Building an AI Meta Ads copilot -- launching today, would love feedback

I've been building Campaiyn for the past couple of months. We launched publicly in April 2026 and I wanted to share the journey, what we got wrong, and what we're seeing now that it's in front of real users.

The problem we're solving

Most small businesses and solo founders running Meta Ads are flying blind. They know their ROAS could be better. They read the blog posts. They watch the YouTube videos. But actually diagnosing their specific account's problems requires either paying an agency ($2-5k/mo) or becoming a full-time media buyer yourself.

There's a huge gap between "I have budget to spend on Meta Ads" and "I know how to optimize Meta Ads." Campaiyn closes that gap.

What we built

Connect your Meta account. Get an instant 0-100 health score and a prioritized list of what's wrong. The AI reads your campaigns the way a senior media buyer would -- looking at structure, audience configuration, creative fatigue, Pixel health, and budget allocation.

The free tier gives you the health score and the top 3 issues. The paid tier unlocks the full recommendation engine, ongoing monitoring, and an AI chat where you can ask questions about your account.

What we got wrong early on

Our first version tried to do everything. Campaign builder, audience research, creative suggestions, performance forecasting. It was slow, confusing, and users had no idea where to start.

We stripped it down to one thing: "tell me what's broken and what to fix first." Retention went up immediately.

Where we are now

Live at campaiyn.ai. Meta App Review passed, production OAuth is open to the public. The health scoring is calibrated and working on real accounts. The recommendations engine produces actionable output -- not generic advice, but specific fixes for your specific account.

What I'd love from this community

If you're running Meta Ads (even a small budget), I'd love you to test it: campaiyn.ai

Especially useful:

• Does the health score feel accurate to you?
• Are the recommendations actually actionable, or do they feel generic?
• What's missing that would make this genuinely useful for your workflow?

Happy to answer any questions about the build, the stack (Next.js, Convex, Claude API), or the Meta API integration. This community has been incredibly helpful throughout the journey.