Two different problems
- Understanding the request — what is needed, how long, what skills, how urgent
- Assigning the slot — constraint satisfaction across people, time and travel
The first is a language problem and suits AI. The second is an optimisation problem and suits scheduling software.
Duration estimation is the valuable part
Most scheduling pain comes from jobs booked for an hour that take three. Estimating from the description and from what similar jobs actually took fixes more than any optimiser.
It uses your own completion history, which means it gets better as you accumulate data and is specific to how your team actually works.
What else it can determine
- Skills or certifications the job requires
- Parts or equipment likely to be needed
- Urgency against your own policy, not a generic scale
- Whether access arrangements are mentioned and complete
- Whether this is a repeat of a previous job at the same site
Rescheduling communication
When something slips, the affected customers need telling. Doing that manually is slow and it is the moment the delay becomes a complaint.
Automatic notification with a specific revised time, sent within minutes, changes the reaction entirely.
Keep the schedule human-controlled
Propose assignments; let a scheduler confirm. They know about the things your data does not: who is having a difficult week, which customer needs the senior engineer.
Systems that override that judgement get worked around within a month.