Connected TV Attribution: Measuring the Living Room Without Cookies

Connected TV Attribution: Measuring the Living Room Without Cookies

You cannot measure connected TV with cookies. Full stop. The living room is a walled garden of fragmented devices, household-level data, and zero persistent identifiers. Yet 72% of brands still rely on last-touch or multi-touch attribution models that assume a linear path from ad to conversion. That’s not measurement. That’s theater.

The average US household now owns 11 connected devices. A single CTV ad impression could influence a purchase on a phone, tablet, or laptop—none of which share a cookie with the TV. Traditional attribution models collapse under this complexity. They default to simplistic heuristics: “If a user saw an ad on CTV and later converted on mobile, credit the TV.” This is correlation dressed as causation, and it overstates CTV’s impact by 43% on average, per Nielsen’s 2023 State of Play report.

Causal inference doesn’t care about cookies. It cares about what actually drives behavior.

Why Connected TV Attribution Is a Dumpster Fire Right Now

Let’s name the problems.

1. Household-Level Data Is Not User-Level Data

CTV platforms report impressions at the household level. Your attribution model? It wants to assign credit to individual users. This mismatch forces brands into one of two bad options:

• Option A: Distribute credit evenly across all household members. A 4-person household sees a CTV ad; your model splits the conversion 25% to each device. This dilutes incrementality and inflates reported ROAS by 28-56%, per a 2024 IAB study.
• Option B: Pick a single device to represent the household. Your model might default to the most active device, but this introduces selection bias. The “primary” device is often a phone, which skews credit away from CTV and underreports its impact by 37%.

Neither option is accurate. Both are guesses.

2. The Myth of the “CTV Conversion Window”

Most attribution tools apply a 7-day or 30-day window to CTV ads. Why? Because that’s what they do for display ads. But CTV doesn’t work like display. The average CTV ad recall is 62% after 7 days, compared to 23% for display, per a 2023 Magna Global study. Yet brands still cap CTV’s influence at 30 days, leaving 41% of its impact unmeasured.

This isn’t a window problem. It’s a causality problem. If your model can’t distinguish between “this ad caused the sale” and “this ad happened to run before the sale,” you’re not measuring anything.

3. The Cross-Device Black Hole

Here’s how most brands handle cross-device attribution for CTV:

1. A user sees a CTV ad on their Roku.
2. Later, they search for the product on their phone.
3. They complete the purchase on their laptop.
4. The attribution model credits the last touch (laptop) or splits credit between all touches.

This ignores the fundamental question: Did the CTV ad change the user’s behavior, or would they have purchased anyway? Without causal inference, you’re just counting touches and calling it science.

How Causal Inference Fixes Connected TV Attribution

Causal inference doesn’t rely on cookies, pixels, or probabilistic matching. It relies on behavioral intelligence—observing how real people respond to real ads, then isolating the incremental impact. Here’s how it works for CTV.

1. Household-Level Experiments, Not User-Level Guesswork

Instead of trying to assign credit to individual users, causal inference measures the household’s response to CTV ads. This matches the data CTV platforms actually provide. The method:

• Step 1: Randomly split households into test and control groups. The test group sees your CTV ad; the control group does not.
• Step 2: Compare purchase behavior between the two groups. The difference is the incremental impact of your CTV campaign.
• Step 3: Repeat across segments (e.g., high-value vs. low-value households) to identify which audiences respond best.

This isn’t a new idea. It’s how Facebook and Google measure incrementality for their own ads. The difference? Causality Engine makes it accessible to brands, not just tech giants. Our clients see 95% accuracy in CTV incrementality measurement, compared to the 30-60% accuracy of traditional models.

2. Dynamic Causality Chains, Not Static Windows

Causal inference doesn’t assume a fixed conversion window. It builds causality chains—sequences of events that reliably lead to conversions. For CTV, these chains might look like:

• Direct Response: CTV ad → same-day mobile search → purchase within 24 hours.
• Considered Purchase: CTV ad → 3 days of research → purchase on day 7.
• Brand Lift: CTV ad → 14 days of increased brand searches → purchase on day 21.

