The Problem
According to our on-chain analysis, prediction markets see 60-70% weekly user churn, and it's not improving. Polymarket averages 62% weekly churn; prediction market apps hit 89% monthly churn in early 2026.
Polymarket saw a 57% decline in monthly active users in 2025 (454K to 193K), and only 12.7% of wallets are profitable. Crypto apps see D30 retention as low as 2-3%, while fintech apps average 11.6%.
The reason is that discovery is still broken. Users can't find markets that matter to them and without any relevant content when opening the app, users don't come back.
We need product surfaces that make the hundreds of prediction markets easy to navigate. That's what we enable.
Introducing Embed’s Prediction Market Recommendations API
This month we released our Prediction Market recommendations API to a select few design partners. Inside is our first generation of AI models trained on Polymarket data, soon expanding to other prediction market sources.
Our core differentiator is personalization. Most prediction markets platforms today show trending markets. We're enabling personalized search and ranking. Markets tailored to each wallet based on their onchain history and engagement patterns.
Similar to X or TikTok, user A is into sports and therefore they'll see relevant sports betting markets. User B sees crypto price predictions. User C sees political markets. All based on what their wallet activity tells our models about their interests.
For prediction market platforms looking to improve retention without building ML infrastructure from scratch.
Try the console → console.mbd.xyz
How It Works
❜embed's best feed-building features are now available for prediction markets:
Search and filter markets, immediately personalized based on user tags and AI labels.
Rank results using our scoring algorithms - trending, interest-based, or custom weighted.
Early Results
For context, AI personalization in sports betting drives 20-25% higher bet placement (Sportradar 2025), and DraftKings saw 18% retention lift from predictive targeting.
Our results:
Latest backtests on prediction market data: Our AI models show 10X trading volume improvement and 3X user activation
Live in production: Personalized social feeds increase transactions by 5X in Base app, our partner since early 2025
Why Embed?
3 years ago we saw web3 data growing exponentially and recognized the opportunity to bring web2-quality personalization to onchain experiences.
We started with Farcaster, working with projects like Neynar and Base app, serving personalized recommendations for posts and mini apps. We built infrastructure that scales across assets and data sources.
Now we're bringing that same personalization engine to prediction markets because web3 UX won't match web2 until we can recommend onchain assets and activities with the same quality users expect from TikTok or Spotify.
Building personalization in-house is hard and costly. Cold start problems, spam filtering, sub-250ms response times, 99.9% uptime. These are table stakes, and getting there takes months and a dedicated ML team.
❜embed is web3's infrastructure for personalization: a self-serve platform for AI models, extensively trained on web3 data of all kinds. The aim: personalized recommendations to any wallet, for any kind of onchain asset.
With Robinhood, DraftKings, and Coinbase all entering prediction markets in 2025-26, the window to establish personalization infrastructure is 12-18 months before the market commoditizes.
Request a demo → https://getembed.ai/contact
Sources
Daily, Weekly, Monthly Churn: Dune Dashboard
Polymarket MAU decline (454K→193K): Fortune
Wallet profitability (12.7%): Cointelegraph
Fintech D30 retention (11.6%): Plotline
General app D30 benchmarks: AppsFlyer
Sports betting personalization (20-25% lift): Sportradar
DraftKings retention improvement: PYMNTS
Competitive landscape: CoinDesk