Start with one process and four weeks:
- Week one: watch the work happen and pick a task that costs a lot of time every day. Describe the process as-is and where the bottlenecks are.
- Week two: describe how the task could be done more efficiently. Be as detailed as you can. Set clear KPIs and then build the simplest version, with a partner.
- Weeks three and four: let the team work with it and check whether the KPIs are being met.
After that you know whether the AI implementation works, how quickly your organisation can change and where adoption is likely to slow down. That is already more than most AI programmes can say after eighteen months.
Watching the work is the step almost everyone skips and the one that yields the most insight. Sit next to someone for twenty minutes and have them narrate what they are doing. You will see things that appear in no manual: the spreadsheet on the side, the copying between two systems, the five minutes of searching before every email that goes out. You are not doing this to be polite: what you see there decides which process you pick.
Two warnings from practice. Agree KPIs before you build anything, for example the lead time of a quote or how quickly a customer gets an answer. Without that agreement every outcome is fine in hindsight and nobody learns anything. And leave work that should not exist alone: AI only makes a bad process faster. Not better.
Does your solution work after four weeks? Then you expand and your own team takes it over. Does it not work, then you probably have a very good idea why.