Coverage across the clock
Compare against a baseline
No baseline pinnedPin a configuration to hold it as a reference, then change the roster or the demand dials. Every figure above will show its movement against the pinned case.
How this works, and what it can and cannot tell you
The staffing calculation
Required advisors are derived hour by hour with the standard contact-centre method: hourly volume and average handle time give traffic intensity in Erlangs, the Erlang C formula gives the probability of queueing at a given headcount, and the model steps the headcount up until 80% of contacts would be answered within 20 seconds. That figure is then divided by 0.70 to allow for 30% shrinkage — breaks, training, absence and everything else that keeps a rostered hour off the queue. Occupancy is measured against the productive remainder, not against every hour paid for.
Why the two headline numbers move apart
Tickets per advisor-hour carries scheduled hours in its denominator. Capacity bought to survive the 10:00 peak is still being paid for at 13:00, when the queue is quiet. Improve coverage and the reported figure falls, with no change in how anyone works. Cut the roster back and it climbs again, while the queue absorbs the difference out of sight.
Occupancy is the counterpart. It asks what share of time on the floor went into live work, which is a property of the day's shape rather than of effort. Read together the two separate a roster problem from a performance one. Read apart, the first will be mistaken for the second.
Two of these numbers are about the schedule, not the staffing
Schedule quality compares this roster against the best arrangement of the same headcount, using no more distinct shifts than the panel offers. When it reads a figure rather than "best shape", that many advisor-hours of exposure exist purely because of when the shifts start. They cost nothing to remove. It is worth knowing which half of a coverage problem is a hiring question and which half is a calendar question.
Annual cost applies a fully burdened hourly rate — wages plus payroll taxes, benefits, equipment and overhead — across 260 service days. The published range for a US contact-centre seat is roughly $28 to $48; a technical desk sits above a general one. Because the figure is driven by rostered hours rather than headcount, two rosters with the same number of people can differ by six figures a year.
Four time zones flatten the lunch dip
A single-region desk sees the textbook shape: a mid-morning high, a clear lunch trough, a smaller afternoon rise. A desk covering the whole country does not. Eastern lunch is Central mid-morning and Pacific breakfast, so the dips of one zone are filled by the peaks of another. What survives the aggregation is a single broad plateau from roughly 10:00 to 15:00 Eastern, with the true high late in the morning while all four zones are open at once.
The practical consequence is that the staffing problem is not "cover two spikes." It is "hold a near-constant high line for five hours, then unwind it" — and eight-hour shifts starting on the hour cannot express that shape without either leaving the shoulders exposed or paying for capacity at 07:00 and 19:00 that the queue does not need.
The last hours belong to the West Coast. At 17:00 Eastern the queue is Pacific mid-afternoon, and those hours run thinner per advisor because the desk is holding a floor for one time zone rather than serving four. Handing routine tickets to an AI agent barely moves the plateau, because peak coverage is set by the shape of the day rather than its total. What changes is what the closing shift is for.
What the model settles, and what it cannot
Roster shape explains a great deal of the movement in a productivity figure, which is exactly why it cannot be assumed to explain all of it. Three causes produce a similar-looking decline, and only one of them is about the team.
| Cause | Signature in the data | Where the fix sits |
|---|---|---|
| Roster shape | Occupancy holds while tickets per advisor-hour falls; the gap sits in the troughs, not the peaks | The schedule. Report occupancy alongside so the figure is read correctly |
| Changed work mix | Handle time rises and reopen rates fall together, with volume flat or down | Nowhere. Harder work per ticket is a different job, not a slower one |
| A genuine decline | Occupancy falls too, reopens climb, and the pattern outlives any change to the roster | The only case where the productivity figure meant what it appeared to mean |
A model built to prove a metric misleading will quietly bury the third row. That is why occupancy is reported beside the headline figure rather than tucked underneath the argument: the two dials that dissolve a false alarm are the same two that would confirm a real problem, and they have to be trusted in both directions to be worth anything in either.