Reviewed and updated October 8, 2026.
About the prior findings: The earlier numerical claims on this page are not used as evidence in this guide. The questions below help evaluate a specific workflow using records from your own organization.
Start with the work you need to change
An adoption rate does not show whether AI will help a particular process. Identify the task, its owner, the people who use its output and the consequence of a failure. Compare a process change, existing software and custom development before choosing a model.
Build an evaluation brief
- Current work: Describe the inputs, handoffs, exceptions and checks people perform. Use available records to establish a baseline for completion time, quality and cost.
- Source readiness: Identify usable information, access permissions, missing records and who maintains the source. Test outdated or incomplete inputs as well as ordinary cases.
- Acceptance evidence: Agree representative tasks, failure cases and reviewer criteria before comparing approaches. Record where an output must cite a source or reach a person for a decision.
- Operating boundary: Map prompts, files, outputs, logs, backups and external dependencies. Assign responsibility for access, monitoring, incident handling and changes.
- Business case: Label inputs as observed, supplied, sourced or assumed. Include implementation and operating costs, and keep speculative value separate from the core estimate.
Separate a demonstration from operating evidence
A successful demonstration is a reason to evaluate the next stage. Compare the proposed approach with your baseline under agreed conditions, including exceptions, permissions and recovery from failure. Record the test conditions, unresolved gaps and what was actually observed. Measure live use before attributing an operating improvement or financial return to the system.
One-workflow evaluation worksheet
| Record | Fill in before approval |
|---|---|
| Task and baseline | Current steps, workload, failure examples and source of the baseline |
| Options | Process change, existing software and proposed AI approach |
| Acceptance | Representative cases, rejected errors and named reviewers |
| Operating scope | Data paths, permissions, dependencies and support owner |
| Next decision | Evidence still needed, owner and condition for proceeding |
This worksheet is unscored. It supports an evaluation brief rather than a citywide adoption measure.
Choose the next decision
Proceed to a scoped evaluation when the workflow, sources and acceptance questions are clear. Gather more evidence when access, data quality or the operating owner is unresolved. Consider existing software when it meets the requirements. Define what your team needs to approve next rather than treating adoption itself as the goal.
Inspect firsthand implementation evidence
The production gap field report labels production, proof, prototype and planned work. It is a qualitative field report, not a citywide adoption survey. The rebuild decision analysis explains one engagement's constraints and where answer quality remained uneven.
Use AI roadmap and business case planning to connect those questions to an agreed scope. Share the engagement brief with business and technical reviewers, or request a project review with the workflow and constraints you need to assess.
