AI / WORK DESIGN AND GOVERNANCE
Design AI work around professional judgment
Choose a real workflow, define the AI task and the human decision, then measure the result.
Consider invoice processing from receipt to payment. AI can extract data, classify items, and flag anomalies. A responsible person still reviews exceptions and approves decisions that affect cash or financial reporting. This division makes both the time saved and the residual risk visible.
Start by recording the current workflow: cycle time, errors, rework, and supporting evidence. Then pilot one bounded task. Define which data AI may use, what happens when its confidence is low, who reviews the output, and how corrections are recorded.
NIST organizes AI risk management around govern, map, measure, and manage, with human oversight documented. The OECD AI Principles emphasize trustworthy systems and accountability. At DHISORA, these ideas lead to a practical test: who owns this result, and can the decision be traced?
Actual workflow → AI task → exception rule → human review → measured outcome