AI & decision-making · JNAGA perspective
What should a business clarify before funding an AI initiative?
An AI initiative is easier to assess when the work, its risks and the person accountable for the result are named first.
Published 25 September 2026
Imagine a team wants AI to answer complex customer questions. It can produce a plausible draft even when a policy has changed or the customer's circumstances are unusual.
A safer first test is a bounded drafting assistant with current source material and a named reviewer. Only after measuring error and review effort should the business consider whether any reply can be sent automatically.
Describe the task without naming the model
Is the intended change to find information, draft a first version, classify incoming work or support a decision? Describe who does that task today, where it goes wrong and what a useful improvement would look like. If the process is not understood, an AI tool can make the confusion faster.
Separate assistance from authority. A tool that suggests an answer is different from a system allowed to send it, change a record or make a consequential decision.
Check information and failure paths
Identify which information the tool would receive, where it comes from, whether it is permitted to be used, and what happens when it is incomplete or wrong. Decide who reviews outputs, how exceptions are escalated and how errors are corrected. These questions are part of service design, not paperwork to add after procurement.
A small trial should use a defined set of representative cases, including awkward ones. Compare the result with the current process rather than a polished demonstration.
Fund a testable next step
The first commitment may be to improve source information or define a review workflow, not to deploy a model. State the decision the trial will inform, its stopping conditions and the owner of the outcome. That preserves room to choose a simpler solution if it proves better.