What should AI be allowed to change in your CRM?
by Banjo Adesuyi
Short answer: let AI update facts it can verify from a call or email, like notes, contacts and dates. Keep a person approving anything a customer will see, and anything that changes your forecast, until you trust it. Then widen the rules step by step.
A simple way to split the work
| Type of change | Examples | Who should decide |
|---|---|---|
| Facts from the call | Notes, new contacts, close dates, next meeting | AI can do it. Keep a record you can check |
| Judgement calls | Stage moves, marking a deal at risk | AI suggests, you say yes |
| Anything a customer sees | Follow-ups, proposals, invites | You approve, at least at first |
| Your biggest deals | Anything on a deal that matters to the quarter | You, always |
Why automatic updates can go wrong
Notetakers and CRMs now write to records after a call, and some only suggest changes for you to approve. Writing straight to the record is faster, but a misheard name or a wrong amount can sit there unnoticed. Three things reduce that risk:
- Show the source: you should be able to see which call or email a change came from.
- Make it reversible: an undo matters more than a perfect first try.
- Flag what's uncertain: a good tool asks when it isn't sure.
How to widen trust step by step
- Start with everything drafted. Approve each follow-up and read the edits you make.
- Let early-stage follow-ups go on their own once you've stopped changing them.
- Set rules it can't break, like never discounting, and keep your biggest deals with you.
- Check back monthly. Read a sample of what went out, and tighten or loosen from there.
How Viro handles it
By default, nothing goes to a prospect until you say "send it", and stage moves come to you for a yes. Pro is draft-only, Growth can send early-stage follow-ups on their own, and Max adds full autonomy within your rules. Read how autonomy works, or our AI Policy.
If you're still comparing tools, see AI notetaker vs CRM.
