Forecasting Using Calls Offered

Forecasting Using Calls Offered

From the research that I have conducted, it appears that the most common approach to forecasting call volume is to use total calls offered. When you are achieving desired service level targets on a regular basis, I can (almost) see this as being relevant.

However, we are currently achieving a service level of 30% of calls answered within 20 secs and, consequently, have an abandoned rate of approaching 20%. Bearing in mind the number of retries and that the majority of callers will eventually be answered, what % of abandoned calls should be used & why??

Question asked by Chris

Forecasting

Hi all

From the research that I have conducted, it appears that the most common approach to forecasting call volume is to use total calls offered. When you are achieving desired service level targets on a regular basis, I can (almost) see this as being relevant.

Why almost?

However, we are currently achieving a service level of 30% of calls answered within 20 secs and, consequently, have an abandoned rate of approaching 20%. Bearing in mind the number of retries and that the majority of callers will eventually be answered, what % of abandoned calls should be used & why??

Initially how are you forecasting your staffing requirements?

I’m not quite sure how you’re getting where you are? Total calls offered
is normally used for setting a gross FTE level and not for actually pattern forecasting. For that you need to map the ½ hr call arrivals

Also your question on abandoned percentage to be used. The short answer is whatever your target is use in the calculation. We use 2%, 5% is common; I’ve even seen it as high as 10%. Remember, as your staffing gets more accurate the ABA level will drop in line with the reduction in repeat calls. However the hard part is to do this cost effectively without producing a DSM environment.

I’ve been meaning to do this for a while so…

Dave’s method for setting a forecast.

1. As a starting point, gather 6 weeks worth of data in ½ hr intervals.
2. Map these into a matrix and calculate the average calls for each ½ hr interval for the week.
3. Work out your CURRENT Average Handle Time (AHT), including after call work.
4. Decide on the metrics you are going to use (AHT, ABA, SVC Level)
5. This is where it gets tricky
6. This then needs to be put through an Erlang calculator to give you a required MINIMUM staff per ½ hr period to handle calls only.
7. Back to data gathering.
8. Collect data on the amount of time spent on Team meetings, training, holidays, sickness levels, lateness (1%) etc…
9. This will give you a % of login time used for ‘work’ that an agent is not available on the phone.
10. Apply this % on top of the staffing required, some will be throughout the day, some will apply to core hours only.
11. From this you will get a ‘staff required’ figure.

With thanks to Dave

Clarification

I think I need to clarify what I am after. I do understand the fundamentals of forecasting having been involved off & on for 10 years. Yes, it is a staff required figure I am trying to get but my question is aimed at the input to the forecasts (i.e call volume) rather than the output.

As I mentioned, we are experiencing poor levels of service & consequently our abandon rate has increased to around 20%. If I use the following scenario it may help to clarify matters:

Calls answered – 10,000
Calls abandoned – 2,000
Calls Offered – 12,000

When determining what call volume should be used in the forecast process, conventional wisdom would seem to suggest that 12,000 is the figure to be used. I would argue that this figure has been artificially inflated by the poor TSF and, due to the number of retries & the probability that the majority of the callers who abandoned will ring back until they are answered, the ‘true demand’ is considerably lower than 12,000. The question is what % of abandons do other listers use to forecast call volumes and does it vary depending on TSF or abandonment rate levels??

If I understand your response correctly (which is by no means certain!) then it appears that you advocate using 2 separate figures, as follows:

FTE (headcount) forecasting – calls offered
Interval forecasting (for scheduling purposes) – calls answered + target abandonment rate.

We operate in an 80/20 service level objective which would lead to a target abandonment rate of about 5%. Are you suggesting that I should use 100% abandons when forecasting headcount and 5% to forecast interval requirements? If so, I can’t understand the logic in this so could you please explain.

Thanks for your assistance and I hope that this helps to clarify my issue.

With thanks to Chris

Abandon Rates Used In Forecasting

Sorry for the confusion.

FTE (headcount) forecasting – calls offered
Interval forecasting (for scheduling purposes) – calls answered + target abandonment rate.

As far as I’m concerned you have a target of 12,000 calls
to be answered.

All you need after that is what YOU deem an acceptable ABA
rate (See my figures above for examples).

In both models I described you use total calls, however for a long
term FTE (Annual / Semi-Annual) forecast you take TOTAL calls
and work with a 0% ABA rate.

For scheduling you accept that in the real world you WILL lose
calls and factor that in.

In all cases it is the total amount of calls presented.

In the case you outlined you MAY have an artificially inflated
call volume due to retrys etc.. The only way to factor this in is
going to be a CLI / Repeat call report from your $TELCO.

In response to the second part of your post. I allow and factor for 2%
ABA.

With thanks to Dave

Fond Memories of Auckland

You are bang on here, working from your example you have 10,000 calls to answer and you should staff to answer this volume rather than the 12,000 calls offered.

Personally I exclude abandoned calls from forecasting the staffing requirement as the models and WFM’s we use are all erlang based and therefore don’t take abandoned calls into account. However Dave’s idea to mitigate with about 2% on top has merit as a “belt and braces” approach, sadly not something those of us in the Outsourcing field can regularly get away with!

With thanks to Eamon

Eamon

Surely you staff to the calls you expect to receive?

