Adobe AI: Agentic CX Cuts CPL 25% by 2027

Using advanced AI inside customer experience (CX) platforms offers a competitive edge in paid media. We’re seeing this with Adobe AI and its agentic features, which promise personalized engagement and campaign efficiency. The big question is whether these systems can actually deliver hyper-relevant ads at scale. Is this really going to transform paid media?

Campaign Teardown: “Project Horizon” for OmniRetail Solutions

In Q3 2025, we ran “Project Horizon” for OmniRetail Solutions, a B2B SaaS company in the supply chain space. This was a six-month paid media push with a straightforward goal: get 30% more qualified leads for their new inventory management software, but keep the Cost Per Lead (CPL) under $150. To hit that, we went all-in on the agentic CX features in Adobe Experience Platform (AEP) and Adobe Journey Optimizer (AJO), plugging them right into our Google Ads and LinkedIn Ads campaigns.

Strategy: Unifying Data for Agentic Personalization

Our strategy created a unified customer profile. We pulled everything, data from their CRM like Salesforce, product usage logs, Adobe Analytics data, and past marketing touches, into AEP to build a complete customer graph. We then set the agentic AI in AJO loose on this data to find high-intent segments, guess their next move, and spit out personalized ad copy and landing pages for our paid channels on the fly.

We bet that letting the AI autonomously tweak messaging based on what users were doing *right now* would blow past our manual campaigns. This was a learning system, not just some simple rule-based automation. For example, a user hits three “warehouse automation” product pages and grabs a whitepaper on “inventory forecasting.” The AI would see that and instantly fire a LinkedIn ad with a relevant warehouse automation case study. That kind of dynamic responsiveness is just impossible to replicate with standard segmentation and A/B tests.

Creative Approach: AI-Generated Iterations and Human Oversight

“Project Horizon” used a hybrid creative model where our team set the stage and the AI performed. Our in-house creatives laid down the brand guidelines, key messages, and a full library of visual assets. From there, the Adobe AI took over, generating hundreds of ad copy variations, headlines, and image ideas for the specific segments AEP was identifying. We used Adobe Firefly to spin up images quickly which ensured consistency and still gave us diverse interpretations.

The AI was surprisingly good at tailoring the message. For a segment like “Small to Medium Business (SMB) Owners,” the copy it generated would hit on cost savings and easy setup. For “Enterprise Supply Chain Directors,” it was all about scalability, ERP integration, and ROI. Our team had a weekly review of the top-performing AI creatives to give it feedback and make it smarter. That loop was everything. We noticed early on it was spitting out super technical jargon for audiences that weren’t technical, so we had to jump in and tweak its prompt parameters to get the tone right.

Targeting: Predictive Audiences and Real-time Adjustments

Our targeting differentiated the campaign. We didn’t just use standard demographics or interests. Instead, we used AEP’s predictive analytics to score individuals in our target accounts based on their “propensity to convert.” We then pushed those scores into Google Ads and LinkedIn Ads as custom audiences. The agentic CX system was constantly refreshing these segments, pulling people in or out based on their real-time behavior.

Here’s how it worked in practice. A prospect from a manufacturing company spends a lot of time on OmniRetail’s blog reading about “lean manufacturing principles” but doesn’t look at any product pages. The AI flags them as “information-gathering” and hits them with a top-of-funnel industry report on LinkedIn. But then, that same person clicks “request a demo” on a competitor’s site (we saw this through a third-party intent data feed in AEP). Their propensity score shoots up, and the AI immediately starts serving them direct response search ads on Google. This targeting allocated budget efficiently to the people showing the strongest intent signals.

Campaign Performance Metrics and Analysis

Project Horizon ran for six months, from July 1, 2025, to December 31, 2025. Here’s a breakdown of the key metrics:

Overall Campaign Performance (July – December 2025)

• Total Budget: $300,000 ($50,000 per month)
• Total Impressions: 12.5 million
• Overall Click-Through Rate (CTR): 1.85%
• Total Conversions (Qualified Leads): 2,350
• Average Cost Per Lead (CPL): $127.66
• Return on Ad Spend (ROAS): 3.2x (based on estimated lifetime value of qualified leads)

Platform-Specific Performance

What Worked: Precision and Adaptability

The AI’s ability to deliver hyper-personalization at scale was the main success. We came in 15% under our target CPL at $127.66, mostly because the ads were just that much more relevant. The proof was in the landing pages: we saw a 25% higher conversion rate on pages AJO dynamically changed based on the user’s profile versus our static control pages. Because the AI was always learning, the campaign just kept getting better over the six months instead of hitting a wall.

A perfect example happened in October. The AI picked up on a spike in searches for “supply chain resilience” after some global shipping problems made the news. On its own, it adjusted our Google Ads bids for those keywords and started pushing new ad copy about OmniRetail’s disruption-mitigation tools. We saw a 10% lift in conversions from that keyword group that month, and our team didn’t have to do a thing. Agentic AI reacts to environmental and user shifts in real-time.

What Didn’t Work: Initial Over-Reliance and Data Silos

It wasn’t all smooth sailing, especially in July. The first batch of creatives the AI came up with were a mess. Some were just off-brand, and others missed the specific tone OmniRetail wanted. This highlighted the need for human oversight and feedback. We made the mistake of giving the AI too much freedom at the start without enough guardrails on tone and style. It’s a common pitfall. The AI isn’t a mind-reader, especially not on day one.

We also hit a wall trying to get certain third-party data sources into AEP. We wanted that unified profile, but legacy data from OmniRetail’s old marketing automation system was a nightmare to standardize. The result was a data lag where the AI was making calls on incomplete or old user data. It took us almost a month to clean up those data pipelines. This just shows that AI effectiveness depends on data quality. If data isn’t clean, agentic AI struggles.

Optimization Steps Taken

Based on our findings, we implemented several key optimizations:

1. Enhanced AI Guardrails and Feedback: We started a strict weekly review for all AI-generated creative. Our team gave direct “thumbs up” or “thumbs down” feedback, with written notes, right into the Adobe Sensei machine learning model to teach it the brand voice.
2. Data Governance and Standardization: We launched a full data governance project to clean up all incoming data. That meant creating strict schemas and automated validation rules in AEP, which cut our data discrepancies by 40% in two months.
3. Dynamic Budget Allocation Rules: We set up rules in AJO to move budget between Google and LinkedIn automatically based on real-time CPL. If LinkedIn’s CPL for a segment went over $160, AJO would pull budget and send it to a similar Google Ads campaign until the cost came back down. This cut a lot of waste.
4. Sentiment Analysis Integration: We hooked a sentiment analysis tool into AEP to watch social media and customer service chats. When it detected negative chatter about “delivery delays,” the AI would automatically change the ad copy to talk up OmniRetail’s “guaranteed on-time delivery,” flipping a problem into a selling point.

These optimizations yielded results. In the final quarter of the campaign (October-December), the overall CPL dropped by an additional 8%, and the ROAS improved to 3.5x. This demonstrated the power of continuous refinement. The AI learns faster and more effectively with human guidance.

Adobe AI and agentic systems are the future of paid media. “Project Horizon” proved these technologies deliver incredible personalization and efficiency, but only with a solid data foundation, clear objectives, and human oversight. Marketers must shift focus to strategic orchestration of intelligent campaign flows. The real challenge is adapting our processes and skills to harness its potential. For more insights on this, consider exploring how AI shifts demand a 2026 strategy overhaul.