Editorial Team Key Takeaways
- For their Q3 2025 product launch, Fresh Step used AI-powered ads to slash their CPL by 30% and boost ROAS by 25%.
- They did it with a dynamic creative optimization (DCO) strategy that generated over 500 ad variations to personalize the message for specific customer groups.
- Predictive AI models automatically moved the budget around in real-time, pushing cash to the best channels and creatives and cutting wasted ad spend by 18%.
- Before the campaign even started, they had the AI analyze historical purchase data and competitor activity, which uncovered some high-intent consumer groups they weren’t talking to.
- The AI’s performance data fed back to the human creative team for constant refinement, which kept pushing conversion rates up over the entire 12-week campaign.
Everyone’s talking about AI in marketing, but Fresh Step actually put it to work for their Q3 2025 product launch, and the results speak for themselves: serious cost savings and a big jump in marketing ROI. They showed how to use these adaptive systems for real impact measurement, not just basic automation. How did they transform their AI ad campaigns and ad spend into such compelling results?
Campaign Strategy and Objectives
For their new odor-control litter, Fresh Step had aggressive but clear goals for the first 12 weeks: get the cost per lead (CPL) below $5.00 and hit a return on ad spend (ROAS) of over 3.0x. They put a $750,000 budget behind it for that 12-week sprint, aiming for cat owners aged 25-54 in US cities and suburbs, especially those who’d bought premium pet products online before. Their strategy was built on using hyper-segmentation, dynamic creative optimization (DCO), and real-time budget allocation. The thinking was that intense AI-driven personalization was the only way to get noticed in a packed market. To get this done, they brought in a specialized AI marketing platform to chew on two years of their own first-party purchase data, mixing it with third-party consumer behavior data from sources like NielsenIQ (a 2024 NielsenIQ study on pet owner preferences, findable at nielseniq.com, gave them some early targeting ideas). This upfront analysis let them find micro-segments based on stuff like buying frequency, brand loyalty, and even preferences for specific product features like “clumping ability” or “natural ingredients.”
Creative Approach and Execution
The creative execution is where things got really interesting. Instead of producing a handful of static ads, they built a modular library of creative assets: various headlines, body copy permutations, call-to-action buttons, and a wide array of images and short video clips featuring different cat breeds, household settings, and problem/solution scenarios. This library fed their DCO engine, which was provided by their AI marketing platform. The DCO system automatically assembled over 500 unique ad variations. An environmentally conscious segment, for instance, received ads highlighting the product’s natural components and sustainability claims, featuring images of cats in bright, minimalist homes. Another segment, identified as value-driven, saw ads emphasizing the long-lasting nature of the litter and its cost-effectiveness, often with images of playful cats in more lived-in environments. This was a continuous, multi-variant optimization process where the AI was constantly learning in real-time which creative combos worked best for each specific audience. The creative assets were primarily distributed across Meta Ads (their team used the Meta Business Help Center documentation on DCO as a key reference) and Google Ads, leaning on Google’s support guides for responsive search ads and dynamic creative.
Targeting Precision and Channel Selection
Their targeting got way more granular than just broad demographics. Using the insights from their initial data analysis, they built out highly specific custom audiences. On Meta Ads, this meant lookalike audiences modeled from their high-value customers, layered with behavioral data points like “engaged shoppers” and “pet food purchasers.” On Google Ads, they combined in-market audiences for pet supplies with custom intent audiences built from people searching for their competitors’ brands, and then layered on precise geographic targeting down to the zip code in key cities like Atlanta, Denver, and Seattle. The bulk of the money went to paid social (Meta Ads, Pinterest Ads) and paid search (Google Search, Google Shopping). A smaller slice of the budget was set aside for programmatic display through a demand-side platform (DSP) that integrated with the AI’s bidding algorithms, making sure impressions were served at the right time to the right people. The whole time, the AI was watching performance across all channels, figuring out which platforms and ad formats were actually converting for each specific segment.
