The primary thing that determines how many staff a call centre needs is how long things take to be done. Talking on the phone is not the only thing the staff do; they also process the calls; ask questions of their supervisors; call customers/clients/others back; take and make personal calls; have meals and rest breaks; take time out for training; go on leave; report in sick and are asked to do numerous other things on the company time.
So together with call duration, you also need to measure the average duration an agent spends after a call doing processing work. You should also measure the amount of time an agent spends “not ready” when they have no calls to process, as well as time away from the phone on breaks, meals, Training, Coaching, etc.
You should probably be measuring the number of calls handled and be relating this to the ideal call model for the type of call transaction you are processing. Agents who are too fast or too slow compared to that ideal call model may be compromising call quality, so you should have some means of assessing accuracy and quality of calls, as this can be a hidden money waster.
You may also want to assess the number of repeat callers or transactions per call (or even calls per transaction), as your objective should be to get it right first time – every time.
If your task is to size a Call Centre for the number of staff needed on duty, then you should be collecting data at an interval level, not just a daily level. The measurement interval can be as fine as 10 or 15 minutes, especially if you are running ad campaigns and know when the ads are likely to be broadcast. However, the call traffic pattern will vary over the day, so your call time and volume measurements can change from interval to interval. I have observed a call centre where the call duration increased later in the day, even though the calling rate was about the same. This meant that there was a performance issue, because more staff were needed later in the day than simple averages suggested.
If you are going to be modelling the call centre, then you should also be assessing how well the real world centre is performing compared to the mathematical model. Thus the model should produce expected service level and occupancy (% of agents’ Time spent occupied with calls either talking or processing.) Do not expect to get over 80% occupancy unless your call centre is very large (hundreds of seats) and there is a low service standard, which means callers may wait several minutes to be answered. Occupancy will also drop as the average call handling duration increases, or the call volume reduces, or the service level is raised. Use Erlang-C calculators to determine the number of staff you need for a given Call Handling Time (not just the talk time) Some calculators will also allow occupancy, call waiting times, service levels etc. to be determined. These values should be compared to the real world situation, as improvements in one area will often result in a reduced performance in another area. If you model the real-world situation properly, then the real world figures will have a benchmark to be compared against. This will stop you trying to do silly things like, For Example, reducing staff because they spend time waiting for calls and are “unproductive” as the model will show you what proportion of time staff need to wait for and be “unproductive” in order to meet service level at optimum cost.
PS: In the end everything can be turned into a dollar value, so you should be measuring costs somewhere, operating costs, costs of calls, cost of sales, cost of missed sales and lost opportunities – even abandoned calls may be valuable, if it indicates lost business.