AI Governance and Usage Limits
Who can use the AI features, what they can reach, how actions are confirmed, the monthly usage allowance and where your data goes.
Who has AI
The AI features (the Assistant, drafted replies, summaries and the customer chatbot) are part of the Enterprise plan. On other plans they're switched off, and nothing is sent to an AI provider.
What the assistant can reach
The assistant works as the person using it. It can only see and act on the modules that person already has access to.
For example, if someone can't open payroll, the assistant can't read payroll for them. You control what the assistant can reach by setting people's roles, the same way you control what they can see.
Actions always need confirming
When you ask the assistant to change something, it first shows exactly what it's about to do. Nothing happens until you confirm.
- It can create and update records. It can't delete anything, email customers, or move money.
- Once you confirm, the change goes through the same permission checks and validation as if you'd made it on the screen yourself. It's recorded in the audit trail under your name.
See Asking the Assistant to Do Things.
Monthly usage allowance
Each tenant has a monthly allowance of AI usage, measured in tokens, which are roughly pieces of words. It's currently 500,000 tokens a month and is shared by all your AI features.
- It resets on the 1st of each month.
- Before each AI request, Plenix sets aside enough of the allowance for it, then adjusts to what was actually used.
- Once the allowance is used up, AI requests fail with "AI monthly token budget exceeded" until the 1st. Everything else in Plenix carries on as normal.
Plenix monitors usage across tenants. If you regularly reach the limit, contact Plenix to talk about your allowance. You can't see a usage meter in your tenant at the moment.
Where your data goes
When an AI feature runs, the content it needs, such as a ticket's text to summarise it, is sent to Anthropic, the AI provider Plenix uses. Anthropic is listed as a sub-processor, used only where AI features are enabled. See the sub-processor list and the Data Processing Addendum for the terms that cover it.
What is sent is the content for the request you made, drawn only from records you can already see.
Tips for sensible use
- Summarise long threads rather than short ones. That's where you save the most time.
- Read drafted replies before sending them. The assistant doesn't send anything itself.
- If your team shouldn't use AI on a particular module, remove that module from their role.
Where to go next
Was this article helpful?