Amazon’s AI Shelf completely changed how apps get found and sold inside their massive store. For us, putting an AI tool on that shelf, right in the middle of the shopping experience, was a test. It gave us a new look at how app discovery is changing for any business trying to get noticed. The big question was simple: could this thing actually drive installs and, more importantly, make us money?
Key Takeaways
- We hit a 1.8% conversion rate, which is decent for a brand new discovery channel but shows it’s not a magic wand.
- Targeting people based on their Amazon purchase history and AI interests gave us a 25% higher CTR than just using broad demographics.
- Creatives that showed the app in action and spelled out the value beat our generic branding with a 15% better impressions-to-install rate.
- The $500,000 budget for a 6-week pilot was enough to let us test and tweak on the fly, which helped us cut our CPL by 12% by the end.
The Campaign Blueprint: Integrating AI into Retail Discovery
We launched a pilot on the Amazon AI Shelf for “SynthSketch,” our generative AI art app, back in Q3 2025. The goal was straightforward: use Amazon’s user base and the hype around AI to get installs and subscriptions. We ran the test for six weeks straight, from September 1st to October 15th, and put a $500,000 budget behind it. That money had to cover everything, making the creative, paying for the ad placements, and having a team on standby for real-time changes. We were shooting for a cost per install (CPI) below $5 and, critically, a 1.5x return on ad spend (ROAS) within three months, based purely on subscription revenue from the new users we acquired.
Our whole strategy was built on Amazon’s unique targeting data. We went after users who had already bought AI-related gear (think smart speakers or AI-powered laptops), signed up for AI newsletters, or were constantly browsing the “Software & Apps” section for creative tools. For geography, we focused our fire on Los Angeles, New York City, and Seattle, since that’s where early adopters usually cluster. Keeping the geo-targeting tight at the start meant we could watch our budget more closely and get cleaner data which was essential for a placement this new.
Creative Approach: Show, Don’t Just Tell
The creative assets for SynthSketch had to do the heavy lifting. We produced a set of quick 15-30 second video ads along with some static image carousels. The videos got right to the point: they showed someone typing a prompt and then the AI generating a slick piece of art. We hammered home how easy it was. Our static carousels displayed the range of art styles the app could create, with each image made by SynthSketch itself and paired with short copy like “Transform ideas into art” and “No artistic skill required.”
We spent a good chunk of the budget, about $75,000, just on making these assets look good. Our bet was that showing people what the AI could do would work far better than just talking about it, and we were right. The videos really stole the show, pulling in a 1.2% click-through rate (CTR), which easily beat the 0.8% we got from static images. It was clear proof that for an interactive AI app, you need to show it in motion.
Targeting Refinements and Performance Metrics
Our initial targeting plan was solid, but the first week’s numbers told us we could do better. Our cost per install (CPI) came in at $6.20, a little rich for our blood. Digging into the data, we saw our interest-based segments were working well, but we also noticed a pocket of users from the general “productivity apps” category who were converting. They weren’t our primary target, but they were installing.
So, in week two, we broadened our reach to include people interested in graphic design software or digital illustration tools, even if they hadn’t bought any specific AI hardware. That one change immediately brought our CPI down to $5.50. By week three, we tweaked the ad copy to focus on how SynthSketch saves time for creative pros, and the CPI fell again to $4.85. All told, the campaign pulled in 12.5 million impressions over its run, which converted into 150,000 clicks and then 27,500 app installs. That gave us an overall CTR of 1.2% and a click-to-install rate of 18.3%. Our final cost per lead (the install) averaged out to $4.55, right in our target zone.
But we tracked everything after the install, too. Of those 27,500 installs, only about 2,500 users actually signed up for a paid monthly subscription within 90 days. At an average of $9.99 a month, that gave us an initial MRR boost of about $24,975. When we projected that out over the three-month ROAS window, the total revenue was just $74,925. Against our $500,000 spend, that’s a 0.15x ROAS, way off our 1.5x goal. It was a tough lesson: getting someone to install is one thing, getting them to pay is another.
What Worked and What Didn’t
The Amazon AI Shelf is definitely a good channel for getting eyeballs and initial installs. Being part of the Amazon platform gives you a level of credibility and reach that you just don’t get from a standard app store ad. Targeting based on what people actually buy on Amazon (like specific gadgets) was a huge plus, giving us a peek at user intent that other ad platforms can’t. And our visual-first creative strategy, especially with the videos, absolutely worked.
But the campaign threw our biggest weakness into sharp relief: converting those installs into paying customers. The terrible ROAS was a bucket of cold water. The AI Shelf could get people in the door, but the journey from a free download to a paid subscription was broken. We saw a huge number of uninstalls in the first 72 hours from people who didn’t immediately find the app useful. It was obvious that just being featured on a fancy “AI shelf” means nothing if the app itself doesn’t grab a user right away and show its value.
Another frustration was the lack of real A/B testing on the shelf itself. We could test creatives and targeting all day long, but we couldn’t test our app’s actual position or prominence against other apps on the shelf. This meant we couldn’t do much to optimize for visibility besides just throwing more money at our bids.
Optimization Steps Taken and Future Outlook
After the pilot, we got to work. First, we completely redesigned the app’s onboarding to be faster and more satisfying, making sure a new user could create their first piece of AI art in under a minute. Second, we scrapped our old three-day limited trial and rolled out a seven-day free trial with full feature access, followed by a much clearer prompt to subscribe. The idea was to give people enough time to get hooked on the app’s power.
We’re also now working on deep linking from the Amazon ads directly to specific templates inside SynthSketch, not just to the main app store page. If a user clicks an ad for a fantasy art style, they should land on the fantasy art generator. It just removes friction. For the next AI Shelf campaign, we’ll be obsessed with post-install engagement and run retargeting ads to users who installed but never subscribed. We’re even thinking about offering exclusive Amazon Prime member discounts on our subscriptions to make the offer more compelling.
Despite the painful ROAS, the Amazon AI Shelf is a promising new place for app discovery, especially for apps like ours that are very visual and tied to specific interests. The platform gives you an audience, but converting and keeping them is still 100% on you and your product. You can’t expect a new discovery channel to magically fix your monetization problems. That takes a complete, well-thought-out plan.
The way people find apps on platforms like Amazon is going to be driven by these complex AI recommendation engines. That means developers and marketers have to get smart about how their products are seen and sorted by these systems. You need to optimize for the user behavior patterns that the AI models understand, not just for keywords. The AI Shelf is just the start of this trend, and you have to adapt to these changes if you want to achieve real market penetration.
What is the Amazon AI Shelf?
It’s a special section on Amazon’s site and app that’s built to feature AI-powered applications and services, making it easier for Amazon’s customers to find and try new AI tools.
How effective was the SynthSketch campaign on the AI Shelf in driving app installs?
It was effective for getting people in the door. We got 27,500 installs over six weeks at a respectable CPI of $4.55, but the real work started after the install to get those users to become paying subscribers.
What targeting methods proved most successful for the AI Shelf campaign?
The best results came from targeting users based on their Amazon purchase history (like AI hardware) and their interest in AI newsletters. We also found a valuable secondary audience by expanding into broader “productivity app” segments, which helped lower our costs.
What was the primary challenge faced during the campaign?
The biggest hurdle was converting those initial app installs into paid subscriptions. This led to a very low return on ad spend (ROAS) of just 0.15x, proving that driving installs and driving revenue are two very different problems.
What optimization steps were taken after the pilot campaign?
We overhauled the app’s onboarding to make it more engaging, extended the free trial to a full seven days to let users experience the value, and started exploring deep linking from ads directly to features inside the app to improve the path to subscription.