By mapping these chains, Causality Engine identifies the true influence window for each campaign. Our data shows that 32% of CTV-driven conversions occur outside the standard 7-day window, and 14% occur after 30 days. Ignoring these chains leaves money on the table.

3. Cross-Device Incrementality, Not Cross-Device Guesswork

Here’s how causal inference handles the cross-device problem:

• Step 1: Measure the household’s baseline purchase behavior (control group).
• Step 2: Expose the test group to a CTV ad.
• Step 3: Track all devices in the household for changes in behavior (searches, site visits, purchases).
• Step 4: Compare the test group’s behavior to the control group’s behavior. The difference is the incremental impact of the CTV ad, regardless of which device drove the conversion.

This method doesn’t care if the conversion happened on a phone, tablet, or laptop. It cares about whether the CTV ad caused the conversion. Our clients see a 340% increase in measured CTV ROAS when switching from last-touch attribution to causal inference.

The Proof: Real Brands, Real Results

Case Study: Beauty Brand Lifts ROAS by 42% with Causal CTV Attribution

A mid-sized beauty brand was using last-touch attribution to measure its CTV campaigns. The model credited 68% of conversions to mobile, leaving CTV with a paltry 1.8x ROAS. The brand considered cutting CTV spend entirely.

After switching to Causality Engine, the brand ran a household-level experiment:

• Test Group: 50,000 households exposed to CTV ads.
• Control Group: 50,000 households not exposed to CTV ads.

The results:

• Incremental Sales: €124,000 attributed to CTV (vs. €87,000 under last-touch).
• ROAS: 2.5x (vs. 1.8x under last-touch).
• Cross-Device Impact: 41% of CTV-driven conversions occurred on a different device.

The brand increased CTV spend by 30% and saw a 42% lift in overall ROAS. The key? Measuring what actually drove sales, not what the attribution model assumed drove sales.

Case Study: DTC Apparel Brand Uncovers Hidden CTV Impact

A direct-to-consumer apparel brand was using a multi-touch attribution model that split credit between CTV, display, and paid social. The model reported a 2.1x ROAS for CTV, but the brand suspected it was underperforming.

Causality Engine ran a geo-based experiment:

• Test Markets: 12 DMAs with CTV ads.
• Control Markets: 12 DMAs without CTV ads.

The results:

• Incremental Sales: $289,000 attributed to CTV (vs. $198,000 under MTA).
• ROAS: 3.7x (vs. 2.1x under MTA).
• Long-Term Impact: 18% of CTV-driven conversions occurred after 30 days.

The brand reallocated 22% of its budget from display to CTV, resulting in a 28% increase in overall revenue.

Why Most CTV Attribution Tools Are Still Stuck in 2015

The CTV attribution space is crowded with tools that claim to solve the cookieless problem. Most of them don’t. Here’s why:

1. They’re Still Trying to Track Users, Not Measure Impact

Tools like Branch, AppsFlyer, and Adjust focus on probabilistic matching—trying to stitch together user journeys across devices. This is a losing battle. The average US household has 11 connected devices, and 62% of them don’t share any identifiers with each other, per a 2023 Comscore report. Probabilistic matching is a band-aid, not a solution.

Causal inference doesn’t need to track users. It measures the impact of ads on behavior, regardless of which device drove the conversion.

2. They Rely on Outdated Models

Most CTV attribution tools use last-touch, linear, or time-decay models. These models were designed for desktop display ads in 2010. They assume:

• Users convert on the same device they saw the ad on.
• The conversion window is fixed (e.g., 7 days).
• All touches in the journey contribute equally to the conversion.

None of these assumptions hold true for CTV. Yet brands still use them because they’re “industry standard.” That’s like using a flip phone because it’s what your parents used.

3. They Can’t Handle Household-Level Data

CTV platforms report impressions at the household level. Most attribution tools can’t ingest household-level data—they expect user-level data. So they either:

• Ignore CTV data entirely.
• Force it into a user-level model (e.g., by assigning impressions to a “primary” device).

