Cross-Screen Attribution: Connecting CTV to Conversions

The cross-screen gap

A viewer watches your 30-second ad on their Samsung TV via Hulu at 8:47pm. The next morning at 10:15am, they search your brand on their iPhone and purchase on your website. From an attribution perspective, these are two unrelated events on two unrelated devices.

Bridging this gap -- connecting the TV impression to the phone conversion -- is cross-screen attribution. It's the fundamental technical challenge of CTV advertising, and getting it right determines whether CTV looks like a waste of money or a profitable channel in your reporting.

How cross-screen attribution works

At its core, cross-screen attribution requires matching an ad exposure on one device to a conversion on another. This matching happens at the household or individual level using one of four methods.

IP address matching

The most straightforward approach. Both the CTV device and the converting device are on the same home WiFi network, sharing an IP address.

The process:

1. CTV platform logs the impression with the device's IP address and timestamp.
2. Your server-side tracking captures the converter's IP address when they purchase.
3. If the IPs match within the attribution window (typically 7-14 days), the conversion is attributed to CTV.

Match rates: 45-60% of CTV-driven conversions can be captured via IP matching, based on published data from Innovid and DoubleVerify.

Failure modes: The converting user is on cellular data (different IP). The household IP changed between exposure and conversion (ISPs periodically reassign). Multiple households share an IP (apartment buildings, college campuses). The user is on a VPN.

Best for: Broad directional measurement. Good enough for campaign-level optimization but not precise enough for granular creative testing.

Device graph matching

Identity resolution companies maintain graphs that link devices to households and individuals.

The process:

1. The Trade Desk, LiveRamp, or TransUnion matches the CTV device ID to a household.
2. The household is linked to its members' mobile, tablet, and desktop device IDs.
3. When a household member converts on any linked device, the conversion is attributed to CTV.

Match rates: 55-75% of targetable households can be matched, with accuracy varying by data density. Urban, higher-income households tend to have higher match rates.

What makes graphs more accurate. Deterministic signals (email login on CTV app matched to email login on mobile app) are high-confidence matches. Probabilistic signals (devices frequently on the same IP, similar browsing patterns) are lower-confidence. The best graphs blend both.

Cost: Device graph matching typically adds $2-$5 CPM to your measurement costs. Some DSPs (The Trade Desk, DV360) include it in their platform fees.

Deterministic platform matching

Streaming platforms with cross-device login data can match impressions to conversions with near-perfect accuracy -- within their ecosystem.

Examples:

• Amazon: A viewer sees your ad on Fire TV. They purchase on Amazon.com the next day. Same Amazon account, deterministic match.
• Google/YouTube: A viewer sees your ad on YouTube CTV. They click a Google search ad on their phone. Same Google account, deterministic match.
• Roku: A viewer sees your ad on Roku. They install your app on their Roku device. Same Roku account.

Match rates: 90%+ accuracy for within-platform conversions. The limitation is that most conversions happen outside the platform's ecosystem.

Data clean room matching

The most privacy-compliant approach. Advertiser and publisher data meet in a neutral environment.

The process:

1. You upload your conversion data (hashed email addresses, conversion timestamps, values) to a clean room.
2. The CTV platform uploads exposure data (hashed identifiers, impression timestamps, creative IDs).
3. The clean room matches the datasets and returns aggregate statistics: match rate, attributed conversions, incremental lift.

Advantages: Privacy-compliant by design. No raw data leaves either party. Supports complex analysis (incrementality, frequency optimization) beyond simple attribution.

Limitations: Results are aggregated, not individual-level. Setup requires technical integration with each platform's clean room. Turnaround is typically 1-2 weeks, not real-time.

Implementation guide for media buyers

Here's how to set up cross-screen attribution for a CTV campaign:

Week 1: Server-side tracking. Ensure your website captures conversion events server-side, including the user's IP address, user agent, and any first-party identifiers (email, customer ID). This is the foundation for all cross-screen matching. Without server-side tracking, you're limited to platform-reported numbers.

Week 2: Exposure log integration. Request impression-level logs from each CTV platform or DSP. Key fields: device ID, household IP, timestamp, creative ID, publisher, placement. These logs are the CTV side of the matching equation.

Week 3: Matching setup. Configure your matching approach:

• Minimum viable: IP matching between exposure logs and conversion events.
• Recommended: IP matching plus device graph integration via your DSP or a dedicated partner.
• Advanced: Add clean room matching with your top 1-2 CTV platforms.

Week 4: Validation and calibration. Run a lift study to measure true incremental conversions. Compare the lift study results to your cross-screen attributed conversions. The ratio tells you your attribution's coverage rate. If the lift study shows 1,000 incremental conversions and your cross-screen attribution captured 600, your coverage rate is 60%. Apply a 1.67x calibration multiplier to future attributed numbers.

Attribution window selection

The attribution window -- how long after a CTV impression you'll credit a conversion -- significantly impacts reported performance.

Too short (1-3 days): Misses the majority of CTV-driven conversions. CTV is an awareness/consideration channel with longer paths to conversion. A 1-day window captures maybe 15% of true impact.

Optimal (7-14 days): Captures 75-85% of CTV-driven conversions while limiting false attribution. 14 days is the most common default, validated by studies from Roku and Samsung Ads showing that 80% of CTV-influenced conversions occur within 14 days.

Too long (30+ days): Captures more conversions but includes many that would have happened without the CTV exposure. At 30 days, the signal-to-noise ratio degrades significantly.

Recommendation: Start with 14 days. If your product has a longer purchase cycle (B2B, automotive, real estate), extend to 21 days. If your product is low-consideration (food delivery, subscription trials), shorten to 7 days.

Deduplication across screens and platforms

When the same user converts after seeing ads on CTV, Meta, and Google, all three platforms claim credit. Cross-screen attribution needs deduplication rules:

Rule 1: One conversion, one attribution. Each conversion event should be attributed once in your reporting, not claimed by every channel that touched the user.

Rule 2: Use a consistent model. Apply the same attribution model (data-driven, position-based, or time-decay) across all channels, including CTV. Don't use last-touch for digital and view-through for CTV -- it makes them incomparable.

Rule 3: CTV gets credit for the assist. In multi-touch models, CTV typically appears early in the path (first or second touchpoint). It should receive appropriate credit as an introducer, even if another channel gets the final-touch credit.

FAQ

What's the biggest mistake media buyers make with cross-screen attribution?

Relying on a single matching method. IP matching alone misses 40-55% of true CTV conversions. Layering IP matching with device graph matching and clean room analysis increases coverage to 70-85%. No single method captures everything -- the combination is what produces reliable numbers.

How do I explain cross-screen attribution accuracy to clients?

Be transparent: "Our cross-screen attribution captures approximately 60-70% of CTV-driven conversions based on our calibration tests. We apply a calibration multiplier to account for the gap. This gives us a reliable estimate, not a precise count. We validate with incrementality experiments quarterly to ensure accuracy."

Is cross-screen attribution improving or getting harder?

Improving, on balance. Data clean rooms, ACR integration, and better identity graphs are increasing match rates. However, privacy regulations and identifier deprecation create headwinds. The net trend is positive -- cross-screen attribution accuracy has improved roughly 10-15 percentage points since 2023.

---

Go Funnel uses server-side tracking and multi-touch attribution to show you which ads actually drive revenue. Book a call to see your real numbers.