Playable Ads with AI: Our Journey, Insights, and Why We Pivoted
The journey to picking this idea
In May 2024, when I decided to give an attempt to starting up (Here's how I built conviction to startup), I started exploring the idea of building voice agents for consumer research but paused in July 2024 (Invalidated building in consumer research). A senior advised us that being idea-agnostic, it was better to stay close to Marketing/Sales due to clearer revenue attribution. We followed this advice and stayed within the marketing function for consumer companies (B2C apps, D2C brands, CPG, Gaming), leveraging our previous interactions with marketing teams.
After pausing the original idea, we faced a challenging period with no strong hypothesis or insights. People were reluctant to talk to us initially, as we didn't offer immediate value. However, persistence allowed us to explore these hypotheses:
- Automating growth ops for consumer companies (personalizing push notifications, WhatsApp, emails, banners).
- Text to HTML email builder.
- Hyper-personalized email marketing for D2C brands (responsible for ~20% of D2C revenue in the US).
Given my previous experience as a consumer tech product manager, I initially chose retention marketing. However, acknowledging that performance marketing had higher needs and spends, we shifted our focus.
Initial experiments and pivot
Our hypothesis centered around the fact that performance marketers upload numerous videos but struggle to learn what works best. They used Excel sheets to find patterns, which we felt could be enhanced with AI. In October 2024, using Gemini, we began extracting video features (actor gender, environment, ambience, quality, music) and correlating them with performance metrics. Here's an attempt on social media videos.
However, we pivoted as insights lacked statistical significance, marketers already knew them, and the problem wasn't frequent enough.
Discovering playable ads
Through our experiments, we realized the primary job of performance marketing teams was creating and testing ad creatives, mostly videos. This led us to explore AI-powered ad creation, initially targeting tools like Kling. However, marketers were already using various AI tools, and quality concerns emerged.
We considered gaming, believing AI could easily generate quality game ads without human actors. Researching gaming ads, we discovered playable ads—interactive game ads offering a "try-before-you-buy" experience. Despite their effectiveness, adoption was limited due to high development costs. We successfully created a match-3 game ad prototype using Claude, sparking excitement.
Quick validation without a product
A friend suggested pitching our idea at IGDC (India Game Developer’s Conference), just three days away. Cold-emailing speakers with the subject "Why is everyone talking about playable ads" yielded a surprising 20% reply rate. Conversations revealed that playable ads were crucial for US-focused games but less relevant for Indian audiences. We had identified a global, not local, problem.
This event led to collaborations with Felicity Games, Nukebox Studios, and MPL. We initially acted as playable ad developers, following the principle "do things that don’t scale," building expertise:
- Created and launched 25–30 ads using Cursor (98% AI-generated code).
- Learned game development and contributed to the Phaser game library.
- Found no other products targeting this niche effectively.
Jevon’s Paradox and hypothesis refinement
I strongly believe in Jevon's paradox: increased efficiency in resource use often increases consumption. If playable ads could be created within an hour instead of weeks by non-technical users, ad creation could expand significantly. We envisioned a vertical platform (lovable.dev) enabling marketers to easily build playable ads, thus expanding the market.
Hypothesis and initial excitement
Our refined hypothesis and pitch included:
- Market: Gaming companies earn more than other entertainment sectors combined, spending over $20B on ads.
- Problem: Performance marketers increasingly need faster ad creation but playable ads take weeks and developer resources.
- Insight: General AI tools weren't sufficient; playable ads had unique requirements (specific assets, landscape/portrait modes, file sizes, network integrations).
- Solution: A vertical lovable.dev for playable ads as an entry into the gaming ad market.
- Competition: Playable Factory ($10M ARR, but not a self-service SaaS) and other limited or clunky products.
This hypothesis excited me so much I enthusiastically discussed it during an hour-long walk with a friend, convinced we had found a promising opportunity.
Conversations in the US and finding design partners
With this insight, we joined the SPC (South Park Commons) Founder Fellowship on March 4, 2025, who funded our pre-seed ($400K for 7%) and flew to the US on March 13. We attended the Game Developers Conference and Pocket Gamer Connects in San Francisco. Despite being uninvited, we gained entry to a Deconstructor of Fun event, connecting with major gaming companies like Rovio Entertainment, Scopely, and Wildlife Studios, who validated our idea.
Product development challenges
Building a product required developing an AI agent capable of converting instructions and assets into interactive game ads. Our technical approach involved:
- Using Bolt’s open-source version for user interfaces.
- Gemini to translate marketing videos into product requirements.
- Task breakdown and testing via Claude Code.
Key challenges
- Difficulty in automating game interaction tests.
- Time-sensitive game mechanics couldn’t be easily tested.
- Lack of AI feedback loops made it hard to assess current progress.
We decided to initially focus on perfecting one game category before scaling.
Disproving the market expansion hypothesis
Encouraged by investors, we built a business case to test market expansion assumptions. We asked gaming companies if faster ad creation (1 hour instead of 2 weeks) would increase ad volume. They confirmed it wouldn't; high testing costs ($5K–$10K per ad set on Applovin vs $20–$50 on Meta) prevented increased testing. Furthermore, improving playable ad performance wasn't a priority for Meta.
Thus, we concluded our solution addressed a problem, but not one urgent or impactful enough to significantly change market behavior.
The difficult decision to pause and kill the idea
The failed market expansion hypothesis, coupled with gaming industry consolidation, led us to pause the idea. SPC supported this move, advising us to pause and reassess after a month. Our decision was positively received by the cohort and investors.
After a month, we didn't feel excited to come back or see enough pull in the idea, so we killed the hypothesis and moved on.
Stay tuned for more insights on what happened next.
Reflections and learning
- Finding a Martech or Salestech niche is challenging but rewarding due to higher payment motivation.
- Always validate or invalidate specific hypotheses instead of vague exploration.
- Pursue the biggest possible idea instead of settling for local maxima.
- Build robust business cases for enterprise solutions.
- Maintain truth-seeking rigor, balancing excitement with rationality.