Step 1: diagnosis
Interview each team for 30 minutes: which tasks take the most time, which are repetitive, which documents they write most often. Note current tools and the data they use.
Step 2: choose three first uses
Pick three uses with visible value and low risk, in different teams if possible. Write for each: the task, the person responsible, the measure of success and the review rule.
Step 3: choose tools
Start with what you already have: many office suites now include AI features. Add a specialised tool only when a use case needs it. Check data settings, admin controls and cost per user.
Step 4: train and set rules
Train in short, practical sessions on real tasks. Publish a one-page usage policy: approved tools, data that must never be pasted, human review rules, how to report problems.
Step 5: measure
Compare before and after on time, quality and satisfaction for 4 to 8 weeks. Keep notes on failures; they show where rules or training are missing.
Step 6: scale gradually
Extend what works to similar teams, build shared prompt libraries and templates, then look at automation and integrations. Stop pilots that do not show value.
Signs you are spreading too thin
- More than five pilots at once.
- Tools bought before a use case is defined.
- No one responsible for each use case.
- No measure agreed before starting.
