For delivery, broadly yes. For deduplicated reach across platforms, no single platform can see the others. Use an independent panel or a cross-platform currency where you need a total.

What is the minimum useful measurement setup?

One primary metric, one independent audience source, and one owned response signal. That combination lets you check the platform's story against something it does not control.

Who should own CTV measurement?

One named person, even if several teams contribute. The owner reconciles definitions, signs off the dashboard and answers questions when the numbers disagree.