Service Level, please help

[imran94]

I have been hearing the definition of service level as "Total number of calls answered by an agent within the time period (20 secs) divided by the total number of call offered (Answered plus abandoned) then expressed as a percentage."

In the above definition, what is the difference between the calls answered and calls offered. It states calls answered in 20 seconds divided by the total number of calls offered....but offered in what time?

I would appreciate if someone can please explain the real definition of service level in an outbound call center, the tool(s) used to calculate service level, how can the service level be controlled and the factors effecting SL.

Thanks a millions!!!

[Matt Schmitz]

Calls offered is just that - the total number of calls placed onto your switch and made available for agents to answer.

Calls answered (in 20 seconds) - the number of calls answered befoe 20 seconds after the call was offered.

These terms are normally used only for Inbound SL computations - not outbound.

Matt

[szaj2]

Since Service Level is the percentage (80%) of calls that you want to answer within a given timeframe, i.e. 20secs, it has no direct association to outbound calling.

Usually for outbound calls management presets the goal. The goal can vary greatly depending on the nature of the outbound call. For example, the goal for a Welcome Call that a wireless service provider routinely conducts is much different than a telemarketing call by a timeshare company.

The measurement usually includes the number of calls placed, number of contacts made and some result of each contact.

[JoeB]

Some of the confusion stems from the fact that calls come at a different (random) rate than those being handled. Because there are usually more calls than can be handled immediately, they are queued and await service, and some will eventually be abandoned. Service level is defined as how many of the total available (“offered”) calls were handled within a given target (e.g. 20 sec.). Some service level calculation methods also factor in the number of abandoned calls.

For example, letÂ’s assume these stats:

Call offered: 100
Answered within service level threshold: 70
Abandoned within service level threshold: 10
Abandoned exceeding threshold: 20


Common methods to calculate service levels:

Abandoned calls have negative impact: 70 / 100 = 70.0%
Abandoned calls have positive impact: (70 + 10) / 100 = 80.0%
Abandoned calls ignored: 70 / (100 - 10) = 77.7%

The more common and probably most accurate method is the first one, because it acknowledges that 30 calls were “lost”. However, you could argue that calls abandoned during the service level window should not be counted because they would have been serviced if not abandoned, and use the second method.

Joe Barkai
DIAGNOSTIC STRATEGIES
http://www.DiagnosticStrategies.com/
Tel. 781-433-0833

[RickB]

Very good reply from Joe, however I would caution against using any method other than the "pure" formula reflected in his first example. While site mgmt (or anyone who has service level on their scorecard) would love to have abandoned calls within service level count to the positive, I favor metrics that keep to the main measurement objective - in this case a very simple "number of calls answered in service level". (You could of course also use ASA - Average Speed of Answer.)

If a segmentation of your abandon call population is needed (i.e. in "buckets" of X % abandoned in Y seconds) you could develop a metric for your abandon service performance. I assume site mgmt might still attempt to use the "under service level" abandons to justify service short falls. The advantage is that it will be clear to all at what intervals your abandons occur. From this data you can assess your servicing strategy, i.e. is your current SLA meeting your customers expectations?

Just my two cents worth.

Rick

[yooo]

I would like to share this reply I got from an online friend, which actually explained it a bit more:
Your 84.9% service level is calculated by your reporting
software. The reporting software captures in how many seconds each call is
answered so then you can analysis reports to not just see how many calls are
answered within 90 seconds but also by whatever time ranges you wish to
consider. 84.9% service level means that 2839 calls(84.9% * 3345) were
answered within 90 seconds. The remaining calls were answered in 91 seconds
or more. Calls Answered / Calls offered gives you your answer rate (97.2%)
or in reverse, your abandon rate (2.8%). It is only a coincidence that
sometimes the service level coincides with your answer rate.

.....I think what confused me was "the number of calls answered in 90 seconds". I would think this to be in a different way. I would think , say, 230 calls are being taken within 90 seconds, meaning, say for instance, from 12:00:00 am to 12:01:30am, 230 calls were taken.

But Now I would say, 90 seconds is the time when a call is taken, right? That each of the 230 calls were answered within 90 seconds.

[JoeB]

The detailed example is consistent with previous discussions. The ACD reports service levels based on the ratio between calls offered and the call answered with the target service level (“service level threshold” above, and as 90 seconds in the example).

The reason Service Level and Answer Rate, as defined above, coincide is not a mere coincidence. Experienced call center planners set their Service Level targets in accordance with actual traffic and avaialble resources, so that under most operation conditions they meet the Service Level target. Consequently, their Answer Rate will always be only slightly better than the Service Level, accounting for a relatively small number of calls answered outside the target service level less abandoned calls.

As noted previously, some ACDs allow the definition of slightly different service level calculations.

