Product
How approvals work in Montr BI
See how Montr BI asks before it acts. A tour of the Review queue, Tool approvals and Post approvals, plus what always needs your yes.

Montr BI’s AI can plan work, draft messages and use your tools. It does not get to do risky things on its own. This post explains how approvals work, so you know what to expect when you start.
Montr is in early access. Some modules mentioned here are in beta, and we say so where it applies.
Three places, three jobs
Anything waiting for a human yes shows up in one of three places. They have different names on purpose.
| Place | What waits there |
|---|---|
| Tool approvals (in the Agent) | Actions the AI agent wants to take: sends, workflow changes, ad drafts |
| Post approvals (in Social) | Drafted posts and the weekly Growth Journey plan |
| Review queue | Escalated conversations, held bulk-call batches, and WhatsApp template submissions that Meta is still reviewing |
The Review queue does not hold everything. Tool approvals and Post approvals are separate lists. The Home screen pulls the counts together so you see what needs you in one glance.
What always needs your approval
Some actions are on a danger list. They need your yes even when you give the agent a lot of freedom:
- WhatsApp sends.
- Voice calls.
- Campaign sends.
- Destructive CRM actions, such as deleting records.
- Anything that spends money on generation, such as images, video or audio.
The approval card shows the full action, not a summary of it. For ads and workflow changes, it shows the complete draft or a plain summary with a safe dry run.
Autonomy levels
You choose how much freedom the agent has, per mission and per brand.
| Level | What it means |
|---|---|
| Watch | Every action needs approval |
| Approval-first | The agent asks before acting |
| Mixed | Only tools on your “needs approval” list ask |
| Autopilot | Runs on its own, except the danger list |
| Autonomous | Runs on its own, except the danger list |
Whatever you choose, you can force approval on any action that changes data. Autonomy is something you grow into. Start with Watch or Approval-first, and loosen it after you have seen the agent work.
Ads are always created paused
If the agent proposes an ad campaign (Ads is in beta), the approval card shows the account, budget, targeting and creative. When you approve, the campaign is created paused. Montr only creates campaigns. You turn them on yourself in the ad platform.
Automations never switch themselves on
If the agent builds a workflow for you, it does not turn it on. Going live is a separate approval that names the trigger, the schedule and any public web address that will accept deliveries. AI-built workflows also come with a dry-run verdict first: what it would send or change, before anything runs.
Posts: chains and client links
For social posts, you can set an ordered chain of approvers, such as admin, brand manager and then client. A rejection at any stage stops the post.
Clients can approve or comment through a private link without an account. The links expire and you can revoke them. When a post is approved, its content, media and targets are frozen, and the publisher uses that exact snapshot.
The weekly plan from Growth Journeys
Growth Journeys (beta) take a goal and, every Monday, put one composed plan in Post approvals. The weekly research and drafting steps are read-only. You approve the whole week at once, approve per post, edit an item or reject the plan. An email draft in the plan becomes a task for you, and nothing is sent. A WhatsApp broadcast draft becomes an unstarted draft campaign. A plan nobody decides by the end of its week expires and cannot be approved late.
Delegate and expire
- You can approve, reject, delegate to a teammate, or decide in bulk.
- Delegated decisions are written to the audit log.
- Tool approvals expire instead of waiting forever.
- If the person who owned a paused run leaves or is suspended, that run does not resume.
Limits that apply on top
Every mission has hard limits on tool calls, tokens, credits and time. At 90% of a cap, Autopilot drops back to a more supervised mode. You can stop a mission at any time, and it stops at its next tool call. Text from outside, such as web pages, emails and customer messages, is treated as data, never as instructions.
Who should approve what
In a small team, the owner approves everything at first. As you grow, split it up.
- A sales lead handles follow-up sends and CRM changes.
- A marketing lead handles posts and campaigns.
- A finance or ops person looks at anything that spends money.
- A client signs off on posts through the private link.
You can delegate a single approval to a teammate, and delegated decisions are written to the audit log. Match the approver to the person who knows the risk best.
A short walk-through
You ask the agent to follow up with quiet deals. A few minutes later, Home shows a card in Needs you: “5 agent steps waiting”. You open Tool approvals. Each card shows a drafted WhatsApp message, the recipient and the deal. You edit one line, reject one, delegate one to a teammate who owns that account, and approve the rest. The sends go out, and each is logged on the contact.
Later that week, the Monday plan from your Growth Journey sits in Post approvals. You approve the posts for the week in one click. Nothing in the plan was sent or published before then.
That is the pattern: the AI prepares, you decide, and the record shows who decided what.
What we do not claim
- The agent does not run your business alone. Risky actions wait for you by default.
- Approving from your phone or desktop app is where we are heading. Those apps are coming soon, not released.
- Controlling the agent from WhatsApp is still being tested with real accounts, so do not rely on it yet.
Try it in order
- Open the Agent workspace and give it a small goal. Follow the guide to giving the AI agent its first goal.
- Watch the Tool approvals list while it works.
- Approve or reject one item, and read the result.
- Move to Mixed mode for actions you trust.
For the thinking behind this, read why your AI should ask before it acts. See the AI agent and social media pages.