Key Takeaways
- We pushed ROAS up 12% to 3.8:1 on the Q3 2025 “ConnectHome Smart Lighting” campaign, beating our previous benchmarks.
- Handing ad copy and landing page work to agentic AI cut our content creation costs by a solid 28%.
- Dynamic targeting, which we ran through a proprietary agentic marketing platform processing real-time behavioral data, got our Cost Per Lead (CPL) down 18% to $18.50.
- Our initial A/B tests showed agentic-generated video ads converted 7% better than the ones our human team had produced for a prior campaign.
- The dedicated $75,000 we set aside for this agentic experiment delivered a positive ROI inside the first six weeks.
Our Q3 2025 review for the “ConnectHome Smart Lighting” campaign is done, and it’s clear the agentic AI experiment paid off. These autonomous systems, which can learn and make their own decisions, are changing how we run campaigns from the ground up. The real question we had going in was how far you can push this automation before it breaks and you lose that necessary human feel. Can these systems actually fit into a complex strategy without creating a mess?
| Metric | Traditional Approach (Benchmark/Target) | Agentic AI-Driven (Actual) |
|---|---|---|
| Return on Ad Spend (ROAS) | 3.5:1 | 3.8:1 (+12% increase over previous) |
| Content Creation Costs | Standard (Implied higher) | Reduced by 28% |
| Cost Per Lead (CPL) | $25.00 | $18.50 (18% reduction) |
| Video Ad Conversion Rate | Human-produced baseline | 7% higher for agentic variants |
| Campaign Duration | 12 weeks | 12 weeks |
| Experimental Budget ROI | Not applicable | Positive within six weeks |
Campaign Teardown: ConnectHome Smart Lighting (Q3 2025)
Strategy and Objectives
For the “ConnectHome Smart Lighting” campaign, our goal was simple: sell more of our new intelligent lighting to homeowners in the Atlanta suburbs. We were specifically targeting 30-55 year olds in places like Alpharetta, Roswell, and Johns Creek. Our main objectives were to generate qualified leads and drive direct sales on our e-commerce site, with a target Return on Ad Spend (ROAS) of 3.5:1 and a Cost Per Lead (CPL) under $25 for the twelve-week run from July 1 to September 23. The big bet was using a new agentic marketing platform, AdCreative.ai, to handle dynamic creative and optimization. We plugged it directly into our existing Salesforce Marketing Cloud instance to create a live feedback loop for scoring and personalizing leads. Our whole hypothesis rested on the idea that these agentic systems could build and personalize ad experiences at scale, something our team could never do manually.
Creative Approach: Human-Guided Agentic Generation
We didn’t just hand the keys to the AI. Our creative process was more like human-guided agentic work. We gave the platform our core messages, brand guidelines, and a library of approved product shots and short video clips. From there, the agent spun up tons of ad copy, headline, and call-to-action (CTA) variations for Google Ads, Meta Ads, and our programmatic display networks. It even used its text-to-video tools to create 15-30 second explainer clips and testimonials, stitching our footage together with dynamic text and voiceovers which took our video lead time from weeks down to a couple of days (a huge relief). We put it to the test, running A/B splits of the AI’s video ads against our older, human-made ones from a previous campaign. The agentic variants, to the surprise of a few people on the team, actually won with a 7% higher conversion rate on average, which we think is because they could iterate and find the right personalization hook so damn fast.
Targeting and Placement
Our targeting was driven by real-time behavioral data streams. The agentic platform was constantly watching for intent signals like search queries and website interactions. So, if someone in the Marietta area searched for “smart home automation” or clicked on a competitor’s smart lighting page, our system would immediately serve them a custom ad for ConnectHome that highlighted a unique feature like our adaptive lighting schedules. Placements were focused on Google Search and Display Networks, the Meta apps (Facebook and Instagram), and premium inventory we bought through a Demand-Side Platform (DSP). The geotargeting was dialed in, focusing on specific zip codes in Fulton, Cobb, and Gwinnett counties where the average household income is over $80,000, which is our sweet spot for this kind of tech.
Campaign Performance: Metrics and Analysis
The total campaign budget was $300,000 for the twelve weeks, with a dedicated $75,000 of that set aside just for testing the agentic platform’s capabilities.
