Qualify
Clarify the workflow, owner, baseline, constraints, and alternatives. Decide whether the problem is worth investigating.
AI IMPLEMENTATION METHOD
Move from an AI idea to a workflow your team can run.
We start with the work that needs improving, then decide whether there is a sound reason to build, adapt, or stop. Governance is part of the work from the first decision through handover.
THE PRINCIPLE
Evidence before commitment.
An AI project should earn the next stage. The point is not to force every problem into a custom build. It is to understand the workflow, the information, the people, and the limits well enough to choose a useful next step.
Start with work
Choose a recurring task where volume, complexity, delay, or error creates a real cost.
Test the uncertainty
Use permitted examples and difficult cases to learn where an approach helps and where it fails.
Keep people accountable
Design the approvals, review points, and exceptions around the decisions your team continues to own.
FIVE STAGES
A decision at every stage.
Prove
Prototype or test the uncertain parts with representative tasks, permitted inputs, and agreed acceptance criteria.
Decide
Bring together the likely benefits, costs, dependencies, and remaining risks. Proceed, change direction, or stop.
Embed
Integrate the workflow, test with users, prepare operating support, and address the gaps found before release.
Handover
Agree documentation, training, ownership, monitoring, and support so the team can operate the solution responsibly.
The exact work, order, timing, and outputs depend on the workflow and the organization. These stages describe the decisions to make, not a fixed delivery promise.
WORKFLOW GOVERNANCE
Controls that help the work move.
Governance belongs in the workflow from the first decision through daily use. It answers practical questions while the solution is still taking shape. Clear controls help teams save time and money by reducing rework, speeding decisions, and making each step easier to review against applicable requirements.
What that looks like in practice
Consider a team preparing a client review from multiple documents. The workflow can be designed around controls.
- Approved inputs: use authorised sources and access permissions, with clear rules for what must stay out.
- Checkable drafts: show the evidence behind a finding so the reviewer can inspect the source, not just trust a fluent answer.
- A route for exceptions: flag missing or conflicting information for a person rather than filling the gap with a guess.
- A human release decision: let a responsible reviewer approve, edit, or reject the output before it reaches a client.
- A useful operating record: retain the agreed evidence and decisions, within the organization’s retention and privacy rules.
This is an illustrative design pattern, not a claim that every implementation includes the same controls. We agree the controls and test their behaviour against your workflow.
Agree the controls before people rely on the result.
Information and access
What sources are permitted? Who can access them? What should never enter the workflow? What must be retained or removed?
Evaluation and review
What does a good answer look like? Which errors are unacceptable? When must a person inspect or approve an output?
Ownership and change
Who owns the business outcome, approves changes, handles exceptions, and decides when the system needs review?
Operating evidence
What records, explanations, and monitoring will the team need to understand performance and investigate a problem?
WHO IS INVOLVED
Bring the people who know the work into the decision.
Process owner
Defines the work, examples, exceptions, and outcome that matter.
Subject-matter reviewers
Help test outputs against real practice and decide what needs human judgement.
Technology and governance leads
Assess integration, data, security, privacy, oversight, and operating responsibilities.
Sponsor
Decides whether the evidence supports investment and enables the change needed for adoption.