| Deciding how many seats to buy | A number from the sales conversation | A pilot group sized from the role shortlist | Start small, measure, then expand where it is working. Seats can be added later more easily than removed. |
| Finance team in desktop Excel all day | Assumed to be a strong fit | Checked against where their files live | If the workbooks sit on a network drive, Copilot cannot use them until that changes. |
| Staff mostly in a practice or job system | Given seats with everyone else | Flagged for a different approach | Copilot can reach some other systems through connectors and agents, but that is extra setup. Custom AI built around that system may suit better. |
| Ten versions of the price list | Discovered when Copilot quotes the wrong one | Current versions named before the pilot | A short list of key documents, with one agreed current version each, removes the most visible failure. |
| Company-wide shared folders | Left as they are | Listed for a relevance decision | You decide which should stay organisation-wide. A security review, if wanted, goes to a specialist firm. |
| Judging whether the pilot worked | General impressions after a month | Measures agreed before day one | Time on defined tasks, weekly usage and user ratings, with stop, extend and expand thresholds written down. |
| Leadership team asked to approve spend | A vendor demo and a quote | A written readiness finding | Includes the honest outcomes: go, not yet, or a different approach for some teams. |
| Files spread across another cloud service | Ignored until results disappoint | Mapped and weighed in the recommendation | Moving or connecting that content is a project in itself and should be decided before seats are bought. |