How Velsano AI agents work
The three kinds of AI in the app — scheduled agents, on-signal agents, and on-request Copilot — and the human-in-the-loop model that keeps you in control.
How the AI works
Velsano is AI-first, but the AI never acts as a black box. Everything an agent does is explainable, attributable, and — by default — reviewable by a human before it changes anything. There are three kinds of AI in the product:
1. Scheduled autonomous agents
These run on a schedule (nightly, hourly, or every few minutes) across your whole book of business. They read signals, reason about which accounts need attention, and produce recommendations. They include the Portfolio Intelligence, Account Strategy, Predictive Intelligence (churn), Conversation Intelligence, QBR Prep, Milestone Intelligence, Renewal, Stakeholder Monitor, Expansion, and Voice of Customer agents — see *"The full agent roster."*
2. On-signal agents
These fire in response to an event — a health drop, an NPS detractor, a renewal approaching. Most of these are Signal Rules you can edit yourself (see "Signal Rules"); a few can also invoke a deeper AI agent for one account (e.g. churn triage).
3. On-request Copilot
These only run when *you* ask — an account brief before a call, a drafted email, a churn narrative. Nothing happens until you click. See the AI Copilot category.
The human-in-the-loop model
When an agent decides an account needs action, it doesn't silently change your data. It writes a recommendation — usually into the AI Agent Inbox as an approve/reject item carrying a *reasoning trace* (the evidence and logic behind the suggestion), or as a Risk/Renewal/Expansion Action in your work queue. An agent only auto-executes when your team's policy explicitly allows it; otherwise a human approves first.
The controls you have
- Per-tenant AI switch — a workspace-wide master toggle. If AI is off, agents fall back to rule-based logic or simply don't run.
- Per-agent enable/disable + cooldown — turn any individual agent on or off, and set how often it may act on the same account (Admin Console → AI Copilot & Agents → Agents).
- Autonomy — whether an agent needs approval, queues customer comms, or runs fully automatically follows the account's coverage model (see *"Autonomy & work styles"*).
- Token quota — a per-workspace monthly budget for AI usage, visible in the AI Copilot config.
If AI is unavailable for any reason, features degrade gracefully — you get rule-based output or a clear "AI is off" state, never a broken screen.
Related: *The full agent roster* · *Autonomy & work styles* · *The AI Agent Inbox* · *AI Copilot: what it is and where to find it*