Adobe CX: 2026 Attribution Models Drive ROAS

Adobe’s AI tools are changing the game for marketers, moving way beyond simple personalization. With agentic CX inside Adobe Experience Cloud, we’re now talking about AI that can guide a user through a complicated setup or anticipate a problem and solve it proactively. The problem is, how do you measure the value of that interaction? If your company is still stuck on last-click attribution, you’re flying blind, because you have no way to credit these AI touchpoints that happen long before the final sale.

Deconstructing the “Project Aura” Launch: A Campaign Teardown

I want to walk through a campaign we just tore down, “Project Aura.” It was a three-month push in Q4 2025 from a big B2B SaaS company we’ll call “InnovateSphere,” and they were trying to drive sign-ups for a new AI platform feature. This feature used Adobe’s agentic CX for better user onboarding, and they put a serious $850,000 budget behind it from October 1 to December 31, 2025. Their goals were pretty aggressive: a 2.5x ROAS (Return on Ad Spend) and a Cost Per Lead (CPL) under $120 while going after enterprise tech and finance execs.

Strategy and Creative Approach

Their whole strategy was to show people what agentic CX could actually do, using interactive micro-experiences instead of just talking about it. They built a series of short, personalized video ads showing the AI assistant walking a user through difficult software setups and fixing common problems. These ads ran on LinkedIn, the Google Display Network, and some niche industry forums, all pointing to landing pages with an embedded demo where you could try a simulated agentic interaction yourself. The creative was built around problem-solving, with headlines like “Your AI Co-Pilot for Complex Workflows” and “Effortless Onboarding, Intelligent Support.”

The targeting was surgical. For LinkedIn, they went after job titles like “Head of IT,” “VP of Operations,” and “Chief Technology Officer” at companies with more than 500 employees. On the Google Display Network, they built custom intent audiences from search queries like “AI customer support software” and “intelligent workflow solutions.” They kept the geography tight, focusing on tech hubs like San Francisco, Austin, and New York City, deciding to hold off on international markets for the first run.

Initial Performance and Attribution Challenges

The campaign pulled in 15 million impressions and a 1.8% CTR (Click-Through Rate), which isn’t bad for B2B. But the conversions weren’t hitting the mark. Their Cost Per Lead (CPL) was stuck around $180 in the first month, way over the $120 target. The problem was their attribution model, straight last-click, which completely ignored a huge piece of the puzzle. We saw users get to the landing page, spend a ton of time in the AI demo, and then leave without converting right away. They’d come back days later through a direct visit or an organic search to complete the sign-up.

When we dug in, it was obvious last-click was killing them. The model completely missed the value of the agentic CX demonstrations. We saw huge engagement on those demo pages, with an average time of 3 minutes 45 seconds and 70% of users finishing at least one task with the AI. The interest was there. The real issue was that the credit assignment was totally disconnected from what was actually influencing people to sign up, which was the AI experience itself, even if the conversion happened a week later.

Implementing a Multi-Touch Attribution Model

So, we switched them to a time decay attribution model within Adobe Analytics. This model gives more weight to touchpoints closer to the conversion but still gives some credit to earlier interactions. We specifically tuned it to value engagement with the agentic CX demo more heavily, and we started tracking micro-conversions like “AI demo completion” and “time spent engaging with AI assistant” to get a clearer picture of what was working.

The change was immediate. Within just two weeks, the conversion paths looked completely different. Paid social campaigns that served as the first introduction to the AI concept started getting the credit they deserved. The AI demo engagement on the landing page which last-click had ignored, was now clearly correlated with eventual sign-ups. For some of their high-value enterprise leads, we also implemented a custom U-shaped attribution model, which better reflects complex sales cycles by weighting the first and last touches most heavily. We could finally see how a person who interacted with the demo was more likely to convert down the road.

Optimization and Results

Once we could actually see the impact of the agentic CX, making optimizations was straightforward:

1. Increased Budget Allocation: We pulled 20% of the budget from top-of-funnel awareness and poured it into retargeting people who had used the AI demo but hadn’t converted.
2. Personalized Follow-up: We used data from the agentic CX module to send personalized emails. If someone explored a specific feature in the demo, the follow-up email talked about that exact feature.
3. A/B Testing Landing Pages: We tested landing page CTAs. The generic “Sign Up” was a loser. The winner, which boosted post-demo conversions by 15%, was a more consultative CTA: “Schedule a Demo with an AI Expert.”

The campaign completely turned around. By the end, the CPL fell to $105, beating the $120 target, and the overall ROAS hit 2.8x, exceeding the 2.5x goal. Total platform sign-ups jumped to 280. Dropping the Cost Per Conversion from a staggering $8,000 at the start to just $3,035 by the end shows what happens when your attribution model actually reflects reality and lets you make smart, actionable decisions.

What Worked and What Didn’t

What Worked:

• The interactive AI demo was a home run. The engagement data proved people want to try this tech for themselves, not just read a whitepaper.
• Shifting to a time decay model was the key. It gave us the real story of the customer journey and let us properly value those critical AI interactions.
• Retargeting users based on what they did in the demo was extremely effective for pushing up conversion rates.
• Those personalized emails referencing their demo activity got much higher open and click rates.

What Didn’t Work as Expected:

• Starting with last-click attribution was a huge mistake that hid the value of the AI demo. It’s a classic error, and frankly, I see it all the time with companies that are slow to adapt their measurement.
• The generic “Sign Up” button after the demo was the wrong move. Users who just had a complex, specific experience wanted a more guided next step.
• Their initial broad keyword targeting on the Google Display Network just brought in low-quality leads until we narrowed it down with custom intent audiences.

This whole campaign proved that agentic CX is a serious part of the customer journey that requires a serious attribution strategy. If you can’t measure its impact correctly, you’ll waste money on the wrong channels and completely miss how much value your AI investments are creating, like thinking a $3,035 conversion actually cost $8,000. The next step for a company like this is to go deeper, integrating predictive data from tools like Adobe Customer Journey Analytics. That’s how you move from just assigning credit for a single conversion to modeling the entire lifetime revenue generated by each of these AI interactions.