Project Horizon: AI Boosts ROAS 15-25% in 2026

The persistent challenge in digital marketing remains understanding the full customer journey across disparate devices. Cross-device tracking, powered by artificial intelligence, offers a solution to stitch these fragmented interactions into a unified narrative, enabling more precise targeting and attribution. But does AI truly deliver on its promise of a cohesive digital experience, or is it another layer of complexity?

Campaign Teardown: “Project Horizon” for a B2C Retailer

We recently executed “Project Horizon,” a six-month campaign for a mid-sized outdoor gear retailer, “Trailblazer Outfitters.” The primary goal was to increase online sales by 15% and improve the efficiency of ad spend by accurately attributing conversions across the customer’s multi-device path. Our hypothesis: AI-driven cross-device tracking would reveal hidden conversion paths and allow for more effective budget allocation. This wasn’t just about showing ads. It was about understanding the user’s journey from a mobile search on the morning commute to a desktop purchase later that evening.

Budget: $120,000 over six months ($20,000 per month). This included platform fees for a third-party identity resolution provider, ad spend across Google Ads and Meta Business Suite, and internal data science resources for AI model training.

Duration: January 1, 2026, to June 30, 2026.

Target Audience: Outdoor enthusiasts, aged 25-55, with demonstrated interests in hiking, camping, and adventure sports. We segmented further by geography, focusing on urban centers near national parks like Denver, Colorado, and Seattle, Washington.

Strategy: Blending Deterministic and Probabilistic Matching

Our strategy relied on a hybrid approach to identity resolution. For deterministic matching, we integrated customer login data from Trailblazer Outfitters’ website and app with hashed email addresses used for newsletter subscriptions. This provided a high-confidence link between devices for users who provided explicit consent. For instance, a customer logging into their account on a tablet and then purchasing on a laptop would be identified as the same individual with near 100% certainty.

The more complex, and frankly, more interesting part, involved probabilistic matching, where AI played a central role. We used a platform that ingested anonymized data points: IP addresses, device types, browser fingerprints, Wi-Fi networks, and even behavioral patterns like browsing speed and scroll depth. The AI model then analyzed these signals to infer connections between devices belonging to the same user. For example, if a user consistently visited Trailblazer Outfitters’ website from the same IP address on a mobile phone during the day and a desktop computer in the evening, the AI would assign a high probability that these were the same person.

We established a confidence threshold of 85% for probabilistic matches to be included in our unified user profiles. Anything below that, and the data was treated as separate user sessions. Too low, and we risked misattributing conversions. Too high, and we’d lose valuable insights into less obvious cross-device behaviors.

Creative Approach: Dynamic Messaging Across Stages

The campaign’s creative strategy was designed to adapt based on the user’s identified stage in the purchase journey. Early-stage users, identified through initial mobile searches for “hiking boots reviews,” received informational content: blog posts on “Choosing the Right Hiking Boot” or YouTube videos demonstrating product features. These were primarily served via Google Discovery Ads and Meta’s Audience Network.

Mid-stage users, who had visited product pages but not added items to their cart, saw retargeting ads featuring specific products they viewed, often with a subtle call to action like “Complete Your Kit.” These were served on desktop through Google Search Ads and display networks. For instance, if someone viewed a specific backpack model on their phone, they’d see an ad for that same backpack, perhaps paired with a complementary item like a water bladder, when they later browsed news sites on their laptop.

Late-stage users, those who abandoned carts or showed high intent signals (e.g., spending significant time on checkout pages), received more direct offers, such as free shipping or a small discount, delivered via email (if collected) and highly targeted display ads. The creative also varied by device. Mobile ads were punchy, with clear product images and minimal text, while desktop ads allowed for more detailed descriptions and multiple product views. This contextual relevance, driven by understanding the user’s device and journey stage, was a core tenet.

Targeting and Attribution Refinements

The AI-powered identity resolution allowed us to move beyond last-click attribution, which we know is often misleading. Instead, we implemented a data-driven attribution model within Google Ads and a custom attribution model in our analytics platform that factored in all touchpoints across devices. This meant giving partial credit to the initial mobile ad that introduced the product, the desktop display ad that reminded the user, and the final search ad that closed the sale.