The Erlang model will work if you take calls presented
as the base figure to staff from.

I may be missing something obvious but
in the example below Chris is saying that
he has 12,000 presented and 10,000 answered.

That gives an ABA rate of 16.5(ish)%

If you staff to that you are saying that that ABA
rate is acceptable?

Please correct me if I’m wrong but I can’t see how staffing with an
acceptable ABA rate that high is a good thing.

Unless of course I’m going off in the wrong direction here (and
probably am).

I appreciate that some of the 2000 ABA calls will be repeats however
to drive down the ABA% when you don’t know the repeat % you have
to staff to the expected call delivery.

The only exception I can see (and know from experience) is if a
client forecast is way under what actually arrives. in that eventuality
you are on ‘Best endeavours’. Even if the client says 10,000 calls
you get 20,000 delivered and answer 15,000, you’ll still get thrashed
by the client even though you answered 150% of forecast because all they’ll
see is a 25% ABA rate.

With thanks to Dave

Dave

The erlang model won’t work as if you are basing it on 12,000 calls (2,000 of which are abandoned) you will overstaff as the vast majority of the 2,000 abandoned calls are repeats.

The obvious bit that you are missing is that if you are answering 10,000 calls and abandoning 2,000 then you are not staffed to hit your required service level for those 10,000 calls. Increasing the staffing to hit the required PCA will not change the number of calls to be answered, simply speed up the answer rate.

With thanks to Eamon

Factor In Your Estimated Retrys

The key is to understand your ‘retry rate’ and then to forecast to answer these calls on the first try.
The retry rate could be substantial given your current GOS, with only 30% answered in 20 seconds the average wait time is likely significant portion of the 2000 abandons. Nothing can be more frustrating then trying to get through to a call center when the wait times are long.

Look at your telco reports, sort by origination number and determine what % recall within a 24/48 or 72 hour period. Depending on the nature of the call it might not have a high degree of urgency and can extend the ‘retry’ window. Compare the call durations of the retry calls with the average wait time during the same period to determine where their threshold (pain tolerance) is.

Of course some of these calls will be completed and this analysis will give you a rough estimate of First Call Resolution…if they have more than one completed call in say 48 hours you probably didn’t fix the problem the first time.

With thanks to Colin

% Abandons – A Case History

Thanks to you all for responding.

Dave – I do understand what you are saying but do tend to lean towards Eamon’s view, just not to that extreme. I wouldn’t advocate using 0% abandons nor would I recommend using 100%. Abandonment rate is not a measure I hold a great deal of regard for, as it is an outcome of other more important measures like TSF & ASA and is affected by the factors that influence caller tolerance. It is worth noting that it is not the forecasting process that is contributing to poor TSF & high abandonment rates but management, in their infinite wisdom, have initiated a recruitment freeze together with an ever increasing work time.

Colin – yes, you have added value and to elaborate on your point, let me give you a history of the % abandons we have used in the past:

– in the beginning….it was deemed that 100% abandons would lead to overstaffing and 0% would lead to understaffing. So, the answer was obviously somewhere in the middle….and you don’t get much more in the middle than 50%!!

– some time later….a one off exercise was carried out to analyse CTI data to determine the ‘true’ abandon rate i.e. those callers that didn’t ring back and eventually get answered. In this instance, it was determined that 10% of callers were true abandons and the formula was changed to answered + 10% abandoned, except for higher value callers where it was maintained at 50%.

– later still….a change in regime led to changes in the forecast process, including using calls offered. They also used to forecast to target to give you an idea of the dubious practices implemented at this time. So, now we are at 100% abandons.

– just recently….a few of the ‘old guard’ questioned the process and a full review was conducted, including a consultant. This led to an arbitrary figure of 25% being used. This is where we are currently at and would be keen to see if others have gone through a similar process.

So, for those of you that have bothered to read this far & are still awake, what are your thoughts? As both Dave & Colin have pointed out, it would be beneficial to delve into the murky depths of the ACD reports to determine what the true abandon level is at this point in time. Would the figure be the same if we were achieving target SL of 80%, rather than the 30% that we are currently experiencing? I would suspect that it isn’t.

With thanks to Chris

Forecasting

Taking Team Meetings or agent coaching happens on real time basis. How would you take this in forecasting? We basically allow team meetings and other related stuff only on non peak hours.

With thanks to Karthik

How Long is A Piece of String?

It’s been a while since this thread, but to add my view…

I fully see where the original point was going and Eamon’s subsequent response is the side of the fence I fall. The problem with coming up with a figure here is that it would be purely arbitrary without fully understanding the patience time, retry rate etc. The answer also depends on company culture:

a) sometimes Dave’s approach is the best one – shock treatment can often get budgetary adjustments whereas trying to stretch the extra inch can prove difficult – “why should I pay to employ the extra 10 people you suggest if I only get 1% more calls answered?

b) sometimes using only answered may get you some extra staff if you full well even the best supported business case is unlikely to see the light of day if it suggests radical expenditure. While not ideal, something is better than nothing, especially for the majority of your customers praying for better service (by service I am using answering speed while understanding in reality so many more things could be important). The rapid rate of improval in answer speeds as you add small amounts of staff are such that this approach could actually be a godsend.

Hope that adds a little to some good points already raised.

With thanks to Robert