Choose among four outcomes: improve the existing process, configure native software, buy human cover, or pilot specialist automation. Do not begin by deciding which AI agency to hire.
| Decision | If yes | If no | Evidence to produce |
|---|---|---|---|
| Can we identify one measurable problem? | Establish its baseline. | Record the workflow before shopping. | One-page problem statement |
| Is there delivery capacity, or a specific administrative benefit? | Calculate a conservative business case. | Fix capacity or choose an admin-only workflow. | Capacity and benefit calculation |
| Can existing software or better call cover solve it? | Test that option first. | Identify the missing capability. | Native-software comparison |
| Are integration permissions, safety boundaries and data responsibilities clear? | Invite comparable pilot proposals. | Commission narrowly scoped specialist discovery. | Requirements and risk register |
| Can we test with spending limits and a working fallback? | Run a bounded pilot. | Do not launch customer-facing automation. | Signed pilot charter |
| Does the pilot improve outcomes after costs and rework? | Expand one permission or workflow at a time. | Repair, extend within a cap, or stop. | Continue/change/stop decision |
Recommended buying rule: approve the next evidence-producing stage, not an entire transformation programme.
PUT THE RESEARCH TO WORK