Magic Improve studies your product data, tells you which experiments to run next, and remembers everything you learn.
PromoUp to 50% off your first year
What Magic Improve shows you, run by run.
Your Product
Mon 08:00 1. It studies your numbers
This week’s analysis
A fresh look at your data and everything you already know.
What it read
Product analytics, last 30 days 
14 past A/B tests and their results Goal: more paying customers
North starTrial to paid
4.2%
Guardrails RefundsSupport tickets
ICE 8.0
Mon 08:04 2. It ranks what to test next
4 new opportunities
Ranked by impact, confidence and effort.
Cut onboarding from 5 steps to 3
ICE 8.0 · Impact 4 · Confidence 4 · Effort 2
Track why trials are cancelled
ICE 7.5 · Impact 3 · Confidence 5 · Effort 2
Show yearly savings on the pricing page
ICE 6.0 · Impact 4 · Confidence 3 · Effort 2
Find where invited teammates drop off
ICE 4.0 · Impact 4 · Confidence 3 · Effort 3
8.0
Most new teams stop at step 3 of onboarding. Teams that finish it are far more likely to pay.
AI-generated · your team decides what runs
Mon 09:15 3. Your agent builds it
Memory
Carried into every run
A yearly plan banner at checkout lost to the original.
No pricing changes until January.
Enterprise trials are handled by sales.
Ideas you already tried won’t come back.
Claude Code your-product > Build the top experiment from Magic Improve
magic-improve · Read brief
Cut onboarding from 5 steps to 3 · ICE 8.0
magic-improve · Read memory
1 learning and 1 decision apply
Adding a 3-step variant behind your feature flag
✓ Variant ready for your team to review
magic-improve · Log decision
Marked as testing
A growth team’s judgment, on every run.
Scored by impact, confidence and effort.
One north star metric per goal. No vanity wins.
Each recommendation shows the data behind it.
Decisions and learnings carry into every run.
Or hand it straight to your coding agent.
It proposes. You decide what runs.
Recommendations are AI-generated. Your team decides what to run.
It reads your data where it already lives.
Where your numbers come from
PostHog 
Mixpanel 
Amplitude 
Google Analytics Past tests and their results
GrowthBook 
Statsig 
Optimizely 
LaunchDarkly Connected over MCP
Claude Code 
Codex 
Cursor New ideas, where you work
Slack 
Microsoft Teams 
Email Optional: read-only access to your code, so ideas fit how your product is built.
MCP
Your agent reads what to test, builds it in your code, and logs the decision back.
Experiment brief
Cut onboarding from 5 steps to 3
Claude Code Codex Cursor Works with any MCP client.
Monthly prices in euro.
Join the waitlist: up to 50% off your first year.
On paid plans.
Try it on one product
€0
For teams testing every week
€99 /month
For several products
Custom
Promo confirmed at setup.
Something we didn’t cover? Write to hello@magic-team.com.
No. It doesn’t write code or open pull requests. It proposes; your team or your coding agents build.
Your product analytics and your A/B testing tool. Access to your code is optional and read-only.
Your analytics shows what happened. Magic Improve tells you what to try next, and why.
Every decision, result and note goes into its memory, and into the next analysis.
Yes. It’s AI. Every idea shows its evidence and confidence, and your team decides what runs.
Up to 50% off your first year on a paid plan, confirmed at setup.
Up to 50% off your first year.
