# Targeting and measuring ads across a fragmenting screen
A household in Ohio streams a cooking show on a connected TV. Mid-episode, a 30-second ad for a mid-size SUV plays. That single impressionimpressionThe total number of times an ad or piece of content is displayed, regardless of clicks. Each display counts as one impression, even to the same person.Voir la définition complète → traveled through an audience segment, a real-time auction, and a measurement pipelinepipelineAll active sales opportunities across the stages of the sales process, together with their combined potential value and probability of closing.Voir la définition complète → before it reached the screen. Following it end to end reveals how modern TV advertising actually works, and where the numbers quietly mislead the people spending the money.
For decades, TV advertising was simple to buy and simple to measure. You bought a slot on a network, and a panel-based rating told you roughly how many people watched.
That world broke apart. Viewers now split their time across linear TV (traditional broadcast and cable), streaming apps, and short-form video on phones. The single biggest shift is CTV, or connected TV: any television that streams video over the internet, whether through a smart TV, a streaming stick, or a game console.
CTV matters because it is addressable. Unlike a broadcast ad that everyone in a market sees, a CTV ad can be targeted to a specific household. That precision is the whole reason marketers moved budget here, and it is also the source of new measurement headaches.
Our SUV impressionimpressionThe total number of times an ad or piece of content is displayed, regardless of clicks. Each display counts as one impression, even to the same person.Voir la définition complète → starts with a segment: a defined group of households an advertiser wants to reachreachThe number of unique people exposed to your message in a given period. Unlike impressions, reach counts each person once, no matter how often they see it.Voir la définition complète →. The auto brand might target "households likely in market for a new vehicle, with children, in specific metro areas."
How does a platform know which households fit? A few common sources:
The advertiser rarely sees individual names. SegmentsSegmentsDividing a market into distinct groups of customers who share similar needs, characteristics or behaviours, so each group can be served with a tailored approach.Voir la définition complète → are built around households or hashed identifiers, not people by name. Privacy regulation like the California Consumer Privacy Act shapes what data can be collected and how consumers opt out.
When our Ohio household opens the cooking show, the streaming app signals that an ad slot is available. This triggers programmatic buying: automated, auction-based ad transactions that happen in milliseconds.
The app sends a bid request describing the available impressionimpressionThe total number of times an ad or piece of content is displayed, regardless of clicks. Each display counts as one impression, even to the same person.Voir la définition complète → (roughly: content genre, device type, and whether this household matches known segmentssegmentsDividing a market into distinct groups of customers who share similar needs, characteristics or behaviours, so each group can be served with a tailored approach.Voir la définition complète →). Advertisers' systems evaluate it and bid. The auto brand's platform recognizes this household as "in-market SUV shopper" and bids accordingly. Highest relevant bid wins, and the ad renders.
This all happens before the next second of the cooking show plays. To go deeper on the plumbing, the IAB Tech Lab maintains the open standards that most of this runs on: see the OpenRTB specification.
Now the measurement problem begins. The advertiser does not just want impressionsimpressionsThe total number of times an ad or piece of content is displayed, regardless of clicks. Each display counts as one impression, even to the same person.Voir la définition complète →. It wants to know two things:
These are the core currency of a campaign. A campaign can serve a million impressionsimpressionsThe total number of times an ad or piece of content is displayed, regardless of clicks. Each display counts as one impression, even to the same person.Voir la définition complète → but reachreachThe number of unique people exposed to your message in a given period. Unlike impressions, reach counts each person once, no matter how often they see it.Voir la définition complète → only 200,000 households if each one saw the ad five times. That may be intentional or wasteful, depending on the goal.
Here is the hard part. The same household watches the SUV ad on a smart TV in the living room, a tablet in the kitchen, and a phone on the commute. Three devices, one household. Without connecting them, the platform counts three "reached" households and undercounts frequency badly.
This is the deduplication problem: matching impressionsimpressionsThe total number of times an ad or piece of content is displayed, regardless of clicks. Each display counts as one impression, even to the same person.Voir la définition complète → across devices and apps back to the same household. Solving it requires an identity graph, a system that links device identifiers to a common household or user ID.
If two streaming apps each report reaching a household, and neither talks to the other, the advertiser thinks it reached two households. Multiply that across dozens of apps and the reported reachreachThe number of unique people exposed to your message in a given period. Unlike impressions, reach counts each person once, no matter how often they see it.Voir la définition complète → balloons while true reachreachThe number of unique people exposed to your message in a given period. Unlike impressions, reach counts each person once, no matter how often they see it.Voir la définition complète → stays flat.
