What a risk signal is
A signal is a dated, evidence-backed observation about one account. Each one has:
- A type: High risk (severity 7 or more), Watchlist (below 7), Upsell likely (an opportunity), or Fact (context that doesn't count against the score).
- A severity from 1 to 10, reflecting what's at stake, how strong the evidence is, the account's size in your book and how close its renewal is.
- A category, such as stakeholder, budget, adoption, product, competitive, sentiment or commercial (13 in total).
- Citations: the call, ticket, email or usage report it came from, with the date and a link to open it in the source system.
Three ways signals are found
1. Serious single events
A blocker or critical Jira issue, a high-concern inbound email, or an escalation alert raises a signal on its own.
2. Hourly checks, no AI
Some risks are plain facts. Horizon checks them hourly: a churn or contraction forecast from the CSM, a "canceling" subscription status, call flags (renewal risk, seat contraction, champion leaving, competitor mentioned), a paid product with less than 10% of its seats in use for a month, an account that went silent, and renewals coming due.
3. AI that reads the whole account
Horizon reads an account's tickets, Jira issues, incidents, calls and email together and raises the signals a person would spot reading all of it. Every account is read at least once every 30 days, and sooner when new evidence arrives.
Escalations caught as they happen
Every new conversation (tickets, incidents, Gong and Google Meet calls, inbound email) is scored as it arrives. Horizon scores each one on activity (reply volume, open cases, executives copied, reopens, missed SLAs) and, with AI, on tone (frustration, churn intent, legal risk). Above your threshold it posts a Slack alert with Acknowledge, Assign to me and Resolve buttons, and raises a high-risk signal on the account.
People stay in charge
Horizon can suggest that a risk looks resolved. Its AI never resolves a risk on its own. When a CSM dismisses a signal they give a reason, which is how Horizon measures its own mistakes.
Questions
Where do risk signals come from?
Gong and Google Meet calls, Zendesk tickets, ServiceNow incidents, Jira bugs, Gmail and email logged in Salesforce, and product usage. Each signal cites the call, ticket, email or usage report it came from, with the date and a link to open it in the source system. See integrations.
How quickly does a new risk show up?
Every new conversation is checked for escalation risk as it arrives. Gmail is read every 5 minutes, Gong every 15 and Zendesk, Jira, ServiceNow and Google Meet every 30. Plain facts, such as a cancellation forecast, are checked hourly, and the AI reads every account in full at least once every 30 days, sooner when new evidence arrives.
Can the AI make a signal up?
No quote, no signal. The AI has to cite dated evidence from the account, or the draft signal is thrown away. Duplicates are judged by the evidence they cite, so one problem doesn't become three signals.
What's the difference between High risk and Watchlist?
Severity. A signal with a severity of 7 or more is High risk; below 7 it goes on the Watchlist. Upsell likely marks an opportunity, and Fact is context that doesn't count against the score.
Does Horizon act on a risk without a person?
No. It can suggest that a risk looks resolved, and it never resolves one on its own. Outreach is drafted for a CSM to review and send, so nothing reaches a customer without a person. When a CSM dismisses a signal they give a reason, which is how Horizon measures its own mistakes.
Are risk signals on the Free plan?
Yes. Free covers up to 25 accounts and 2 users on your own Anthropic or OpenAI key, with Salesforce plus one more source, and every signal comes with its evidence. Professional is $5,000 a year for up to 250 accounts, with unlimited users and AI included. See pricing.
More on this topic: churn, definitions, guides and tools.
Updated November 3, 2026