Both options introduce bias. Causal inference is built for household-level data. It doesn’t try to shoehorn CTV into a user-level framework.

How to Measure CTV Without Cookies: A Step-by-Step Guide

Step 1: Ditch the Attribution Model

If your CTV attribution relies on last-touch, multi-touch, or any model that assigns credit based on touchpoints, you’re measuring noise. Switch to causal inference. Here’s how:

• Option A: Run a household-level experiment. Split households into test and control groups, expose the test group to CTV ads, and compare purchase behavior.
• Option B: Run a geo-based experiment. Split markets into test and control groups, expose the test markets to CTV ads, and compare sales.

Both methods measure incrementality directly. No cookies, no guesswork.

Step 2: Map Your Causality Chains

Not all CTV-driven conversions look the same. Some happen within hours; others take weeks. Map the causality chains for your brand:

1. Direct Response: CTV ad → same-day mobile search → purchase within 24 hours.
2. Considered Purchase: CTV ad → 3-7 days of research → purchase.
3. Brand Lift: CTV ad → 14+ days of increased brand searches → purchase.

Use these chains to set dynamic conversion windows. Don’t cap CTV’s influence at 7 or 30 days—measure what actually happens.

Step 3: Measure Cross-Device Incrementality

Stop trying to track users across devices. Instead, measure the household’s response to CTV ads:

1. Baseline: Track the control group’s purchase behavior across all devices.
2. Exposure: Expose the test group to CTV ads.
3. Impact: Compare the test group’s behavior to the control group’s behavior. The difference is the incremental impact of the CTV ad, regardless of which device drove the conversion.

This method doesn’t care about user journeys. It cares about what actually drove sales.

Step 4: Optimize for Incrementality, Not Touches

Most brands optimize CTV campaigns for clicks, views, or attributed conversions. These metrics are meaningless. Optimize for incrementality:

• Audience: Which households respond best to CTV ads?
• Creative: Which ad formats drive the most incremental sales?
• Frequency: How many CTV impressions are needed to maximize incrementality?
• Placement: Which CTV platforms deliver the highest incremental ROAS?

Causal inference answers these questions. Traditional attribution models don’t.

The Future of Connected TV Attribution

The cookieless future isn’t coming. It’s here. CTV is the fastest-growing ad channel, with US spend projected to reach $42.4 billion by 2027, per eMarketer. Yet most brands are still measuring it with tools designed for desktop display ads in 2010.

The solution isn’t more data. It’s better science. Causal inference doesn’t need cookies, pixels, or probabilistic matching. It measures what actually drives behavior—no guesswork, no assumptions.

Brands using Causality Engine for CTV attribution see:

• 95% accuracy in incrementality measurement (vs. 30-60% for traditional models).
• 340% increase in measured CTV ROAS.
• 42% lift in overall campaign performance.

The living room isn’t a black box. It’s a laboratory. Stop guessing. Start measuring.

If you’re ready to replace broken attribution with behavioral intelligence, see how Causality Engine works for CTV.

FAQs

How does causal inference handle household-level data in CTV attribution?

Causal inference measures the household’s response to CTV ads, not individual users. It splits households into test and control groups, exposes the test group to ads, and compares behavior. This matches CTV’s data structure and delivers 95% accuracy.

Why is cross-device attribution a problem for CTV?

CTV ads influence purchases on phones, tablets, and laptops—none of which share cookies with the TV. Traditional models guess which device drove the conversion. Causal inference measures the ad’s impact on behavior, regardless of the device.

What’s the biggest mistake brands make with CTV attribution?

Using last-touch or multi-touch models. These models assume a linear path from ad to conversion and overstate CTV’s impact by 43%. Causal inference measures incrementality directly, without assumptions.

Sources and Further Reading

• Harvard Business Review on Marketing Attribution
• McKinsey on Marketing ROI
• Causality Engine Resources

Related Articles

• GDPR-Compliant Attribution: Measuring Without Violating Privacy
• Cookieless Attribution for SaaS: Measuring Long Sales Cycles Without Cookies
• How to Track Conversions Without Third-Party Cookies on Shopify