Download - Contact Center Forecasting… and
Contact Center Forecasting… and
Reforecasting
Ric Kosiba
Very large bank
Very analytically driven
organization
Hires mathematicians as
forecasters
During a loan loss scare,
relied heavily on their
forecasting team to predict
business drivers
A “Not Atypical” Forecasting Story
Forecast Versus Actual
0
20,000
40,000
60,000
80,000
100,000
120,000
J F M A M J J A S O N D
Date
Vo
lum
e
Forecast
Actual
Month One: There is a contact volume variance to plan!
Volumes
significantly higher
than expected!
A New Forecast!
0
20,000
40,000
60,000
80,000
100,000
120,000
140,000
J F M A M J J A S O N D
Date
Vo
lum
e
Month One Decision: Reforecast!!
New forecast builds
off of the new trend-
more volume!
Reforecast Versus Actual
0
20,000
40,000
60,000
80,000
100,000
120,000
140,000
J F M A M J J A S O N D
Date
Vo
lum
e
Reforecast
Actual
Month Two: There is a contact volume variance to plan!
In February,
volumes significantly
lower than expected!
Re-Reforecast
0
20,000
40,000
60,000
80,000
100,000
120,000
J F M A M J J A S O N D
Date
Vo
lum
e
Re-Reforecast
Month Two Decision: Re-Reforecast!!
New forecast builds
off of the new trend-
less volume than the
last reforecast!
Re-Reforecast Versus Actual
0
20,000
40,000
60,000
80,000
100,000
120,000
J F M A M J J A S O N D
Date
Vo
lum
e
Re-Reforecast
Actual
Month Three
In March and April,
volumes significantly
higher than
expected!
• Assuming the mathematicians are good (they were), for
them, at this time, forecasting contact volumes was simply
unpredictable
• Possibly external or internal changes were transforming the
historic seasonality (e.g. heavy marketing in non-traditional
times)
o Forecasters were likely missing critical information (these
external or internal changes)
o Even with an accurate picture of what was changing,
they did not have history (because a loan loss spike
didn’t occur often) on how their customer would respond
(through additional contacts)
o Therefore, they had no data that would enable them to
accurately predict their contact center performance
drivers
o They had an unstable forecast
What do we know about their forecasting process (or operation)?
We forecast to:
o Put together investment plans and allocate resources (staffing,
technologies, cost to service, etc…)
o Answer what-if questions (and the end result of a what-if is still a plan!)
o Forecast variance serves as an early warning sign
The best companies view their forecast as their baseline
The best companies view forecast variance as an anomaly to be explored
(internally caused, externally caused, controllable, uncontrollable, to be
monitored, etc…)
The art of forecasting itself becomes less important when there is a lot of
uncertainty (because history is not a great predictor)
The art of strategic business planning (enterprise analytics) becomes much
more important when there is a lot of uncertainty
The point of forecasting is not to forecast well, it is to make better decisions
Why do we forecast?
We forecast primarily our contact volumes and handle times (see SWPP
forecasting survey)
Many use workforce management algorithms (usually annual weighted
average)
Many use spreadsheets and proprietary methods (such as regression
models with external data)
Many (not most) focus primarily on a short time horizon (say 90 days)
Most measure error using standard statistical methods (RSME, MAE,
etc…), but only tactically (error out 30 days, say).
Current “state of contact center forecasting”
Smart analysts would use their forecasting information to
help make better decisions. They know this:
oTheir operation is in a state of flux
oTheir forecasts are unstable
oThe company still requires resource allocation decisions to
be made
They can use their strategic planning process as their
analytic engine and still make good strategic decisions
… Back to our example, what do we do?
The Three Forecasts
0
20,000
40,000
60,000
80,000
100,000
120,000
140,000J F M A M J J A S O N D
Date
Vo
lum
e
Forecast
Reforecast
Re-Reforecast
So how can we use these three forecasts?
How about as
bounds to what
might happen?
These three forecasts represent real possible scenarios!(each of these forecasts were considered at one time “the official” forecast)
When there is serious uncertainty, and accurate forecasting is beyond the art, the
forecasting exercise may become one of putting bounds on the range of possibilities. For
example:
o Is it conceivable for volumes to drop 20%?
o Is it believable that the economy will turn around in Q4?
o If so, then these scenarios are very possible and should be evaluated
Forecasts then are judged very differently: the entire exercise is not about “correct”
forecasts but has shifted to “what do I do?”
o Forecast error is inconsequential (nobody believes the forecasts anyway)
o Alternative forecasts exist as a statement of possible outcomes (not necessarily probable
outcomes)
o These alternatives must be evaluated against risk to the business
o Business risk can be determined during the planning process!
