We are currently running a project asking call centre consultants to use a data capture tool to give us a view what there call was exactly about.

My call centre currently gets 175k calls a week, and Im stuck in terms of deciding what an ideal sample size should be.

Has anybody ran anything similar or have any idea's on ideal sample size?

Thanks

Geek Warning!

LINK ONE

LINK TWO

Both are very good descriptions, and as it's part of the degree course I'm
on at the moment something that's fresh in my mind.

The key is the fact you want an even distibution, and as you say, a relevent
sample size.

WHilst the good old fashoned 10% is PROBABLY good enough in this
case have a look at the 'Central Limit Theorem' in the second link. This is
currently probably the best way to decide.

The other interesting part of course is how you take said data sample
in a 'random' way.

Have Fun

If you need anymore drop a line back.

Regards

DaveA

You will reach a point where the additional effort required to collect the data will be too great to substantiate the increase in accuracy that you obtain.

Depending on how you select the sample, 400 calls may be enough with a reasonable level of accuracy (above 90%). The difficulty is if the advisor finds it difficult to select the call type, or indeed the customer calls about several issues in one call. Then you get a lot of "other"!

Also, make sure you can link back to the call recordings in someway during the data collection so that you can verify your findings. You wouldn't want your report to send company strategy down the wrong route ;-)



Hope that helps,

James

http://www.answersresearch.com/pdf/SampleSizeMagic400.pdf

Hope that helps,

James

400 out of how many?

400 out of 4 million is not likely to be able to give you
the 5% CI you're looking for.

It's a rollnig dice (p(X)x = 1/6) question.

Look at the difference 30 / 300 / 3000 levels.

It's dependent on the number of variables you have,
at p(x) = 1/6 the 5% CI doesn't come in until r=>2700.

The 'paper' is based, correctly, on questional sampling
and obviously related to qualitative rather than quantitative
data so may fit what you're after just be a little cynical.

To paraphrase Tom Cruise. "Show me the numbers!"


Regards

DaveA

I think that paper could be dangerous in the wrong hands. Yes there is a point where extra samples are just noise but it will vary depending on the population and variance of the data. 400 would not be appropriate for a large population with great variance.

If you are taking over 9 million calls per year then, unless you have very few demand types, it is unlikely that 400 calls would be representative.

My advice would be to run the data capture for a few weeks (bearing in mind that not all calls will be logged) and then look at the broad headings of your demand. At this point I would ignore sample size. If you find great variety in the logged calls then you may have to extend the logging to get more data.

HTH

Unless your contact centre only handles simple calls, it is likely that many calls will have multiple contexts making call classification subjective and potentially distorting the results. For example, the customer wants to make a credit card payment after checking their balance (simple example, accepted!) - do you classify this as "balance enquiry" or "credit card payment"? (Technically it's both.)

One way to get an accurate view of why your customers are calling is to utilise the wealth of information that you have within your call recordings (assuming you record calls). There are a number of vendors who can provide speech analytics tools that can take a sample (up to 100% in some cases) and produce very detailed analysis of the calls - some will even offer this as a "service" (i.e. you provide the calls, they will do the analysis and feedback). I recognise that this option requires money (although maybe not as much as you would think) but the speed and accuracy of results can make it pay for itself very quickly and many times over.

I hope this helps,
Dave

:-)

I agree with the statements above about getting the calls in the right category and being in multiple categories. Interaction Analytics can certainly help provide the mass analysis if that is what is required.