Control charts for call centers.

Six Sigma – iSixSigma Forums Old Forums General Control charts for call centers.

Viewing 6 posts - 1 through 6 (of 6 total)
  • Author
  • #42973


    I have a very interesting challange ahead of me.  I am a Manufacturing Black Belt by certification, but am now a call center Black Belt / Manager.
    I notice very quickly in my call center, that there is an obvious lack of visual statistical communication of each and every individual group within the call center.  For that I have began a few groups going over a daily metric / dashboard as I would when I was manufacturing parts.
    The obvious problem I have is, what is the best tool to use to take a look at the data and make it easy enough for a rookie to use in minitab.
    My goal is to take a look on a daily basis and identify the agents that are performing out of range with the rest.  Problem 1, some agents take 50 calls, others 6.  So to calibrate fairly without removing all of my points I set the limit to at least 10 for averaging purposes.
    Now the average of an average is the largest problem  I am seeing.  But there is not much I can do with our current Average Handle Time reporting tool. (cant look at within variation by interval)
    When I run the IMR char on this data set I have about ten total agents.  Since I have no subgroups and IMR chart is my best option I can see easily.  But when I actually pump it in, the R is so large (like 200 seconds) that it is not really telling me a story that I can see.
    Initally 1 operator pops out every week as having an elevated AHT, but with the Variation so large,  I cannot assertain the ones at the top as easily.
    Any ideas?  I want an easy method to teach with a story style board and get the managers our there walking their process with somthing other then a piece of paper full of crap numbers with little to know obvious story telling skills.
    God I miss my widgets, but this a interesting new challange.



    A couple of thought…
    -It sounds like the number of calls is you “Y” and it is out of control as well as out of spec…You might want to kick off some projects on that.
    -Short term, look at some of the obvious “X” that your team can control (call length, post call processing time,…) and control chart those. That will get your managers focused on X’s to drive the Y.Your right…Widgets are fun, but transaction and call center enviroments can keep you on your toes and let you flex some Six Sigma “muscles” you haven’t used since training.



    I agree with the Y being # of calls.  The process is so far out of control that I have never seen this type of variation in anything.  I agree also with the project to take a look at variables such as call length, after call work.
    The only real measurment target we are tasked with is Service Level.  Service level is the % of call answered within a certain time. (ex 70% of calls within in 30 seconds)
    Mabye a more representative task for Y is # of calls being answered within service level.  In addition I can make a proportion defect style control chart where calls answered below a certain service level are automatically binned as bad news. (ok I like that Idea, amazing how a simple post on Isixsigma is answering my own questions logically)
    From an 30k level metric, the process is lieing to me as averges are very nasty in this world.  I obviously need to start rocking some interval data for more specific approaches to my world.
    There is also a tremendous gaging problem when it comes to the issue types an operator can select for a disticnt call.
    But back to control charts.  Is an IMR acceptable to monitor Average Handle Time per day/week/month for an operator. (no subgroups for the daily stuff




    Joe:Since you like your widgets, do a ‘Distribution ID’ plot of
    Hold Times and Call Times. We find we can usually fit the distributions with a
    Weibull or lognormal for inbound calls to information and help desks. The
    Weibull for Hold Times also allows you to model the call abandonment behaviour
    using the usual reliability analysis tools. Once the distribution is
    identified, the data can be transformed to normality for use in control charts.Others may say differently, but I would prefer an IMR chart
    with all calls rather than an XBarR with averages of groups of calls. Operators
    can be very inventive with metrics if they know they are being tracking using
    averages. On the practical side, there are problems in Minitab when you have
    differing subgroup sizes. This will be a problem if you are updating data
    points every 30 minutes (differing subgroup size), or using a workaround of
    constant subgroup size (operators will be out of sync and complain that call
    volume as a function of time of day unfairly penalizes them or credits others).Once the data was transformed, we could clearly identify
    operators that were systematically different than the rest of the operators
    using ANOVA, t-tests and Levene’s test. This could be nicely shown using
    ‘Distribution ID plot’ broken out ‘By Group’, or ‘Probability Plot’ broken out
    ‘By Group.’Once you have determined the best transformation, it is safe
    to use the same transformation for all subsequent data. In other words, you can
    build the transformation into an Excel spreadsheet for use by all the neophytes
    using it.Cheers, BTDT



    Thanks for that thought. I was pulling a rookie mistake on the distribution identification.
    What size sample / time in your opinion(expereince) would be good to look at to draw a group wide conclusion?
    Like sample 30 AHT in the day and use it, I have averages available to me very readily but am scared as they hide the true variation.
    Good thoughts though.  I agree on the Imr, the chart for the Xbar is nasty due to subgrouping constraints.  I would like to think a little further though.  My ranges are really nasty between operators with constant calls and this obviously biases my ouput in an IMR char because three sigma up or down is alot when you are talking about a sigma of like 100 or 300 seconds.
    Yeah, I like my widgets.



    This may be a case of presentation for different audiences. I would think you would use a “p” chart like you were suggesting to monitor proportion above the spec for hanging in the halls, working with your managers,… For you and others (after bringing them up to speed), a box-wiskers plot of call wait times with median confidence intervals with each box a day/week of data. You would be able to see differences in days (non-overlapping confidence intervals) + see the spread.

Viewing 6 posts - 1 through 6 (of 6 total)

The forum ‘General’ is closed to new topics and replies.