Start with a baseline
Before building anything, measure how the process works today: how long a task takes, how many people do it, how often errors happen and what those errors cost. Without a baseline, any result is just an opinion.
Count the full cost
- Software licences or API usage
- Build and integration time
- Data preparation and cleaning
- Training the team
- Ongoing monitoring, maintenance and review time
Review time is often forgotten. If every AI output needs a person to check it, that time belongs in the calculation.
Measure the right benefits
- Time saved: hours per week returned to the team, and what they now do with them
- Quality: fewer errors, faster responses or better consistency
- Revenue: more leads handled, higher conversion or new services offered
- Experience: customer satisfaction and employee satisfaction
Run a fair test
Where possible, compare the AI-assisted process with the existing one over the same period, on similar work. A two- to four-week pilot with clear numbers is more convincing than a long project with vague impressions.
Decide in advance what success looks like
Agree on a target before the pilot starts, for example "cut ticket handling time by a third without lowering satisfaction scores". If the target is met, scale up. If not, adjust or stop.
Report honestly
Share what worked, what did not and what it cost. Honest reporting builds trust for the next project, and it keeps AI investment focused on real business results rather than novelty.