How the AI is governed
Last updated 2 August 2026
AI that talks to your customers needs a governance model, not a disclaimer. This is ours.
This page is maintained by Inifye Tech Ltd to answer common questions about how Sales Intelligence handles sales data. It describes our current practices — it is not an independent audit or certification.
Autonomy levels
| Level | What the agent does | Typical use |
|---|---|---|
| 1 | Observes and suggests only | New motions being evaluated |
| 2 | Drafts, a human sends | Sensitive accounts and high-value deals |
| 3 | Acts, with approval required before external actions | Most outbound motions at launch |
| 4 | Acts automatically inside defined limits, exceptions escalate | Proven inbound response motions |
| 5 | Acts automatically with post-hoc review | Low-risk routine follow-up and admin |
Human control
- Autonomy is set per agent, per motion, and can be lowered instantly.
- An approval queue holds anything awaiting a human, with the draft and the reasoning attached.
- A single control pauses all outbound AI activity in a workspace.
Grounding and accuracy
- Answers are drawn from your approved content, product data and records — not from open-ended generation.
- When confidence is low, the agent escalates instead of guessing.
- Pricing, legal and contractual statements are restricted to approved sources.
Traceability
Every agent run is traced: model calls, tools used, records retrieved, policy checks, handoffs, cost and duration. Audit entries are immutable and carry a content hash, so what was sent can always be reconstructed.
Model use and training
Your business data is not used to train foundation models. Models are selected per task, and model choice is recorded in the trace so behaviour changes can be explained.
Bias and fairness
Scoring uses commercial signals — fit, engagement, purchase history, payment behaviour. Protected characteristics are not scoring inputs. Score drivers are shown on the record so a human can challenge them.
Incidents
If an agent does something it should not, pause the motion, review the trace, and tell us at oracle@inifye.tech. We treat agent-behaviour incidents with the same seriousness as security incidents.
Inifye Tech Ltd
Global HQ
oracle@inifye.tech