Control and assurance · Australian teams
Human in the loop AI
Design review points that give authorised people the evidence and control needed to make a real decision.
Understand the service
What this means in practice
Human in the loop AI places people inside a workflow at a meaningful decision point. A useful checkpoint shows the proposed action, the evidence, the consequences and the alternatives. A button labelled approve is weak control if the reviewer cannot see what will happen or if the action has already occurred.
Temrik can scope an approval model for a defined process: which actions can proceed under rules, which require review and which remain prohibited. The review workload matters too. If every trivial event demands approval, staff may stop examining the decisions that deserve attention.
- Microsoft: approval workflows Human approval can be part of an automated process.
- NIST: Generative AI Profile Guidance on generative AI risks and controls.
A practical workflow example
A proposed assistant prepares a supplier email containing revised delivery terms. The reviewer sees the source request, recipients, attachment list and exact outgoing text. Changing the message invalidates the earlier approval. A rejection returns the draft for correction, and no external message is sent merely because the draft exists.
Proposed engagement
How we would approach the work
Classify consequences
Identify actions that change records, contact people or commit money. Assign an approval owner and a clear limit for each action class.
Design the review screen
Show the exact proposal, evidence and relevant changes. Provide approve, reject and request-information paths with understandable feedback.
Test bypasses and overload
Check unavailable reviewers, stale approvals, changed inputs and duplicate actions. Confirm staff can stop the workflow and recover from an error.
Deliverables to agree in the scope
- An action and approval matrix.
- A proposed review experience and decision record.
- Acceptance tests for changed inputs, escalation and stop controls.
Access, sample information and reviewer availability affect the plan. Any implementation, provider costs, support arrangements and acceptance criteria are agreed before work begins.
Limits worth understanding
- Human review does not eliminate automation bias or guarantee a correct decision.
- Reviewers need time, authority and enough information; approval cannot be treated as a decorative step.
Questions to bring to the first conversation
- What exactly is the reviewer authorising?
- What changes invalidate that approval?
- Who can stop the workflow if something goes wrong?
Australian teams
Scope the work for your operating context.
For Australian construction or service workflows, distinguish preparation from commitments made to clients, suppliers or site teams. Map the actual authority held by an owner, coordinator or project manager and test how the process behaves when that person is unavailable.
A starting reference for your review: OAIC: using commercially available AI products. Local obligations and deployment settings need to be assessed for the actual use case.
A focused next step
Work with Temrik.
Tell us about the workflow you want to improve and the outcome you need. We can review the context and discuss a focused assessment. Scope and price are agreed before paid work begins.