How to Measure CTV Ad Performance Beyond Impressions

CTV ad performance is measured in two layers. Delivery metrics, such as impressions, completion rate, reach and frequency, show that the ad played on a TV. Outcome measures, such as site visits, branded searches, calls and incremental lift, show what it changed. Decide the question first, check where the ads ran, then compare people who saw them with people who didn't. This guide is part of our paid advertising guide.

You'll need conversion tracking on your website and access to your CTV partner's reports. A holdout group or a comparison area makes step 7 possible.

What CTV reports can and can't tell you

A connected TV ad has no click. Nobody taps the screen in a living room to visit your site. So the click-through rate that anchors search and social reporting has no equivalent here, and a CTV campaign judged on clicks will always look like a failure.

What a CTV report does show is delivery: whether the ad was served, how much of it played, how many households saw it and how often. Those numbers come from tracking events built into the ad itself. They prove the ad ran. They don't prove anyone did anything afterward.

Outcomes need a second layer. People usually act after a TV ad on a different device, often days later. Measuring that means linking households that saw the ad to later visits or sales, and then comparing them with households that didn't see it. If you're new to how these ads are bought and delivered, start with OTT advertising.

Steps

1. 01
   Write down the question the campaign must answer.

   Pick one main question, such as "Did more people from our target ZIP codes visit the site?" or "Did branded searches rise during the flight?" The question decides which data you need and how long to run. Write it down with the campaign dates, budget split and creative before launch, so the result can't be reinterpreted afterward.

2. 02
   Agree how each delivery metric is counted.

   Ask the partner to define an impression, a completed view, reach and frequency in writing. Ask whether reach counts households or devices, and whether frequency is capped per household. Two partners can use the same word for different counts, so agree the definitions before you compare anyone's numbers.

3. 03
   Ask for delivery by device type and app.

   A buy sold as streaming TV can include phones and computers. In OpenRTB, connected TV is its own device type, so a partner can split delivery by screen. Ask for impressions on TVs versus other devices, and a list of the apps or channels where ads ran. We explain the device split in OTT vs CTV vs streaming TV.

4. 04
   Check that tracking fires, including with server-side insertion.

   CTV delivery is counted through tracking events in the VAST ad response. With server-side ad insertion, those events may be fired by a server rather than by the TV. Ask how the partner handles this, and whether independent verification through Open Measurement is supported on the apps you buy.

   A trimmed VAST snippet shows where the quartile events live:

   html

   <Linear>
     <Duration>00:00:30</Duration>
     <TrackingEvents>
       <Tracking event="start">https://example.com/t?e=start</Tracking>
       <Tracking event="firstQuartile">https://example.com/t?e=q1</Tracking>
       <Tracking event="midpoint">https://example.com/t?e=mid</Tracking>
       <Tracking event="thirdQuartile">https://example.com/t?e=q3</Tracking>
       <Tracking event="complete">https://example.com/t?e=done</Tracking>
     </TrackingEvents>
   </Linear>

5. 05
   Give every spot its own creative ID.

   Use a separate ID for each version, such as the :15 and the :30. VAST has a field for a registered ad ID, and IAB Tech Lab says registration helps with frequency capping and cross-platform measurement. Even without registration, consistent creative names let you compare versions and spot reports that mix them up.

6. 06
   Connect exposure to visits and conversions.

   Make sure your site tracks the actions that matter: form fills, calls, bookings and sales. Then ask the partner how they link households that saw the ad to later site visits, usually through the home's internet connection. Track branded searches and direct visits for the same period, because many viewers will type your name rather than follow any link.

7. 07
   Set up a comparison group before launch.

   Where your partner offers it, hold back a share of the target audience that sees no ads. If not, compare similar areas with and without the campaign. Either way, set it up before the first impression. A comparison group is what turns "people who saw the ad visited" into "the ad caused extra visits".

8. 08
   Read results over the full window and report by question.

   Wait for the measurement window you agreed in step 1, then report against your question first. Put delivery, device split, outcome measures and lift side by side. Note anything that changed mid-flight, such as new creative or a sale, because it can explain a result as well as the ads can.

If you'd rather have the spot produced and the measurement set up and read for you, our in-house ads team runs OTT and CTV campaigns alongside the creative.

Delivery metrics and what they count

Delivery metrics come first in every CTV report. This table shows what each one counts and the question to ask about it.

Use the table to build your own log. One row per campaign, with the question, the definitions agreed, the delivery numbers and the outcome measures, makes the next campaign much easier to judge.

Reading a lift result

Lift is the difference between the exposed group and the comparison group. If households that saw the ad visited your site more often than similar households that didn't, the gap is the campaign's effect. Visits from people who would have come anyway are already in both groups, so they cancel out.

Read lift with care. A small comparison group gives a noisy result, so a gap that looks large may not hold up. A comparison area can differ in ways that have nothing to do with ads, such as weather, a local event or a competitor's sale. And lift only covers the outcome you measured. Branded searches can rise even when site visits don't.

The same logic applies when CTV follows another channel. A geofenced audience that later sees a streaming TV spot can be measured against fenced visitors who didn't. We cover fence setup and holdouts in geofence radius and audience best practices.

To put CTV results next to your other channels, see how to track paid media ROI.

Troubleshooting

Completion rates look almost perfect

Cause: the ads run in breaks that can't be skipped, so nearly every start reaches the end. That's normal for many CTV formats and says little about attention. It can also mean completions are counted by a server that assumes the stitched stream played through, rather than by the TV reporting each quartile.

Fix: ask whether the format was skippable and how quartile events were fired, using the server-side insertion question from step 4. Stop using completion rate as a success measure for unskippable placements and use it only to compare formats or spot lengths. Judge the campaign on reach, frequency and the outcome measures from steps 6 and 7 instead.

Site visits can't be tied to the campaign

Cause: conversion tracking on the site is missing or incomplete, the partner can't match households to visits, or the visits happen on phones that don't share the home's connection. Many viewers also search for your name instead of typing a web address, so the visit shows up as search traffic rather than as a CTV visit.

Fix: check that site tracking fires on every key action before the next flight. Ask the partner what share of exposed households they could match, and how. Add branded search volume, direct visits and calls to your report for the same dates. Use the comparison group from step 7, so you measure the gap rather than chasing single visits.

Two partners' numbers don't match

Cause: they count different things. One may report households and the other devices. One may include phones and computers in its impressions and the other TVs only. Quartile events may be fired by the device for one partner and by a server for the other, and their reporting windows may differ.

Fix: go back to the written definitions from step 2 and line the reports up column by column. Compare TV-only delivery from each partner, on the same dates and the same window. Use consistent creative IDs, so each spot is counted once. If the gap remains, judge both partners against the same outcome measure, such as lift in site visits.