вход по аккаунту


How to Forecast Work That Is Not a Phone Call - Human Numbers

код для вставки
forecast focus / oct 2011
How to Forecast
Work That Is Not a
Phone Call
Today’s centers handle
multiple types of contacts. Key
considerations for chat, text, web
and IT support calls.
by Tiffany LaReau, Human Numbers
Pipeline Articles
How to Forecast Work That Is Not a Phone Call
Tiffany LaReau
Human Numbers
fter you master the art of call volume forecasting, your reporting
audience may ask you to produce other forecasts of seemingly similar workload that does not arrive by phone. If you insist on accurate forecasts you must resist the urge to treat this as shrinkage, and should not recycle
your same phone forecasting models, or else you may suffer the bane of inflated results.
In a phone volume forecast, a special Erlang calculation is used to account for the random
arrival of calls, which, in turn, swells your base staff (“bums-in-chairs”). How much this swells
depends on the service goal, and is designed to work at an interval level, usually measured in
seconds or minutes. Non-phone contacts that have service goals (or “time-to-process” goals)
that are greater than a two-hour turnaround won’t perform well against a forecast using
Erlang, so the model needs to be backward-engineered a bit to work properly.
You may be thinking, “What about chat, text and web contacts that use small-interval
service goals?” When these are handled as single-tasked, start-to-finish contacts, then you’re
right—the call volume forecast is completely interchangeable. But the more common
approach with these is to have an agent multitasked on more than one chat session at a
time. Multitasking is not the same thing as multiskill, which means an agent is able to handle
different types of contacts. Multiskill means an agent can take more than one type of customer group, whereas multichannel describes the point of contact: phone, web, text and fax.
Erlang Can Still Be Used
to Calculate Chat
…the only difference that you need to consider is the fact that multiple chats may be
happening at the same time. This has two impacts on the calculation:
1. In the formula, when it’s time to enter the volume (which is usually the 30 min vol x 2,
or the 60 min volume x 1), you would actually take that final volume calculation and divide
it by the total number of calls an agent can chat at the same time.
So, if the normal Erlang calculation is fractional agents=(SLA, Service Time, Calls Per
Hour, AHT)
But your agents take 3 chats at the same time, the new Erlang calculation is fractional
agents=(SLA, Service Time, (Calls Per Hour)/3, AHT)
And remember, if you’re using 30-minute intervals instead of 60-minute intervals, the
calculation becomes fractional agents=(SLA, Service Time, (Calls Per Hour*2)/3, AHT)
2. When multiple chats happen at the same time, there is less real-time interaction and
more lag time in between sentences to/from the agent-customer—this means AHT may be
higher, so don’t automatically assume that the chat AHT is equal to a phone AHT. And yes,
the increased handle time WILL negate the divisor gained by multitasking.
The other thing to consider when multiple chats are happening, and there is increased lag
between the agent/customer, is that quality may suffer—customers are becoming smarter
about the fact that the agents are not paying strict attention to them, they’re copying/pasting scripts, etc. and it’s creating a more aggressive customer. No one enjoys being ignored.
If you’re planning to forecast for social media contacts, be aware that it comes with a lot
of extra noise due to ads, friend requests and other time-sucking distractions. Social media
has a way of making people believe that they are starring in their own reality show. This can
be captured in your forecasting models as a loss of agent occupancy time, or a reduction in
In addition to Erlang inflating required staff, it also fails to use some reality thresholds in
Pipeline Articles
How to Forecast Work That Is Not a Phone Call
centers with longer handle times because there are simple cases where it produces more
people than your forecasted volume. Required staff should never exceed the forecasted calls;
after all, two people cannot team up on one call to make it go faster.
Erlang Fail…
4 calls, 23-minute AHT, 80% SL in 30 seconds, Erlang requires 5.3 people
That’s more people than calls!