I am not sure I followed the last two paragraphs. Service level is simply defined as percentage of calls answered within a predefined time, e.g. “85% of the calls will be answered within 90 seconds”. The call center staffing is designed to sustain this performance level independent of the number of calls offered at any period of time (but, obviously staffing level is calculated for a specific average call volume). Therefore, to report servie level we need to use percentage, not absolute numbers. The reason you can “guarantee” service level in a well-designed call center is because you leverage the statistical nature of the calls queued by the ACD (see http://www.diagnosticstrategies.com/...20Modeling.pdf)

[RickB]

Joe,

From your previous reply:

"The reason Service Level and Answer Rate, as defined above, coincide is not a mere coincidence. Experienced call center planners set their Service Level targets in accordance with actual traffic and avaialble resources, so that under most operation conditions they meet the Service Level target. Consequently, their Answer Rate will always be only slightly better than the Service Level, accounting for a relatively small number of calls answered outside the target service level less abandoned calls."

I hate to disagree with you here, but Answer Rate does have very little DIRECT correlation to your service level statistic. Yes, there is a general trend relationship, however there is not a direct numeric one. This is an issue that I have been dealing with for some time. Speaking from an antecedent level, the amount of time your customer will hold has a direct correlative relationship to the amount of time they have to hold and or the severity of their issue. For example, if I have a bill payment issue that is messing up my credit rating I will hold and hold and hold until you answer the call. From a stats view point I reviewed 5 years worth of AR / SL data to show that in a 98 - 99.99% answer rate window (i.e. standard deviation of .0052) the ASA data yielded a standard deviation of 14.53 - hardly a close relationship.

In my conversations with other call center professionals the answer rate vs service level debate is one of the most contentious. Often times, those with scorecards directly related to service quality champion the use of an answer rate statistic / target over that of service level. The reason is obvious as they can deliver a wider range of (unacceptable?) service levels while present a "good face" to senior mgmt - i.e. we are making our answer rate goals. (Which, by the way, say nothing about the quality of the customer experience.)

Best Regards,

Rick

[JoeB]

Rick,

I think that this is a good discussion and while there may still be some disagreement in regards to terms and measurement methods, we are very much in agreement in regards to the fallacy in managing purely by statistics, as reflected in your last paragraph – meeting (arbitrary) performance targets such as answer rates doesn’t tell much, if anything, about the quality of service and about customer satisfaction. As you state in your previous post “is your current SLA meeting your customers expectations?”

The reason why there is a relationship between service level and answer rate is simple. The formulae proposed in previous discussions were:

Answer Rate (AR) = Call Answered (CA)/Calls Offered (CO)
Service level (SL) = Call Answered Within Service Level Window (CSL)/Calls Offered (CO)

By definition, total call answered is the sum of calls answered with in the service level window and those answered later (CX): CA=CSL+CX. Therefore:

AR = (CSL+CX) / CO
SL = CSL / CO

And the ratio between the two is CSL / (CSL+CX)

The point is merely that within a *single* call center there will be (an average) fixed relationship between answer rate and service level. Intuitively, because both answer rate and service level deal with sub-groups within the answered calls, there will be a relationship between the two, determined by the number of calls answered within the service level window (and abandoned calls, if you decide to factor them in). Again, this is for a given call center – if you meant to suggest that one cannot generalize this behavior and induce one metric from the other – I agree.

I am not sure that I agree with the use of standard deviation as a correlation coefficient. Standard deviation measures the variance in sample or population data, and while it is a very important statistical tool, it doesnÂ’t imply correlation (or lack thereof) between sets of data.

Finally, if I read your reply correctly, you compare STDEV in answer rate, which is expressed as a percentage, and STEV in ASA, which is expressed in time units (presumably seconds). This is not a valid comparison (STDEV measures the level of spread around the mean and is always expressed using the same units of the sample data).

You said: “…the amount of time your customer will hold has a direct correlative relationship to the amount of time they have to hold and or the severity of their issue” this sounds like plain tautology although I think I understand what you meant…

This discussion may be going into a level of detail that might make it less useful for many. In summary, however, it does demonstrate that call center metrics, if taken seriously, can be fairly involved and that back of an envelope calculations may be risky if one doesnÂ’t fully understand the alternatives and subtleties. The, lack of standards and common analysis methods only worsen the situation. I hope that this discussion helped others in seeing it.

Thanks for your thoughtful comments and good discussion points!

Joe

[RickB]

Joe,

Tautology, yes but just plain? Hummm... let's see is it the law of noncontradiction, or maybe the law of identity or.........

In any event, I have enjoyed the discussion. I think that we do have some points in common, however I do regret that I am not as eloquent in my rebuttals.

One final comment: Of course there is a difference in the percentage vs time spread analysis. My error in this discussion was in trying to over-simplify (considering the generalist nature of the forum) my case. Since the point was missed I have failed in my task - yes you can run Bivariate and or Partial correlations to the day ends but the result will not tell you the impact a 3 - 5 second variance in ASA will have on your staffing model. (One center or many, ARIMA model or simple moving average.) Let alone the impact on customer satisfaction.

Well, it appears that my envelope IS full now (I use 8 x 10 inter-office as my standard). I DO take call center metrics seriously, sometimes too much so. Oh well, as Deming said, "How could they know..............."

Rick

[imran94]

This thread was started by me and I have no words to express my appreciation for all the valuable inputs. As far as I am concerned, this has been a wonderful experience and many of my concepts have got cleared.
Once again, I thank all. You have all been wonderful. Actually, from the above, I am putting it in my diary and presentation for a promotion.

Cheers!!!