AI Marketing: 18% ROAS Boost in 2025 App Launch

Understanding the true impact of each touchpoint in a user’s journey to conversion is a persistent challenge in app marketing. Traditional attribution models often fall short, miscrediting early interactions or overemphasizing the last click, leading to misallocated budgets and suboptimal campaign performance. This teardown examines a recent campaign where an AI-driven multi-touch attribution model revealed critical insights, fundamentally shifting our understanding of user acquisition dynamics.

Campaign Overview: “Quest for Echoes” Mobile RPG Launch

Our client, a mid-sized gaming studio, launched “Quest for Echoes,” a new fantasy mobile RPG, in Q3 2025. The studio had a strong reputation for engaging gameplay but struggled with efficient user acquisition in a crowded market. Their previous campaigns relied heavily on last-click attribution, leading to a focus on direct response channels. For “Quest for Echoes,” we proposed integrating an AI multi-touch attribution model to gain a more well-rounded view of the customer journey, from initial exposure to final install and in-app purchase.

The campaign ran for 12 weeks, from September to November 2025. The total budget allocated was $1.5 million, with a target Cost Per Install (CPI) of $3.50 and a 90-day Return On Ad Spend (ROAS) of 1.2x. The primary objective was to acquire 400,000 high-quality installs (users with at least one in-app purchase within 90 days) while maintaining ROAS targets. Secondary objectives included increasing brand awareness and fostering community engagement.

Initial Strategy and Channel Mix

Based on historical data and last-click models, the initial budget allocation prioritized channels known for direct conversions:

• Paid Social (Meta, TikTok): 40% ($600,000) focused on in-feed video ads and playable ads.
• Search Ads (Google App Campaigns, Apple Search Ads): 30% ($450,000) targeting high-intent keywords.
• Ad Networks (Unity Ads, AppLovin): 20% ($300,000) for interstitial and rewarded video ads within other gaming apps.
• Content Marketing & Influencer Partnerships: 10% ($150,000) for YouTube reviews, Twitch streams, and gaming blog features. This was primarily considered a brand-building exercise with little direct attribution expected.

Creative assets included short gameplay trailers, character shows, and user-generated content (UGC) style ads for social platforms. Targeting was broad initially, refined weekly based on early performance data.

The AI Multi-Touch Model: Unveiling Hidden Value

Our chosen AI multi-touch attribution solution, Branch, ingested data from all advertising platforms, the client’s CRM, and in-app analytics. Unlike traditional models like linear, time decay, or position-based, this AI model used machine learning to assign fractional credit to each touchpoint based on its actual contribution to the conversion path. It analyzed sequences, time between touches, and the type of interaction to determine influence scores. This allowed for a nuanced understanding of which interactions truly moved a user down the funnel.

Phase 1: Initial Performance (Weeks 1-4)

In the first month, the campaign performed largely as expected under the traditional last-click lens. We saw:

• Total Impressions: 150 million
• Click-Through Rate (CTR): 1.8% average across all channels.
• Installs: 110,000
• Average CPI (Last-Click): $4.09 (above target)
• 90-Day ROAS (Last-Click Projection): 0.9x (below target)

Paid Social and Ad Networks delivered the highest volume of installs at the lowest last-click CPIs. Search Ads showed higher quality users but at a higher CPI. Content marketing and influencer campaigns, as anticipated, showed very few direct installs attributed, primarily generating impressions and engagements.

Table 1: Initial Campaign Performance (Weeks 1-4) based on Last-Click Attribution

Phase 2: AI Insights and Budget Reallocation (Weeks 5-8)

The AI model began to reveal a different story. It identified numerous conversion paths where content marketing and influencer engagements were early, critical touchpoints, even if they weren’t the final click. For example, a user might watch a YouTube review from a gaming influencer, then later see a paid social ad, click it, and install. The last-click model would attribute 100% to paid social. The AI model, however, might assign 30% credit to the YouTube video, 60% to the paid social ad, and 10% to a previous organic search.

Specifically, the AI model highlighted:

• Influencer content had a 25% influence score on conversion paths for users who eventually made an in-app purchase. This was a stark contrast to its near-zero last-click attribution.
• Short-form video ads on Meta and TikTok, particularly those featuring authentic user reactions, consistently appeared as strong early-stage touchpoints, driving awareness and consideration. Their influence score was 1.7x higher than static banner ads for initial engagement.
• Search Ads, while still valuable for high-intent users, were often preceded by exposure to display or social ads, indicating they acted more as a mid-funnel validation point rather than a pure discovery channel in many cases.

