Security and reliability
Leave the key decisions to people, and keep a record of every step.
- 01AI suggests
- 02A person confirms
- 03The system acts
- 04Everything is logged
01
A human in the loop
- Writing or changing business data, approvals, sending messages outside, anything involving payment, and deletion all need a person to confirm before they run.
- AI gives its reasons with every suggestion, and a person decides whether to take it.
02
Permissions and data
- Only the minimum permissions a task needs.
- Data separated by tenant and role.
- Content sent to a model is de-identified first.
- Where data lives and which models are used are set by the client, including on-premises or private deployment.
03
Traceable and reversible
- An audit trail of who did what and when, what AI suggested and on what basis.
- Key data keeps its versions and can be rolled back.
04
Quality assurance
- AI writes the code, people review it: every change has automated tests and code review.
- AI features are checked against an evaluation set before launch.
- Delivery comes with tests and documentation.
05
When AI is unsure
- Answers cite their sources so they are easy to check.
- If it doesn’t know, it says so.
- Anything beyond its ability goes to a person.
Curious which work AI could take on?
Tell us where your team spends the most time and effort, and we’ll work out together whether it’s worth doing and how.