Our targeting also became significantly more granular. Instead of simply targeting “outdoor enthusiasts,” we could now target “outdoor enthusiasts who viewed specific tent models on mobile and then researched them on desktop.” This reduced wasted impressions and focused our budget on users with higher propensity to convert. We also created exclusion lists of users who had recently purchased, preventing repetitive and irritating ads.

What Worked: Uncovering Hidden Journeys

The campaign revealed fascinating insights. We found that 80% of conversions involved at least two devices, a much higher figure than our previous last-click models indicated. Mobile was overwhelmingly the starting point (65% of initial touchpoints), but desktop accounted for 70% of final conversions. This underscored the importance of a smooth experience from initial discovery on a small screen to detailed consideration and purchase on a larger one.

Our return on ad spend (ROAS) improved by 18% compared to the previous quarter, reaching an average of 3.5:1. This was largely due to the more accurate attribution, which allowed us to reallocate budget from less effective channels to those truly contributing to conversions earlier in the journey. For example, we reduced spend on generic desktop display ads by 15% and increased budget for mobile search ads by 20%, seeing a direct correlation with increased top-of-funnel engagement.

The cost per conversion (CPC) decreased by 12%, dropping from an average of $25 to $22. Our ad impressions remained relatively stable at 15 million over the six months, but our click-through rate (CTR) saw a modest but significant increase of 0.5%, from 1.8% to 2.3%. This indicates that the more relevant ads, served at the right time on the right device, were resonating better with the audience.

One specific example: we identified a segment of users who would browse backpack reviews on their commute using public Wi-Fi, then later search for specific models on their home desktops. Without cross-device tracking, these would appear as two distinct users. With it, we could serve them a “compare models” ad on their desktop, which proved highly effective, leading to a 30% higher conversion rate for that specific ad set.

What Didn’t Work: The Privacy Balancing Act

The biggest hurdle, and frankly, an ongoing challenge, was working through data privacy regulations. While we strictly adhered to consent requirements, particularly under GDPR and CCPA, a segment of users opted out of tracking. This created “blind spots” in our data, meaning we couldn’t create a truly unified journey for everyone. Approximately 15% of website visitors opted out of non-essential cookies, impacting the completeness of our probabilistic matching.

Another issue was the reliance on third-party cookies, which are gradually being phased out by major browsers. While our identity resolution partner was already shifting towards cookieless identifiers (e.g., first-party data, contextual signals, and advanced fingerprinting techniques), the transition wasn’t entirely smooth. We saw a slight dip in match rates during browser updates that tightened privacy controls, requiring constant adjustments to our AI models.

The initial setup and integration with Trailblazer Outfitters’ existing CRM and analytics systems were also more complex and time-consuming than anticipated. It took almost a full month to ensure data flowed correctly and was properly anonymized and ingested by the identity resolution platform. This upfront investment in integration is often underestimated.

Optimization Steps Taken: Adapting to the Field

To address the privacy concerns and future-proof our strategy, we took several key steps. We invested in a more strong first-party data strategy, actively encouraging users to create accounts and subscribe to newsletters, offering clear value in return. This increased our deterministic match rates by 5% over the campaign’s duration.

We also implemented a more dynamic consent management platform, allowing users finer control over their data preferences, which paradoxically, increased trust and led to more users opting into some forms of personalized advertising. A transparent privacy policy, clearly explaining how data was used for cross-device identification, was also important.

Plus, we continuously retrained our AI models with new data to adapt to evolving browser privacy features and user behaviors. We focused on strengthening probabilistic matching using contextual signals and anonymized behavioral patterns, rather than relying solely on device IDs that may become obsolete. Our data science team, for example, developed a model that could infer device connections based on the sequence and timing of page views, even without a shared identifier. This involved processing billions of data points daily to identify subtle patterns.

The integration challenges were mitigated by hiring a dedicated data integration specialist for a three-month contract, ensuring smoother data pipelines. We also established weekly review meetings with the identity resolution vendor to address any data discrepancies or integration issues promptly.

In the evolving digital marketing field, AI-driven cross-device tracking is no longer an optional enhancement. It is a fundamental requirement for understanding and engaging with modern consumers. The insights gained from Project Horizon demonstrate that while challenges like privacy and integration persist, the ability to unify customer journeys leads to significantly more effective and efficient campaigns. Marketers must prioritize strong first-party data strategies and continuously adapt their AI models to stay ahead of privacy changes, ensuring they can connect with their audience wherever they are.