About
Why we built it
Maya Kessler spent several years running customer success and then retention at two subscription software companies. Every quarter ended with a postmortem. A customer had cancelled, or an account had contracted down, and the team would walk through what the data had been saying for weeks before the call came. The signals were always there. The warning always arrived too late.
Daniel Okafor had spent nine years shipping machine learning into production, from fraud scoring to ranking systems. The behavioral pattern of a subscription account heading toward cancellation, falling logins, shrinking feature use, unpaid invoices, is not a hard problem for a well-built model. It is a logistics problem: getting the score in front of the customer-success lead while there is still time to act.
They started Churnora in September 2023 and have worked on it full time since. The product has one job: know who is about to leave, why, and what to do, before the cancellation email arrives. The model is a calibrated logistic regression over eight behavioral features, chosen because it is inspectable. A language model translates the factors into plain language and drafts the save play. The two pieces stay separate because prediction is a math job, not a language job.
Churnora scores an estimate, not a certainty. A quiet account might be on vacation. You know things the model cannot see. The product says so wherever a score appears, and the factors panel shows exactly which numbers in that account's own data moved the score, so you can agree or disagree.
The team
Five of them are below. Everyone is based in the United States and nobody here has a second job. Churnora is what we do.
Co-founder and CEO
Austin, TX
Ran customer success and then retention at two subscription companies, where every quarter ended with a churn postmortem that the data had predicted weeks earlier. She started Churnora so the warning arrives while there is still time to act, and works on it full time.
Co-founder and CTO
Austin, TX
Nine years shipping machine learning into production, from fraud scoring to ranking systems. He owns the behavioral churn model, the feature pipeline behind it, and the evaluation set every change to the scorer is measured against. Full time on Churnora.
Head of ML
Austin, TX
Builds the engineered features the model scores from: login recency, usage trend, ticket load, payment health, tenure. Her rule is that every factor the product shows a human must trace back to a number in that customer's own data.
Founding Engineer
Denver, CO
Works on the parts nobody sees: the import pipeline that accepts a messy CSV without complaint, the scoring run that finishes before the page loads, and the export that gives a team back everything they put in.
Design
Portland, OR
Designs the risk table, which is the whole product. Her rule: a customer-success lead should see who is slipping, why, and what to do first, within a few seconds of opening the app.
The people and investors named on this page are illustrative. Churnora is a demonstration company and we would rather say so plainly than have you find out later. Everything about the product is real: the behavioral model, the scoring engine, and the free tier all work exactly as described.
The company
Backed by
We raised $850,000 at Pre-Seed in April 2024, led by Harbor Line Ventures, and the money is spent on building the product rather than on advertising it.
Harbor Line Ventures
Lead investor, Pre-Seed
An early-stage fund that backs applied machine learning for B2B software.
Elena Vasquez
Angel
Former VP of customer success at a public subscription software company.
Tom Berrigan
Angel
Founded and sold a revenue-analytics product for SaaS finance teams.