Taking AI into production
Five decisions that decide whether a government AI pilot becomes a service, from a dozen proofs of concept in Australian agencies
Government AI pilots rarely fail on the technology. They stop because the decisions that carry a pilot into production were never in scope. This paper sets out the 5 that matter, drawn from about a dozen proofs of concept in Australian agencies. For each it describes what failure looks like and how to check. It points to the DTA’s guidance for AI proof of concept to scale. The register published with it can serve as the decision register that guidance asks for.
- 5 decisions, each with a named owner, the evidence that closes it, and when
- Written from about a dozen proofs of concept in state, territory and Commonwealth agencies
- Why most agencies should buy rather than tune, with the exceptions named
- pdf · 252 KB
- xlsx · 19 KB
- pdf · 62 KB