CTV as a Performance Traffic Source: Testing Results for a FinTech Application

Connected TV (CTV) – a video advertising format on Smart TV and streaming platforms – was actively discussed as a promising traffic source with high audience reach.

The challenge was to determine whether CTV could work not only for brand awareness but also as a performance channel with measurable install metrics.

CTV Format Specifics

Connected TV is an ecosystem of Smart TVs, gaming consoles, and streaming devices (Roku, Apple TV, Amazon Fire Stick) where users consume video content via the Internet.

Unlike traditional TV, CTV as a channel enables programmatic buying with targeting by demographics, interests, and behavior, making the format attractive to digital marketers.

Historically, CTV was used by brands for awareness campaigns with reach and frequency metrics, but the emergence of probabilistic attribution methods opened the possibility to measure performance metrics.

The client required an evaluation of CTV specifically as a performance channel. The existing media mix included Facebook, Google UAC, TikTok, and DSP platforms that ensured predictable CPI and ROAS.

The question was whether CTV could complement this toolkit without cannibalizing organic traffic and with comparable unit economics.

View-Through Attribution Mechanics in CTV

View-through attribution (VTA) in CTV campaigns is based on an IP address matching model that differs from deterministic click-based attribution in mobile advertising.

When a user sees an ad on TV and installs an app from a mobile device within the attribution window, the MMP matches the CTV device IP address with the mobile device IP address.

The install is attributed to the CTV campaign if the IPs match and the time constraints of the attribution window are met. The attribution formula can be described as follows:

CTV Install = IP Match (TV device = Mobile device) × Time Constraint (install within window) × No Click Attribution

where:

• IP Match – IP address alignment between devices
• Time Constraint – time interval after impression (typically 1-24 hours)
• No click attribution – absence of click-based attribution from other sources (click takes priority over view in attribution partner logic).

For statistically significant traffic volumes, the method provides accurate assessment of incremental impact. VTA accuracy depends on household size and VPN service usage, creating a 10-15% margin of error for typical campaigns.

The critical parameter is the attribution window choice – too long a window leads to organic cannibalization, too short underestimates the effect.

Test Planning: Key Decisions

As we approached CTV testing, we formulated questions that would define the experiment architecture:

1. What metrics should be considered successful for a CTV campaign compared to primary channels (target CPI, D7 ROAS)?
2. Which media source will provide transparent attribution and protection against manipulation?
3. How does the client's attribution partner process VTA for CTV and what limitations exist?
4. How should we configure windows to avoid cannibalizing organic traffic?
5. What minimum budget is required for a statistically significant test?

The first step was selecting a CTV platform from accredited media sources. We evaluated purchasing options: programmatic DSPs and CTV-specialized platforms.

Considering the budget constraints of the test and technical integration requirements, we chose a CTV platform with a managed service model and ready-made integration.

Limitations

Integration revealed critical technical limitations affecting test design and results interpretation. The first problem concerned protection against manipulation: despite MMP assurances of reliable protection, we discovered a vulnerability in tracking link parameters.

When necessary parameters added, any traffic source could imitate a CTV campaign, affecting all active media sources.

Source verification from the MMP (matching mediasource_id with whitelist) did not prevent this manipulation, creating risk for the client whose media mix included CPA partners: incorrect attribution would lead to overpayment for manipulated traffic.

We implemented additional controls through parameter analysis in tracking links. This fraud protection layer became essential for maintaining attribution integrity across the entire media mix.

The second limitation concerned reporting: CTV campaign ad impressions did not display in the standard dashboard, with data residing only in raw exports.

Working together with the client's data engineers, we built a custom dashboard in Tableau. This complicated daily monitoring but ensured full campaign visibility.

Creative Configuration and Tracking Links

The creative strategy for CTV required adapting existing performance videos to the format's specifics. We took a horizontal 15-second video with the best metrics and adapted it for 16:9 format with the addition of a mandatory static end card.

The endcard – a static banner with QR code and call-to-action – served a dual function: it increased engagement and created an additional cohort of click attribution data.

The QR code was generated through a separate link, allowing us to track additional activity parallel to impressions.

The calculation formula we used accounted for both sources:

Total CTV Installs = VTA installs + QR installs

where deduplication occurred automatically through click attribution prioritization. This gave a more complete picture of creative effectiveness: if there were many QR scans, the creative generated immediate intent; if VTA predominated, the delayed conversion effect was at work. We found that the click-to-view attribution ratio became an additional indicator of creative approach quality for the CTV format.

Attribution Window Selection: Balancing Accuracy and Cannibalization

Window configuration became a subject of negotiations between us and the CTV platform. The media source insisted on a standard 24-hour window for VTA, arguing this reflected delayed conversion behavior among fintech audiences: a user sees the ad in the evening and installs the app the next day. We objected based on organic traffic pattern analysis.

The data showed that 40% of organic installs occurred in the same time windows as potential CTV views (evenings, weekends). Setting a 24-hour window would attribute a significant portion of organic traffic to the CTV campaign, artificially inflating its effectiveness.

After we conducted A/B testing with 1h, 2h, 6h, and 24h windows, we settled on a 2-hour window as the optimal balance.

Formula for estimating Organic cannibalization:

Cannibalization Rate = (Organic installs in CTV flight period – Organic baseline) / Total CTV attributed installs

The analysis showed: with a 24h window, the share of organic traffic cannibalization exceeded 30%; with a 2h window, it decreased to an acceptable 5-10%. The reduction decreased install volumes by 60% but provided a cleaner assessment of incremental effect from the CTV campaign.

Testing Results

Several incidents occurred during the test: at one point, impressions began displaying in the standard MMP report, contradicting the platform's technical limitations.

The media source contacted the MMP to determine the cause, but details of what happened remained undisclosed.

CTV-campaign metrics significantly underperformed primary channels:

• Conversion to registration event was 35% below the benchmark of existing channels
• D7 ROAS did not reach the minimum profitability threshold for the fintech vertical
• CPM was 4 times higher than typical indicators
• CPI exceeded client target metrics by 60-70%

Such results made the channel ineffective from the client's unit economics perspective. The QR code generated only 3% of all installs, with the main volume coming through view attribution.

Low CTA share indicated that CTV operated primarily as an awareness channel without generating immediate install intent.

Systematic Approach to Validating New Channels

Understanding the mechanics of each traffic source and identifying structural limitations is a key skill in performance marketing.

By influencing individual system elements (attribution windows, creative format, targeting parameters), we impact the overall effectiveness of the entire channel.

CTV as a performance tool did not meet expectations in the current configuration, but it can work for brand awareness tasks with reach and ad recall metrics.

What proved critically important was not the test result itself but the methodology: strict isolation from organic traffic, manipulation control through MMP, and report building.

What proved critically important was not the test result itself but the methodology: strict isolation from organic traffic, manipulation control through MMP, and report building.

These principles apply to validating any new channel – from influencer marketing to podcast advertising.

The main thing is to formulate the right questions before launching the test, rather than searching for explanations of unexpected results after the fact.