Surprisal Labs

Open-ended discovery at enterprise scale

Contact
Live with first design partners
> scroll to continue_

Can you search for something you don’t know exists?

Your data already holds the answer: a new fraud ring, a shift in how customers behave, a treatment that works better than expected. But rules, dashboards and analysts only look where someone points them. Surprisal looks everywhere else.

×
Discovery

A full discovery loop: from hypothesis to evidence

What it found when nobody pointed it anywhere

Customer behaviorHow is user behavior changing?
Users who stop using search are 3× more likely to churn.
Fraud & riskWhat threats are we missing?
40% of suspicious accounts share just 12 devices.
OperationsWhere are things going wrong?
Failed payments spike 4× after a specific API timeout.
HealthcareWhat changes the outcome?
A history of failed antidepressants doubles TMS appeal wins, from 36% to 74%.
ComplianceWhat risks are emerging?
30% of high-risk transfers bypass existing monitoring rules.

The research behind it

Fig. 1 Evidence rises past 1/α, then decays: the finding expires.
NeurIPS 2026 workshop on E-values

Discoveries Have Lifetimes: E-Values for Autonomous Discovery in Adversarial Environments

Patterns in data expire. We detect when they do.

P. Goldman-Kalaydin, V. ShebanovRead on alphaxiv →
Fig. 2 Most results stay in the band; the one that departs is worth a look.
Preprint

Do Not Forget Normal Science: Designing Discovery Loops in Adversarial Domains

Why discovery loops fail when the data fights back, and how to build ones that don’t.

Surprisal LabsComing soon
End of this run
Team
CEOPavel Goldman-Kalaydinpavel@surprisal.io LinkedIn ↗
CTOViacheslav Shebanovvs@surprisal.io LinkedIn ↗
Where
Paris Berlin SF
Next step
 Taking on a few more partners.
Run the loop on your own data, before anyone else does.
Run a loop on your data
Surprisal Labs