Sett builds AI-powered playable ads—interactive mini-games used to acquire users for mobile games. The 45-person company based in Tel Aviv serves nearly 50 game studios, from mid-size publishers to top-grossing enterprise studios, helping them overcome creative fatigue and find winning ad concepts faster.
Mobile game studios face a tough market: you can't predict which ad creative will win. The only solution is to explore widely, testing dozens of concepts to find the few that drive performance. But creating high-quality playable ads—interactive mini-games that showcase a game's appeal in seconds—is slow and expensive. A single playable ad can take a month or more to produce manually.
Sett's customers are studios spending hundreds of thousands to millions of dollars monthly on user acquisition. They've tried agencies. They've tried scaling internal teams. Neither solved the core problem: producing enough creative diversity to consistently find winners. These studios run campaigns across Meta, AppLovin, Unity, and Google, and need dozens of playable ads monthly just to discover which concepts perform well enough to scale ad spend behind them.
"We needed a model that could go beyond pure logic to understand abstract concepts like 'taste' and create an end product that is genuinely attractive and fun for users," said Yoni Blumenfeld, CTO and Co-Founder at Sett.
Opus 4.5 serves as the core model at the heart of Sett's architecture, handling the planning and creative coding that brings playable ads to life. Beyond generating content, Claude functions as an orchestrator agent. That means managing the entire production workflow and intelligently delegating specific tasks to other specialized diffusion models, using them as tools to achieve the best possible outcome.
Sett's decision to build on Claude came from comparative testing against other leading models. The company developed an internal metric called the "Last Mile"—measuring how much human work remained after AI completed its portion of a task.
"In our comparative tests against other leading models, Claude demonstrated a superior ability to handle our creative coding tasks," said Blumenfeld. "Our 'Last Mile' metric confirmed that Claude's output required the least amount of refinement to reach a finished state."
The results were decisive. Before implementing Claude in their agentic system, producing a single playable ad required about a month of human planning, design, creative coding, and iteration. The last-mile work for a single ad is about one week. With Claude as the core model, that dropped to 75 minutes. Sett's system still requires a human in the loop to QA and polish output for maximum production quality, but Claude's higher-quality output means far less refinement within that process.
Implementation moved quickly once the decision was made. Sett invested significant effort in building evaluation systems, breaking down the complex creative task into component sub-evaluations to measure performance at each stage. The teams at Anthropic and AWS Bedrock provided support on operational aspects, helping resolve scaling challenges efficiently.
The impact shows up directly in customer results. For one game studio, 90% of ad spend is now driven by Sett-generated ads powered by Claude. The two leading ad concepts command about a quarter of all spend and drive the highest return on ad spend.
"Creative is the fuel for growth, and Sett makes sure you never run out," said Daniel Fridman, Performance Lead at Papaya, another Sett customer. "It helped us test bold ideas, adapt to new trends, and stay ahead of the game. What used to take weeks or even months now takes days."
The real advantage is volume and diversity. When studios can produce massively more creative diversity, they find more winners faster. Creative performance in user acquisition is a numbers game, and Sett lets game studios play that game with confidence.
Looking ahead, Sett is building toward a system that achieves autonomous user acquisition and creative performance. The current architecture already orchestrates creative generation and production. The next evolution connects that to performance data and contextual signals about how players engage with playable ads. Early versions are already in development, with some components in production. The system can already identify when parts of an ad aren't working, such as a puzzle that's too difficult for players to complete, and automatically ideate simplifications for the next iteration.
"The trust we've built in Claude's capabilities and consistency has led Sett to make foundational architectural decisions based on its technology,” said Blumenfeld. “We believe that Claude is the way to win."