Kinda new to the Call Center world, this may be old hat to you veterans.

I have 3 skill Q's

All agents are qualified in all skills.

Call volume by %

Skill A = 25% (Priority 1)
Skill B = 50% (Priority 3)
Skill C = 25% (Priority 2)

(Call length is roughly the same on all skills)

With Priorities set as listed above and all agents logged in to all skills
the speed of answer is always much better on skill B as compared to the other
2.

My thinking is since the rate of inbound calls is much higher on skill B compared to the other 2, the priority settings are negated and skill B effectively "Hogs" the Q's.

Is this accurate ? Is putting Agents in Skill's A & C alone an appropriate solution ? (Leave productivity out for now, just ASA and Service Level)

Also ....

How does one use Erlang equations in this situation ? I wish to address each Q
individually as well as total. Since the metrics I am responsible for address each Q individually.

Any insight would be greatly appreciated ...


"Tarrant"

Tarrant,

I suspect the difference in ASA is largely down to call distribution as if your priority system works then it shouldn't be the case regarding the ASA by queue.

What i would do is look at a shorter period, say one 15 minute interval, where you know you recieved calls on all three queues and compare the ASAs for that period, you should see that the higher priority queues have a lower ASA.

Two assumptions i'll make to explain what i mean as simply as possible, one is that your staffing profile is fairly flat, the second is that your call distribution for queue B is fairly flat.

Over the an entire day your staff can deal with the interval traffic on queue B and therefore periods where you have no or few calls on queues A & C you have a low ASA.

Any interval where the other two queues start to use more resource then the natural consequence is that your ASA will rise for that interval, for all queues, though admittedly the increase should be mainly on Queue B.

However over the period of the day your ASA on queue B will be weighted, the many periods of low ASA will partially negate the effect of the periods of high ASA, when the other queues were also active.

However the ASA of the other two queues will be heavily weighted in favour of the high ASA intervals.

However if you don't see lower ASAs on queues A & C when you look at a specific interval then i would look at the prioritys on your ACD and ensure that they aren't set back to front, some systems i've used 3 is highest priority, others 1 is the highest, may be as simple as that?

As for erlang modelling multi queues with priorities. Its tricky to get per queue forecasts for ASA, SLA, etc, but given your equality in handling times then you should be able to overlay the call profiles and get a pretty accurate resource forecast.

I suspect the difference in ASA is largely down to call distribution as if your priority system works then it shouldn't be the case regarding the ASA by queue.

A: I think the difference in call volume between Q's negates the pri settings.
If the inbound rate is 200% on one q relative to the others it forces high ASA's
on the lower inbound rate Q's. One of my questions was if this was true ?


What i would do is look at a shorter period, say one 15 minute interval, where you know you recieved calls on all three queues and compare the ASAs for that period, you should see that the higher priority queues have a lower ASA.

A: This is the basic problem, I agree it should be. It's not.

Two assumptions i'll make to explain what i mean as simply as possible, one is that your staffing profile is fairly flat, the second is that your call distribution for queue B is fairly flat.

a: None of my call distribution is flat, i'm sorry to say. Staffing is.

Over the an entire day your staff can deal with the interval traffic on queue B and therefore periods where you have no or few calls on queues A & C you have a low ASA.

Any interval where the other two queues start to use more resource then the natural consequence is that your ASA will rise for that interval, for all queues, though admittedly the increase should be mainly on Queue B.

However over the period of the day your ASA on queue B will be weighted, the many periods of low ASA will partially negate the effect of the periods of high ASA, when the other queues were also active.

However the ASA of the other two queues will be heavily weighted in favour of the high ASA intervals.

However if you don't see lower ASAs on queues A & C when you look at a specific interval then i would look at the prioritys on your ACD and ensure that they aren't set back to front, some systems i've used 3 is highest priority, others 1 is the highest, may be as simple as that?

A: Priorities are set correctly. Open to suggestions however.

As for erlang modelling multi queues with priorities. Its tricky to get per queue forecasts for ASA, SLA, etc, but given your equality in handling times then you should be able to overlay the call profiles and get a pretty accurate resource forecast.

a: I understand it's tricky, hence my question how ? I'm not a guessing kinda person. I would like to add reason instead of the crytal ball method.

Surely this must be a somewhat common issue among call centers ?

How is this analyzed ?

Some of this seems so counter intuitive.....

Tarrant

Tarrant,

If analysing that stats down to that level still leads you to think that your priority system is not working as it should then the perhaps you could physically test that hypothesis

Do you have any control over the ACD?

Perhaps you could just test your theory as regards the delivery by setting up a couple of dummy skillsets with different priorities and a dummy agent.

Make calls to the dummy queues so they are queueing and then log in the dummy agent and see which call is delivered first?

As for the complexity of forecasting differing prioritised calls using erlang, and i presume some sort of spreadsheet, then unless Dave A or some other genius has found a way then i just don't know if its possible without some serious programming.

I consider myself fairly competent in Excel and VBA and cannot get a model that i am happy with, i also spent time with a company called Teleopti working on getting their WFM tool to do just that in line with the prioritisation system i had in Symposium, and that took a team of professional developers some time to figure out and build into a simulation programme.

Guess thats why WFM tools are so damn expensive. ;-)

Sorry i couldn't be any more help

Tarrant,

The last company I was with had a very similar set up (i.e. 1 multi-skilled agent group taking three types of call at different priorities).

The ASA of the third priority line should, at least on average, always be greater than on the other two lines, providing there are not a disproportionate number of abandons on the third priority line and providing you don't limit the Queue lengths in a discriminatory manner. This is the case whatever the relative volumes on the three lines.

Conceptually it would be possible to have a lower ASA on the third priority line if it had a much greater number of abandons (if you answer 80% of calls offered without any delay, then this will apply equally to each line. However if 15% of the third priority line's calls are abandoned, then for the ASA calculation it would appear that 94% of calls answered on the third priority line were answered without delay). However while possible, I guess it would be rare in practice.

Like William, we found that many apparent 'oddities' in daily figures could be explained by the different intraday profiles on the lines, but since this is not the case with yourself, I can only suggest that if you can get reports on the call profiles for each line (i.e. 80% answered within 5 seconds, 10% between 5 and 20 seconds etc, and similarly for the abandons), it may show something interesting. At very least it should confirm that calls on the lines with higher priority are not being 'delayed' by wait steps / messages or whatever.

As far as calculating what the ASA would be for each of the three lines for a given level of staff, I believe it should be possible using call queueing theory, providing each line has the same handling time. However I have never come across such a calculation in use, probably because it is not practical to do it for lines with differing handling times, and nor is it practical to calculate service levels for the three lines. In these situations some form of call simulator would need to be used.

Don't know if that helps,

David.