Email that optimizes itself.
Eigen runs Thompson sampling over your variants in production — killing losers, compounding winners, and spawning new variants the moment a test reaches significance. Continuously. Bayesianly.
The optimizer, in motion.
Each send is evidence. Posteriors update. Allocation tracks Pr(variant is best). Winners compound, losers retire, and new variants spawn from the leader — automatically.
10 founding spots.
$100 each.
- •Lifetime 50% discount on whatever pricing ends up being
- •Direct Slack/email line to the founder
- •First access when MVP ships
Frequently asked.
01What if I don't have an email list yet?
Eigen is built for senders with at least a few thousand subscribers — that’s the volume needed for posteriors to separate quickly. If you’re smaller, the math still works, it’ll just take longer to converge. Reach out and we can talk through it.
02How is this different from Mailchimp's A/B testing?
Mailchimp picks a winner once, then sends the winner to everyone. Eigen never stops testing — every send is both an experiment and a delivery, and traffic is allocated continuously based on each variant’s posterior probability of being best. New variants spawn automatically when winners emerge.
03What happens after the 10 spots fill?
The founding round closes and the next tier prices significantly higher. Founding members keep their 50% lifetime discount regardless.
04What's Thompson sampling?
A bandit algorithm: at each decision, sample one conversion rate from each variant’s posterior, pick the variant with the highest sample, send. It naturally explores uncertain variants and exploits confident ones — no manual hyperparameters.
Built on Bayesian decision theory.
Each variant has a hidden conversion rate. We place a Beta(1, 1) prior on it, and after every send the posterior updates in closed form: success or failure, the parameters tick up. Beta-binomial conjugacy means there's no MCMC, no approximation, no simulation drift — just two integers per variant, updated forever.
Beta(α, β) + (s, f) → Beta(α + s, β + f)
s = successes since last update, f = failures. Two integers. That's the whole state.
Read the methodology paper →