Performance Metrics and Results
The numbers were strong. Over the 12-week campaign, Fresh Step saw:
- Total Impressions: 85 million
- Click-Through Rate (CTR): 1.8% (average across all platforms)
- Total Conversions: 150,000 (defined as a completed purchase on their e-commerce site)
- Cost Per Lead (CPL): $4.17
- Return On Ad Spend (ROAS): 3.59x
These figures meant they beat their old campaign benchmarks with a 30% reduction in CPL and a 25% increase in ROAS. The campaign’s $750,000 budget was spent completely, with the AI reallocating funds on the fly. For example, in week 7, the AI spotted a surge of high-intent searches for “odor control cat litter” on Google Shopping that were converting like crazy, so it automatically moved 15% of the Meta Ads budget over to capitalize on it. The AI’s ability to automatically make these changes is its main advantage. A human team just can’t manage that level of granular adjustment at scale and speed.
What Worked Well
The dynamic creative optimization was the most effective part of the whole setup. Having so many specific ad variations meant Fresh Step was almost always hitting the right person with the right message, which directly led to better engagement and higher conversion rates. The predictive analytics also helped them find and target customer segments ready to buy that they would’ve otherwise missed with traditional demographic targeting. And the automated budget allocation was key for efficiency, it stopped wasting money on tired segments by moving it to where the campaign was actually working, making every dollar count.
What Didn’t Work as Expected
It wasn’t all perfect, of course. Early on, some of the longer video assets (anything over 15 seconds) just weren’t performing on Pinterest Ads, with really low completion rates. The AI caught it fast and pulled back on those assets for that platform, but it was a good reminder that we need even more granular testing for video length on different social channels next time. Also, the initial data integration, getting all the different sources to talk to the AI model, took longer than we planned. It needed a lot of data cleaning and standardization upfront. Honestly, that’s a common headache. Getting legacy data systems ready for AI ingestion is almost always more work than you think.
Optimization Steps Taken
This campaign was dynamic from day one. Here are some of the key adjustments made on the fly:
- Continuous Creative Refresh: When the AI flagged creative for declining performance (like a drop in CTR or a rise in CPL), the creative team would jump in and build new variations based on the AI’s suggestions, maybe trying new color schemes or more direct calls to action to keep everything fresh.
- Bid Strategy Adjustments: The AI’s bidding algorithms were always optimizing for max conversions under the target CPL, and as more conversion data came in over the 12 weeks, the AI got smarter, bidding more aggressively on high-value segments and pulling back on unprofitable ones.
- Audience Refinement: We fed post-purchase survey results and website behavior data back into the AI model every week. This helped refine the audience segments on the fly, uncovering new lookalike chances and cutting out segments that had high clicks but low conversions. For example, after week 5, the AI recommended cutting some zip codes within our target cities that were just wasting spend.
- Landing Page Optimization: The AI also provided insights on which ad variations led to better on-page engagement. The web team took that data and ran A/B tests on the landing pages, tweaking things like hero images and product description layouts to help optimize the user journey from ad click to conversion.
This back-and-forth between AI analysis and human judgment was what made it work. It’s proof that AI in marketing is a co-pilot, not a replacement for a good strategist who can interpret the data and guide the machine. In the end, Fresh Step’s Q3 2025 launch is a solid case study for how to use AI into advertising strategies to get real results. By adopting an adaptive, data-first approach, they hit and surpassed their ambitious goals, setting a new playbook for their future marketing efforts.
What kind of AI tech actually ran this campaign?
It was mainly a mix of three things: machine learning algorithms for the predictive analytics (finding audiences, setting bids), a dynamic creative optimization (DCO) engine to build and serve all the personalized ads, and some natural language processing (NLP) to analyze customer feedback and find new keywords.
How was the campaign’s ROAS measured?
Simple formula: total revenue from campaign conversions divided by total ad spend. They used a solid attribution model to connect ad clicks and views on their e-commerce site directly to sales from the AI ad campaigns, so the tracking was accurate.
How was the budget split between channels?
The starting point for the $750,000 budget was about 45% for Meta Ads, 40% for Google Ads (Search & Shopping), and 15% for programmatic display. But that was just the start. The AI constantly shifted those percentages based on what was working, so if Google Shopping was crushing it one week, it would get more of the budget automatically.
How much prep time did the AI setup and creative take?
The prep work took about seven weeks total. The first four weeks were for data work, cleaning and integrating two years of historical sales data, analytics, and consumer insights to train the AI. Then it took another three weeks to build the modular creative library and get it hooked up to the DCO platform.
What about data privacy with all that targeting?
They were careful to stick to privacy rules like GDPR and CCPA. All the third-party data was anonymized and aggregated, so it wasn’t tied to individuals. The first-party data they used was collected with explicit consent from customers through their site and loyalty programs, keeping everything compliant.