Forecasting for an IT Department
…is one of the most interesting variations. The results are very similar to the look and feel
of a call staffing models: They still need to reflect workload, presence/utilization, and actual
vs. required staff, because these are the three common elements to determine when you
have extra capacity or if you are understaffed. But there are some important differences in
building a staffing model for an IT department vs. what you are accustomed to experiencing
from your old call center models. First, calculating workload is not as simple as multiplying
the number of IT tickets times an average handle time because there rarely is an “average”
with handle times for IT personnel. For example, two similar tickets could range from 30
minutes to 30 days in IT depending on priority and severity levels, the availability of parts,
and many other odd factors that make this group unique.
The approach we like best is to use an IT Design Factor. This method includes analyzing the
following components of IT workload to establish the current level of efficiency (baseline):
Ratio of systems to IT reps—as new systems are introduced (software, hardware, etc.)
the IT burden increases to support these from behind the scenes.
4-wk avg Smoothed
Will Clean
Up the
Spikes and
4-week history
Ratio of employees to IT reps—as employees increase, the IT burden increases at a
comparable rate.
Pipeline Articles
How to Forecast Work That Is Not a Phone Call
IT Process Efficiency—rates the team on how effective they are able to work within
their current confinements on a scale of good-average-poor.
IT Ticket Volume—seasonality is critical for forecasting, specifically focused on
recurring times during the year when IT tickets spike, as well as a review of the
conditions that existed if your time-to-process goals were missed.
Once a baseline is established that demonstrates where you are now, your leadership
team can set goals defining where you want to be. This may include expanding or reducing the ratio of employees to IT reps, adding process improvements, or redefining system
support requirements. Each of these “what-if” assumptions will impact the total headcount
requirements, and your staffing models should quickly show you the resulting +/- result
from any changes that you anticipate.
After collecting the workload components, you can revert back to the normal and familiar
staffing model methods (collecting vacation, meetings and sick time to establish shrinkage,
etc.) to show Presence & Utilization. There is a small piece that is different for IT in regards
to training shrinkage, because it is an ongoing regular occurrence due to the nature of the
work, necessary to occur at the IT reps’ discretion rather than formally scheduled at specific
times. When all of these things are in place, you can make decisions about when to allow
attrition to occur without rehiring immediately, and when you need to start posting jobs.
How do you know when attrition will happen?
Any metric that has been measured and captured historically can be forecasted, including
attrition. Each individual metric has unique considerations, and for attrition that includes
separating the trending patterns from full-time, part-time, and temporary employees. Like
raw ACD data, this history may also need to be normalized, especially if there were exceptional situations that led to large hiring periods or layoffs. To give it a little more finesse, you
can also apply a smoothing curve.
Smoothing Normalizes Peaks and Valleys
Non-phone forecasts will benefit from smoothing because the forecast accuracy conditions and volume drivers are more vague than predicting phone calls. You may not have
access to detailed interval-level history since there’s no ACD capturing anything for you. The
act of smoothing will add a nice layer of normalization to the high peaks and low valleys in
your history. The formula is quite easy—and MS Excel has a little wizard to walk you through
it the first time. First, you’ll need to load your Analysis TookPak add-in, then you’ll see a new
feature on your Data ribbon called Data Analysis. When you click that, select Exponential
Smoothing and then you can experiment with different results by altering the damping factor. I tend to like a 40/60 split. After you run the wizard the first time, go back and click on
one of the smoothed result cells to study the formula and it will become instantly clear how
easy it is.
Tiffany LaReau is a Certified Workforce
Manager at Human Numbers, a firm that provides
contracted forecasting and scheduling services.
(678) 494-1506
Pipeline Articles
How to Forecast Work That Is Not a Phone Call
About Contact Center Pipeline
Contact Center Pipeline is a monthly instructional journal focused on driving business
success through effective contact center direction and decisions. Each issue contains
informative articles, case studies, best practices, research and coverage of trends that
impact the customer experience. Our writers and contributors are well-known industry
experts with a unique understanding of how to optimize resources and maximize the
value the organization provides to its customers.
To learn more, visit:
This issue is available online at: October 2011, Contact Center Pipeline
Online Resource
Pipeline Publishing Group, Inc. PO Box 3467, Annapolis, MD 21403
(443) 909-6951 вќ�
Pipeline Articles
Без категории
Размер файла
593 Кб
Пожаловаться на содержимое документа