- 看不到交付可用性的用戶，並且 確實看到交付可用性的
One possibility is to look at idle time (time a driver spends waiting for the next order). If the drivers are on your payroll (as opposed to working on commission, i.e., doing "gig" work), idle time has a direct cost. If the drivers are gig workers, a relatively even distribution of idle time might be perceived as "fairer" and might contribute to driver retention. (A better measure in the gig case might be driver revenue per hour, adjusted for mileage costs, but your data does not suggest any measure of revenue, or of mileage for that matter.)
If we use idle time, then "under-utilization" ("over-utilization") would mean more (less) idle time per shift than some standard. (The standard would likely be time-dependent, since people are more likely to order food at some times than at others.) You could make the standard either absolute (more than 20 minutes idle per hour is under-utilized, less than 5 minutes per hour is over-utilized) or relative (if you idle time is 1.5 standard deviations below the average for a certain time span, you're over-utilized, etc.), where the standard deviation and mean would be sample-based. Note that the sample-based approach does not detect over- or under-staffing as well as the absolute standard approach (assuming you pick your absolute standards wisely).