I built a churn prediction tool for B2B SaaS, looking for founding beta customers
Hey IH,
I've been building ChurnSense for the past few months, a churn prediction platform for B2B SaaS companies.
The problem it solves:
Most SaaS founders find out a customer is leaving when they cancel. By then it's too late. ChurnSense identifies which accounts are likely to churn 30 to 60 days before they do, so you can reach out while there's still time to save them.
How it works:
Connect your customer data (CSV upload, JS snippet, or Stripe webhook)
ChurnSense trains an XGBoost model on your specific account behavioral data, login frequency, feature usage, support tickets, NPS trends, seat utilization, payment failures
Every account gets a risk score updated nightly
When an account crosses your risk threshold, you get a Slack alert plus an AI-generated save email you can send with one click
The model is per-tenant, it learns from your own data, not someone else's. The longer you use it, the more accurate it gets as real churn outcomes feed back into retraining.
Who it's for:
B2B SaaS companies with 50 to 2,000 customers. If each customer is worth $200+/month to you, saving one account a month pays for the tool many times over.
What I'm looking for:
I'm looking for a small group of beta customers for honest feedback, and if it works for you, a short testimonial for my launch. In exchange: full access to ChurnSense (up to 200 customer accounts) for $149/month — that price stays locked in as long as you remain a subscriber. Plus a 30-day money-back guarantee once you sign up. I'll personally do white-glove onboarding, helping you set up the integration and making sure you're getting real scores within 48 hours.
Before you pay anything, I'll run a free churn-risk scan on 90 days of your client data, no cost, no obligation.
Live product: https://getchurnsense.com
If your churn rate is something you think about, drop a comment or DM me. Happy to answer anything here too.
The free 90-day scan removes a lot of validation friction. Have you tested predictions against known historical churn yet, and do the alerts identify accounts early enough for a CS team to actually intervene?
Good question, and I'll be straight with you: not against a live customer's actual churn outcomes yet. Here's why, and it's relevant to your second question too.
The model starts in a heuristic cold-start mode for a new tenant, since it needs a real sample of confirmed churns (accounts marked "churned" by the customer's own team) before it switches to training on actual outcomes rather than proxy signals. No beta tenant has hit that volume yet, so I can't honestly claim it's been validated against known churn in production.
What it's built on is the pattern behind the 30-60 day window: comparing each account against its own historical baseline (login frequency, feature usage, etc.) rather than one fixed threshold across your whole customer base, since a quiet week means something different for a brand-new account than a 2-year one. That's a real, testable signal even in cold-start mode.
This is exactly what the free 90-day scan is for; it's not a sales gimmick, it's the actual way to answer your question on your own data before you'd trust it enough to act on an alert. Happy to run one for you if you want to see it firsthand.
That customer-data validation angle is interesting. If you’re open to it, what’s the best email to reach you on?
you can reach me at founder@getchurnsense.com
Thanks! I’ve just sent it over.
Looking forward to hearing your thoughts whenever you have a chance.