What to look for
| Criterion | Why it matters | How Horizon does it |
|---|---|---|
| Evidence behind every flag | Account owners act on a risk they can check | Every risk signal cites the call, ticket, email or usage report it came from, with the date |
| Reads calls, tickets and email | Champion changes and budget cuts surface in conversations first | Gong and Google Meet calls, Zendesk tickets, ServiceNow incidents, Jira bugs, Gmail and product usage |
| Testable on your history | You should know how it would have scored last year's renewals | Admins can backtest the score against past churn and compare it with the score you use today |
| Honest about thin data | An account with no data shouldn't look safe | Accounts with evidence behind less than 60% of the score's weight show "Insufficient data" |
| Review before customer emails | A wrong guess shouldn't reach a customer | Horizon drafts the email; a CSM reviews it and sends it from their own Gmail. Nothing sends on its own |
| Impact measured against a holdout | A saved renewal might have renewed anyway | A random holdout of up to 20% stays on your normal process, and the report shows the lift |
| Time to first value and price | You should see results on your own accounts, at a price you know up front | A free 30-day pilot on your real accounts. Professional is $5k a year for up to 250 accounts, and pricing is published |
Approaches compared
| Approach | Good for | Watch out for |
|---|---|---|
| Spreadsheets | A cheap first version with logic you control | Hand-typed fields go stale, they can't read calls or tickets, and nobody sees why a score moved |
| General CS platforms | Large teams that want one system for the whole customer lifecycle | A bigger rollout, and someone to configure and maintain the scoring |
| Focused renewal-risk tools | Teams that want risk found, acted on and measured without a platform project | A narrower scope than a full CS platform |
For side-by-side detail, see how Horizon compares, including Horizon vs Staircase AI and Horizon vs Hook. Each comparison is sourced from the vendor's own pages.
How Horizon does it
Horizon reads each account's calls, tickets, Jira issues and email from your integrations at least once every 30 days, and sooner when new evidence arrives, and raises a renewal risk signal for each problem it can back with dated evidence. The signals roll up into a customer health score from 0 to 100, and each risk comes with a playbook for the account owner.
Read churn prediction for B2B SaaS for how to test any model, see what renewal risk means, or work out what churn costs you with the cost of churn calculator.
Questions
How accurate is churn prediction software?
It depends on your data, so test it on your own history. Run it on last year's renewals as they looked beforehand, and check how much churned ARR it flagged and how early.
Does Horizon predict churn with AI or with rules?
Both. AI reads the conversations and must cite dated evidence, or the signal is thrown away. The score is plain rules, with no AI in the arithmetic.
What data do we need?
Salesforce, which gives Horizon its account list, plus the calls, tickets, bugs, email and usage you connect. Free covers Salesforce plus one more source.
How much does churn prediction software cost with Horizon?
Professional is $5,000 a year for up to 250 accounts, with unlimited users and AI included. Free is $0 on your own Anthropic or OpenAI key, for up to 25 accounts and 2 users. Larger books are on the pricing page.
Can we try it on our own accounts?
Yes. The 30-day pilot is free and runs on your real accounts, set up with you on a call. If you don't continue, you move to Free.
More on this topic: churn, definitions, guides and tools.
Book a 30-minute call
Pick a time below. We start on a demo book or your own top 25 accounts, and can set up your free 30-day pilot on the call.
Calendar not loading? Open it in a new tab.
Updated November 3, 2026