Alcatel Unleashed

Hi,

I can roughly estimate average waiting times through Erlang C calculations, but I'm not quite sure how to estimate lost calls based on my empirical data.

I have data on:
1. Calls Recieved.
2. Calls answered without queuing.
3. Calls abandoned within 5 seconds, 30 secions, 60 seconds and total abandoned.
4. Calls answered within 5 seconds,30 seconds, 60 seconds.
5. Average waiting time, lost calls.
6. Average ringing time.

I'm thinking that since i can to some extent ignore early abandons (5 seconds) that linear regression might suffice. But I would love to hear which approaches others on this board would suggest?

I manage lots of small queues and we're moving from monitoring service level to just monitoring efficiency.

Thank you very much.

What do you call lost call in your case ? Calls not answered , or called that hung up before pickup ?

By lost I mean customer who hang up the phone for some reason. Either in queue or while ringing.

A small percentage hangs up quite quickly, discovering that they dialed the wrong number for some reason.
Another slightly larger percentage hangs up after they hear an automated message telling them to wait.
A few people figure out the answer while waiting.
A few people get interrupted for some reason and need to hang up.
A few people are forwarded from somewhere else, so clumsily that the person isn't aware they're being forwarded. So they hang up.
Some people get bored waiting and just hang up, sending their request as a ticket instead.

The sum of all these people who hang up needs to be approximately 10% of calls. I'll be trimming the resources to hit somewhere between 90-92% in the course of a month, but never below 90% ... the call time varies between 1 minute and 1 hour for each call, with 2 - 8 agents on a queue at any given moment ... which is why I need to predict abandon rate and also figure out a reasonable model to capture the variations in customers willingness to wait among my different services.

I have good empirical data, just need to use it to make a reasonable model.


Thanks

Not quite there yet:

Empirical data shows a non-linear relationship. A couple of percent lost within the first few seconds. The spread increases as time progress.

Since lost calls doesn't take processing time, the number of calls needs to be reduced by some amount to estimate the new waiting time. The processed calls and subsequently average waiting time will be lower when some people abandon than projected, which needs to be taken into account. (although some will retry).

Anyone know how to describe this cyclical relationship in math?