DP365 / AI Intelligence Brief 001 / 14 September 2026

Start with the workflow,
not the model.

Many organisations begin AI work by choosing a model or platform. A more reliable starting point is the workflow that needs to improve, the evidence available and the decision the project must support.

What businesses should care about

Make the experiment
answer a question.

An AI initiative should be specific enough that a team can say what changes if it works. Define the workflow, owner, inputs, exceptions, human approvals and baseline before selecting the intervention.

A broad objective such as “use AI to improve operations” is a starting signal, not an implementation brief. The next step is a workflow hypothesis that can be tested.

Practical applications

Workflow-first assessment can support document-heavy work, recurring customer enquiries, research and analysis, recruitment processes, knowledge retrieval, hand-offs and bounded decision support.

Implementation lesson

The first useful deliverable is often a decision boundary: what the system may do, what it may not do, when it must request review and how activity is recorded.

AI-agent boundary

An AI agent should have defined access, permitted actions, exception handling, human approvals and monitoring. Autonomy is not the objective by itself.

Limitations

AI outputs can be incomplete, wrong or unsuitable for a decision. This brief is not a financial forecast, implementation guarantee or vendor recommendation.

Recommended action

Choose one workflow
worth testing.

Document the current steps, inputs, exceptions, approvals and baseline effort. Then assess whether AI, automation, improved information access or a process change is the appropriate intervention.

Find a possible AI opportunity