AI approval policy template for sales

A single on-off toggle for AI forces one decision to cover a hundred different risks. Logging a call and sending an email to a customer are not the same kind of action, and they shouldn't share a switch.

A policy table splits them by class. Low-stakes, reversible, internal work runs on its own. Anything a customer sees waits. Anything that spends money never runs. That structure is worth copying whether or not you use SalesCrew, so the table is published rather than described.

An AI approval queue is one list where every action waiting on a person sits until someone approves it. In SalesCrew the queue is grouped by agent. Each item shows the draft itself, the evidence behind it (the thread, the score, the knowledge-base sources it used), the model's confidence and what the run cost.

You have three buttons: Approve, Edit-then-approve and Reject. A rejection asks for a reason, and that reason feeds the agent's next run, so the same mistake doesn't come back tomorrow. The same queue is reachable over MCP through approval_list, approval_approve and approval_reject, so a reviewer working in Claude or ChatGPT clears the same items you'd see on screen.

Mode is only half of it. Guardrails apply regardless of what mode an agent is in: at least three days between touches per contact, a per-cadence cap, per-mailbox and per-domain daily caps, a business-hours send window, suppression re-checked at send time, and a confidence threshold that drops an action back into review when the model is unsure.

Above all of it sits the kill switch: one control on the Agents page that stops every agent at once, with an MCP twin so it can be hit from a script as well as a screen. Whoever owns that switch is a decision to make at setup rather than discover in an incident.

Take the three columns and apply them to whatever automation you run. Write out every action your system can take, sort each into auto, review or human-only, and be honest about which ones a customer sees. If your human-only list is empty, look again at anything that spends money.

Then add the part people skip: a record of what happened. Every tool call in SalesCrew is written to the audit log with the actor named as a user, a token or an agent. Without that, a policy table is a statement of intent rather than something you can check after the fact.

The table tells a tool what to do. Your team also needs the rule in words, so here is the policy as a short document. Change the names and the numbers, then keep it next to your sales playbook.

1. Scope. This policy covers every AI agent that reads or writes our CRM, our inboxes or our outreach tools.
2. Default mode. A new agent starts in draft. It moves to review after a named person has read its first 20 drafts.
3. External sends. Any email, LinkedIn message, SMS or proposal a prospect or client will see waits in the approval queue until a person approves it.
4. Internal work. Research, enrichment, tagging, logging, scoring, summaries and reply labels run without approval, and every one is logged.
5. Pipeline changes. Stage changes, merges and deletes need approval. New deals and booked meetings need approval for the first clean week, then run on their own.
6. Money. No agent spends money or acts outside the CRM, in any mode.
7. Rejections. Every rejection carries a one-line reason, so the agent can learn from it on its next run.
8. Stop. One named owner holds the kill switch, and their name sits at the top of this policy. Anyone who sees an agent misbehave tells them at once, and they stop every agent first and ask questions after.
9. Review. Once a month, the owner reads the audit log and decides whether any action class moves between review and auto.