Table 1: Campaign Performance Summary (Q3 2025)
| Metric | Target | Actual | Variance |
|---|---|---|---|
| Total Budget | $300,000 | $298,500 | -$1,500 |
| Duration | 12 weeks | 12 weeks | 0 |
| Impressions | 15,000,000 | 16,800,000 | +12% |
| Click-Through Rate (CTR) | 1.8% | 2.1% | +0.3 pts |
| Leads Generated | 10,000 | 12,700 | +27% |
| Cost Per Lead (CPL) | $25.00 | $18.50 | -$6.50 |
| Conversions (Sales) | 1,200 | 1,550 | +29.2% |
| Cost Per Conversion | $250.00 | $192.58 | -$57.42 |
The numbers speak for themselves. We beat our impression target by 12% to hit 16.8 million, and the CTR improved to 2.1%, which tells me the AI-generated creative connected with people. The big wins were on the financial side: CPL came in at $18.50 (an 18% drop from our $25 target) and ROAS hit 3.8:1, beating our 3.5:1 goal. That efficiency comes straight from the agentic platform’s ability to find winning ad combinations and scale them instantly while killing the losers before they burn too much budget. A late 2024 eMarketer report mentioned early generative AI adopters were seeing similar gains, so it’s good to see our results are tracking with the industry. If you want to get into the weeds on measurement, you can read up on experiential ROI and proving AI value in 2026.
What Worked
Dynamic Creative Optimization: The system’s ability to generate and test hundreds of ad variants in real time was the biggest win. This let us personalize at a scale that was just impossible before, with the agent automatically tweaking headlines and images based on live user data. Automated Bid Management: Tying the agentic platform into the Google and Meta APIs allowed for some very smart, real-time bid changes. It made sure we were paying the right price for clicks, which really maximized our budget efficiency, especially during the evening peak conversion hours in the Atlanta metro. Cost Reduction in Content Creation: We used the agentic tools to produce first drafts of ad copy and assemble basic videos, and it cut our content creation costs by about 28% compared to what we spent in 2024. That freed up budget we could put into better analytics.
What Didn’t Work
The campaign wasn’t perfect. Early on, some of the AI-generated ad copy it produced was way too technical and missed our brand voice, even though we fed it the guidelines. This meant our team had to do a lot of manual review and editing in the first two weeks. We also noticed the system started to rely too heavily on retargeting. Left on its own, the agent would aggressively go after anyone who showed the slightest interest which just led to ad fatigue and wasted impressions. We had to step in and set stricter frequency caps by hand.
Optimization Steps Taken
To fix the brand voice issue, we set up a tighter feedback process. Our copywriters would flag bad examples and feed the system good ones, basically training it on our specific style. That human-guided iteration got the AI’s content on-brand pretty quickly. For the retargeting problem, we tweaked the agent’s rules to be smarter about recency and frequency. We started segmenting audiences based on how deep their engagement was, so someone who abandoned a cart got a very different retargeting message than someone who just glanced at a product page. This cut down on the ad fatigue and actually improved our conversion rates in those retargeted groups, something the IAB’s programmatic guides always preach. We also beefed up our negative keyword lists for Google Search, getting rid of a lot of vaguely related terms that weren’t showing any real purchase intent. It seems like a small tweak, but it made a big difference in our traffic quality and helped lower the CPL even more.
Future Outlook
This ConnectHome campaign proved the value of these agentic tools in our MarTech stack. The future here is clearly a partnership between human strategists and these smart agents. Our experience with the $75,000 experimental budget showed that while the AI is incredible at optimization and working at a scale we can’t match, you absolutely still need human intuition for brand direction and high-level strategy. The point isn’t to replace marketers. It’s to get them out of the weeds of A/B testing a hundred headlines so they can focus on what’s next. We’re already planning to feed real-time sentiment analysis into the system, letting it adapt creative based on how audiences are emotionally responding to the ads. This campaign gave us a blueprint: a human-guided, AI-driven approach is how you get through the noise of modern digital marketing.
What is an agentic shift in MarTech?
It’s about using AI systems that don’t need constant hand-holding. These “agents” can make their own decisions to execute marketing tasks, like changing bids, generating new ad copy, or adjusting targeting, in real time based on what the data is telling them.
How did agentic AI impact the cost of content creation for the ConnectHome campaign?
It cut our content creation costs by about 28%. We used agentic AI to generate tons of ad copy variations, headlines, and even rough drafts of videos which meant we could reallocate that part of the budget to things like strategic oversight and better analytics.
What specific platforms or tools were used for agentic marketing in this campaign?
Our main tool was AdCreative.ai for the dynamic creative work. We had it integrated with Salesforce Marketing Cloud, which handled the CRM and lead management side. We also used the built-in AI functions inside Google Ads and Meta Ads for some of the automated bidding and targeting.
What were the key performance indicators (KPIs) that saw the most improvement due to agentic shifts?
The two biggest improvements were in Cost Per Lead (CPL), which we got down by 18% to $18.50, and Return on Ad Spend (ROAS) which climbed to 3.8:1. We also saw our Click-Through Rate (CTR) go up to 2.1%, which told us the agent-generated creative was working better than our old stuff.
What was a key challenge encountered with agentic content generation and how was it addressed?
The biggest problem at first was that the AI’s ad copy didn’t sound like our brand. To fix it, we created a tighter feedback loop where our human copywriters would give the system specific examples of good and bad phrasing. The agent used that feedback to learn our style and got the content aligned with our brand guidelines pretty quickly.
