Pilots with measurable gain
A bounded trial where you know in advance what to measure and when to stop or scale.
Pick a problem, not a technology
Good pilots start from a concrete, recurring problem: time spent on manual work, errors in a handover, waiting time in a process. Poor ones start from a tool looking for something to do.
Useful criteria are that the problem is large enough for a gain to be felt, that data is available and of reasonable quality, that there is an owner in operations, and that a mistake cannot do harm before you have learned how the solution behaves.
Measure before, during and after
A pilot without a baseline proves nothing. Before the start, note how long the task takes today, how many errors there are and what it costs. During the pilot measure the same things, and watch for side effects, for example the work simply moving to another department.
A good measure is simple and understood by the people who do the work. Measuring three things properly beats measuring twelve loosely.
End with a decision
A pilot has a deadline and ends in one of three decisions: stop, adjust and run again, or scale. Stopping is a valid result if you have learned where AI does not help.
It is worth writing down what worked, what did not, and what must be in place before anything can run in production. That is the entry point to the next step.
Other topics
Get in touch
If you have a question or a comment about anything on this site, you are welcome to write. We reply when we can.
Your message was not sent. Check the email address and try again — or write to [email protected].