Based on these insights, we made a significant budget reallocation:

• Paid Social: Reduced to 30% ($450,000 remaining budget), with a shift in creative focus towards more interactive, early-stage engagement formats.
• Search Ads: Maintained at 30% ($450,000 remaining budget), but with refined keyword targeting to capture both discovery and validation intent.
• Ad Networks: Reduced to 15% ($225,000 remaining budget), as the AI model showed diminishing returns and less influence on high-value users compared to other channels.
• Content Marketing & Influencer Partnerships: Increased to 25% ($375,000 remaining budget). This included expanding partnerships with micro-influencers and investing in more long-form content that demonstrated gameplay depth.

This reallocation meant shifting 35% of the original budget from last-touch heavy channels to upper-funnel and content-driven initiatives.

Results and Optimization (Weeks 9-12)

The impact of the AI-driven budget reallocation was evident. By focusing on channels that genuinely influenced the user journey, even if they weren’t the final click, we saw a significant improvement in overall campaign efficiency and ROAS.

• Total Installs (Campaign End): 450,000 (exceeded target of 400,000)
• Average CPI (AI-Adjusted): $3.33 (below target of $3.50)
• 90-Day ROAS (AI-Adjusted): 1.42x (exceeded target of 1.2x)

Table 2: Revised Campaign Performance (Weeks 5-12) based on AI Multi-Touch Attribution

The most striking change was the performance of content marketing and influencer partnerships. What was initially seen as a high-cost, low-return channel under last-click attribution transformed into the most efficient channel for driving high-quality installs when viewed through the AI model. Its AI-adjusted CPI of $2.08 was significantly lower than any other channel. This confirmed that these early-stage engagements were important for building trust and initial interest, which then led to conversions through other channels.

The overall campaign ROAS saw an 18% increase compared to the initial projections based on last-click data. The client acquired 50,000 more installs than their target, and these users showed higher engagement metrics post-install, including a 10% higher average revenue per user (ARPU) compared to previous campaign benchmarks.

What Worked and What Didn’t

What Worked:

• AI Multi-Touch Attribution: This was the single most impactful element. It provided actionable insights that traditional models simply could not. We were able to see the true value of upper-funnel activities. According to a eMarketer report from late 2025, companies using advanced attribution models reported a 15% average increase in marketing ROI. Our results align with this trend.
• Influencer Marketing: When properly attributed, influencer content proved to be a highly cost-effective way to build awareness and drive consideration. The authentic nature of these recommendations resonated deeply with potential players.
• UGC-style Video Ads: These ads, particularly on TikTok and Meta, performed exceptionally well in the early stages of the funnel, driving initial interest and clicks. Their authenticity made them highly engaging.
• Iterative Optimization: The weekly analysis and reallocation based on AI insights allowed for continuous improvement, preventing budget waste on underperforming channels.

What Didn’t Work as Expected (under AI lens):

• Ad Networks for High-Value Users: While still delivering volume, the AI model showed that users acquired primarily through ad networks had lower influence scores from early-stage touchpoints, suggesting they were often impulse installs with less pre-existing interest. This translated to lower ARPU.
• Generic Banner Ads: These consistently had low influence scores across the board. While they contributed to impressions, their impact on moving users through the funnel was minimal compared to interactive or video formats.

Optimization Steps Taken

1. Creative Refinement: We doubled down on short-form video content and user-generated themes for social and display ads, emphasizing authentic gameplay moments and community interactions.
2. Influencer Strategy Expansion: Increased budget for existing high-performing influencers and onboarded new micro-influencers whose audiences aligned with specific game features. We provided them with deeper game access and creative freedom.
3. Targeting Adjustments: For Paid Social, we shifted focus from broad interest-based targeting to lookalike audiences based on early-stage engagers (e.g., video viewers of influencer content) rather than just direct converters.
4. Bid Adjustments: Optimized bids across all platforms based on the AI-assigned value of each impression and click, rather than just the last-click conversion value. This meant bidding higher for early-stage impressions that the AI model identified as highly influential.

Conclusion

The “Quest for Echoes” campaign demonstrates that relying solely on last-click attribution in app marketing is a missed opportunity, often leading to misinformed budget decisions. By adopting an AI-driven multi-touch attribution model, we unlocked a deeper understanding of the customer journey, enabling strategic budget reallocation that significantly improved ROAS and overall campaign efficiency. Embrace advanced attribution to truly understand where your marketing dollars are making an impact.