# Workflow Automation: AI under guardrails | Vertex

**Source:** https://vertexcrm.io/features/automation  
**Summary:** Workflow automation with AI under guardrails: steps that do real work, approvals on money and bulk actions, every run audited. Pay per task.  
**Publisher:** Vertex

---

Workflow Automation

## Automation that thinks, and knows when to ask.

Connect visual nodes (trigger → audience → action) and drop in AI nodes for the judgment calls. Safe steps run on their own; anything that touches money or moves in bulk waits in an approval queue for one click from you.

How it works

## Trigger. Audience. Action.

Three node types, one canvas. Add an AI node anywhere you'd otherwise need a human to decide, and let the guardrails handle the rest.

### Trigger

Start from any event on the shared database: a new lead, cancel intent, a quota hit, a stage change, a payment, an inbound message or a schedule.

### Audience

Filter to exactly who should flow through: plan, region, lifetime, last activity, score. The segment recalculates live as data changes.

### Action

Send an email or SMS, draft an offer or order, move a deal, create a task, hit a webhook, or hand it to an AI node to decide and draft.

AI nodes

## Judgment, with a safety net.

An AI node does the thinking a rule can't: picking the right offer, writing the message, choosing who to skip. It runs under the same guardrails as the assistant: safe actions fire automatically, money and bulk actions are drafted and held for approval.

- Safe, read-style actions execute on their own
- Money and bulk changes wait in an approval queue
- Scoped to the workflow owner's permissions
- Every action audited: who, what, when, fully traceable

Approval queue

## You stay in the loop, not in the weeds.

Everything a workflow drafts that needs a human lands in one queue. See the exact values, the recipients and the cost before you approve, then watch it run. No surprise charges, no rogue bulk sends.

- Every risky action collected in a single review queue
- See impact and cost before you commit a thing
- Approve, edit or reject: the decision is audited
- Safe steps never queue; they just run

Put it to work

## Workflows that pay for themselves.

### Win back churn

Cancel intent → match plan & tenure → AI drafts a save offer → approve → send email and SMS. Recover revenue automatically.

### Nurture new leads

New lead → segment by source → AI writes a 3-step sequence → safe sends fire automatically, no approval needed.

### Upsell at the quota

Tenant hits a plan limit → AI offers the next tier in-app → upgrade drafted → one approval locks the new plan.

### Revive stuck deals

Deal idle 14 days → AI books a check-in slot → drafts the invite → owner approves and it goes out.

### Chase failed payments

Payment fails → smart dunning sequence → AI drafts a personal nudge for high-value accounts, held for approval.

### Watch the funnel

Leads drop while traffic holds → AI flags the capture gap → drafts the fix → routes it to you to ship.

Better together

## Automation everywhere your data lives.

### AI Assistant & Autopilot

The same agentic engine that powers the chat box powers your AI nodes, drafting actions and waiting for approval on the risky ones.

### Marketing Analytics

Trigger workflows from real funnel events and watch recovered revenue land back in the same analytics that fired them.

## Questions, answered.

**Q: How do AI nodes stay safe inside a workflow?**

AI nodes run under the same guardrails as the assistant. Safe, read-style actions execute automatically; anything involving money or bulk changes is drafted and held in an approval queue until a person clicks approve.

**Q: What can trigger a workflow?**

Any event on the shared database: a new lead, a form submission, cancel intent, a quota hit, a stage change, a payment, an inbound message, or a schedule. Triggers feed an audience step, then one or more actions.

**Q: Do I have to write code?**

No. Workflows are visual nodes you connect (trigger, audience, action) with optional AI nodes for the judgment calls. You can also ask the assistant to build a workflow for you.

**Q: What gets audited?**

Every action a workflow takes is written to the audit log with who, what and when. Approvals and the person who granted them are recorded too, so a workflow is fully traceable after the fact.

## Set it up once. Let it run.

Build a workflow in minutes, drop in an AI node, and keep one hand on the wheel with the approval queue.