The practical consequence: a media buyer thinks a campaign is broadly reaching new audiences when it is actually hitting the same households again and again. Independent measurement firms and industry bodies like the Media Rating Council exist partly to audit and accredit these numbers so buyers can trust them.
Here is a simplified view of what deduplication does to the same raw data:
Raw impression log:
household_A smart_tv 10:02
household_A tablet 19:30 (same household, different device)
household_B phone 21:15
Naive count: 3 impressions, 3 "reached" devices
Deduplicated: 3 impressions, 2 households reached
household_A frequency = 2The raw log is not wrong. The interpretation is. Everything depends on whether the identity graph correctly ties those two household_A rows together.
Our household saw the SUV ad on Tuesday. Three weeks later, someone in that household buys the SUV. Did the ad cause the sale? AttributionAttributionA framework for assigning credit to the touchpoints that contributed to a conversion, so you can measure which channels and interactions actually drive results.Voir la définition complète → is the practice of assigning credit for a conversion (a purchase, a sign-up, a dealership visit) to the marketing touches that preceded it.
The most common and most dangerous model is last-touch attribution: give 100 percent of the credit to the final ad or click before the conversion.
Last-touch is popular because it is simple and easy to implement. It is also frequently wrong, for a specific reason.
Imagine the real buyer journeybuyer journeyThe full sequence of touchpoints a customer has with your brand before, during and after purchase, spanning awareness, consideration, decision, retention and advocacy.Voir la définition complète →:
1. A CTV ad first makes the household aware of the SUV (upper funnelfunnelThe customer journey from awareness to purchase, typically Awareness, Interest, Consideration, Decision, Action, with prospects narrowing at each stage.Voir la définition complète →).
2. A few social ads keep it top of mind.
3. The buyer searches the brand by name and clicks a search ad.
4. Purchase.
Last-touch hands all the credit to the branded search click. But that click only happened because the CTV ad created the interest. Search "caught" a demand that CTV "created."
The buyer looking at last-touch data concludes CTV does not work and search does, then shifts budget out of the very channel that started the journey. Over time, demand dries up because nothing is filling the top of the funnelfunnelThe customer journey from awareness to purchase, typically Awareness, Interest, Consideration, Decision, Action, with prospects narrowing at each stage.Voir la définition complète →. This is one of the most common and expensive mistakes in media measurement.
Better approaches include multi-touch attributionmulti-touch attributionA method that distributes conversion credit across all marketing touchpoints in the customer journey, rather than crediting only the first or last interaction.Voir la définition complète → (distributing credit across touches) and incrementality testing: holding out a control group that sees no ad, then measuring the lift in conversions among those who did. Incrementality answers the real question: what happened *because* of the ad that would not have happened anyway?
🎬 [VIDEO: "How Attribution ModelingAttribution ModelingAttribution modeling is the method of assigning credit for a conversion across the marketing touchpoints a customer interacted with before buying or signing up.Voir la définition complète → Works" — youtube.com — a clear walkthrough of last-touch versus multi-touch and where each breaks down]
A media buyer running our SUV campaign should read the numbers in a specific order:
A recurring trap: judging an upper-funnelfunnelThe customer journey from awareness to purchase, typically Awareness, Interest, Consideration, Decision, Action, with prospects narrowing at each stage.Voir la définition complète → channel like CTV by lower-funnelfunnelThe customer journey from awareness to purchase, typically Awareness, Interest, Consideration, Decision, Action, with prospects narrowing at each stage.Voir la définition complète →, last-touch metrics. CTV rarely earns the final click. Held to a click-based scorecard, it always looks weak, even when it is doing its job.
Every number here rests on data qualitydata qualityThe degree to which data is fit for purpose: accurate, complete, consistent, timely, valid and unique. Poor quality data undermines analytics, reporting and AI.Voir la définition complète → that varies. ACR coverage depends on which TVs are in a market and whether households consented. Identity graphs are estimates, not perfect truth. Panel-based measurement covers a sample, not everyone. Treat vendor reachreachThe number of unique people exposed to your message in a given period. Unlike impressions, reach counts each person once, no matter how often they see it.Voir la définition complète → figures as informed estimates and ask how deduplication was done before trusting a comparison across platforms.