---
title: "AI Agent Controls — Approved Actions"
description: "You approve what the agent can say and do, from approved actions and escalation rules to draft-only mode for sensitive flows. Every action is logged to your helpdesk."
canonical_url: https://www.getadelante.com/product/control
type: product
---

# Autonomy, within limits you set.

Adelante gives ecommerce support teams control over their AI support agent. You approve what it can say and do and set escalation rules, and every action is logged to your helpdesk.

An agent that can act on your store must also know where to stop. You approve what the agent can say and do before launch, and every action it takes is logged to your helpdesk with the full story attached.

[Get your agent](https://www.getadelante.com/demo)

Product pages: [Overview](https://www.getadelante.com/product/overview) · [Launch](https://www.getadelante.com/product/launch) · [Control](https://www.getadelante.com/product/control) · [Improve](https://www.getadelante.com/product/improve)

---

## The control layer

Control means boundaries designed into the agent during launch and enforced on every conversation, with no dashboard to babysit.

- **Approved actions:** The agent can only take actions you've approved, such as refunds within limits, address changes before fulfillment, and exchanges by policy
- **Escalation rules:** Refund thresholds, VIP customers, fraud signals, and policy exceptions route to a human automatically
- **Draft-only mode:** For sensitive flows, the agent prepares the response or action and a human approves it before anything happens
- **Helpdesk logging:** Every conversation, decision, and action is logged to your helpdesk; autonomy without an audit trail is a liability
- **Handoff with context:** When a human steps in, the request, checked context, and suggested next step are already attached
- **Scoped access:** The agent connects with workflow-scoped permissions, reading and acting only where the workflow requires

---

## When a human should decide, a human decides

The agent knows the difference between a repeatable decision and a judgment call. It finishes the first kind inside your policies and hands the second kind to your team in the same thread, with the full story and no rebuilding context.

---

## Trust, documented

On the launch call we map which decisions are repeatable, which need judgment, and where the agent must stop. SOC 2 Type II and GDPR documentation is included. See [getadelante.com/security](https://www.getadelante.com/security).

[Get your agent](https://www.getadelante.com/demo) · [How it improves](https://www.getadelante.com/product/improve)

---

## Frequently asked questions

### Can an AI support agent issue refunds without asking us?

Only if you approve that action, and only within the limits you set. Refunds above your threshold, VIP customers, fraud signals, and policy exceptions route to a human automatically, and sensitive flows can run in draft-only mode.

### How do I stop an AI agent from making up answers about our policies?

Scope it. Adelante's agent works from your policies, macros, and live order data, and it can only take the actions you approve. Sensitive flows can start in draft-only mode, and every correction your team makes feeds the QA loop.

### What happens when a customer asks for something outside our policy?

The agent hands the ticket to your team in the same thread, with the request, the checked order context, and a suggested next step attached, so nobody rebuilds the story.

### Is Adelante SOC 2 and GDPR compliant?

Yes. Adelante is SOC 2 Type II and GDPR compliant, connects with workflow-scoped permissions, and provides security documentation and DPA options for your review.

### Can we see what the AI did on every ticket?

Yes. Every conversation, decision, and action is logged to your helpdesk with the full story attached.