Probabilities Versus Possibilities
What Happens(Usually out of your control)
What You Planned
For: Your Decisions(Completely in your control)
The Resulting Performance
and the Business Risk(You can calculate this!)
… let’s plan for risk!
Planning for business risk
What Happens Your Business Decision Result/Risk Probability
Forecast 1 is correct
(Low Volumes)
Staff to Forecast 1 (L) Low costs, consistent service 15%
Staff to Reforecast (H) Very high costs, over-service, until trend
is realized
Staff to Re-Reforecast (M) Higher costs, over-service, until trend is
realized
Reforecast is correct
(High Volumes)
Staff to Forecast 1 (L) Very high overtime costs, horrible service,
possible service catastrophe
35%
Staff to Reforecast (H) High costs, consistent service
Staff to Re-Reforecast (M) High overtime costs, poor service,
possibly recoverable in a few months
Re-Reforecast is
correct
(Middling Volumes)
Staff to Forecast 1 (L) High overtime costs, poor service,
possibly recoverable in a few months
50%
Staff to Reforecast (H) High costs, over-service, until trend is
realized
Staff to Re-Reforecast (M) Consistent service
Results are evaluated by producing full-on staff (hiring and overtime) plans and budget plans. Operational performance
(service levels, ASA, abandons, sales, occupancy) is simulated and costs are known. Choose the best plan!!
1. Monitor the plan for variance
2. Determine the impact of the variance to the network’s performance
3. Reforecast and determine the range of “possibilities”
4. Determine the effects of different management responses
5. Choose the resourcing decision that is appropriate for your company’s risk
tolerance
How do we decide to change the official forecast?
Sensitivity Analysis
A great way to measure the impact of a performance driver
What is the cost of variance?
Every new plan has a corresponding cost (or benefit)
A quick aside… cost of shrinkage
Getting shrink correct is as important as getting volumes right!
In an environment where forecasts often change,
forecasters can lose credibility
It is important that as analysts that we focus on the correct
things
o Forecasting is a single step in solving a bigger problem
(resource allocation)
o Investing in the accuracy of the next steps- strategic
planning- will allow you to produce amazing analytics even in
the face of uncertainty and forecast unreliability (e.g. risk
analyses)
o Focus your analytics on the decisions you’re making,
resource allocation, and not forecast accuracy
Automating and optimizing the next steps of the planning
process will improve your planning accuracy and will
enable:
o Developing forecasts then capacity plans then budgets in
minutes
o Real-time, interactive, and accurate what-if analysis
o Rapid scenario and risk analysis
Analyzing business risk should impress the big boss!
How to deal with “error” and variability
You need a capacity planning process that is both
quick and accurate
How can you provide risk analysis?
1. Requirements building
Most companies still use Erlang equations to determine staff required by
week. It is commonly known that Erlang has an overstaffing bias
Some companies use an occupancy forecast/estimate: but occupancy is
an output of your decisions!
Try using discrete-event simulation
2. Hiring/overtime/undertime planning
Most companies produce over/under charts (based on Erlang) and
“eyeball” when to hire
Try using integer programming
3. Scenario evaluation
Erlang is inadequate for evaluating scenarios (it is inaccurate)
The standard Erlang spreadsheet is too slow and error prone
Try automating!
Traditional capacity planning
There are better ways
With uncertainty,
determine the range of
possible scenariosVariance is
your early
warning signal
How can the
resource
decision go
wrong?
Strategic Planning Systems
Predict the
Future
Analyze
Impact
Set Goals
Build Staff
Plan
Create
Financial
Plan
Perform Risk
and
Sensitivity
Analysis
Compare
Performance
To Plan
Strategic
Planning
System
Try sensitivity
analyses!
What are the resource
decisions available?
Because all of the business uncertainty, NOBODY,
including you or I, believe that the following mathematically
rigorous and highly impressive forecast is correct
However, since our forecasts are bound to incorrect, we
have provided you the following business risk analysis to
help us determine how to allocate our resources
(free cool Decisions t-shirt to anybody who shares the reaction of their
boss with me!)
My dare to YOU: produce the following slide
At a large financial services firm, the marketing department wanted to create
separate small teams of agents, so when their customers called, they might
hear a familiar voice
o The value of “one to one” marketing was thought considerable
o Customer retention was a significant concern
Their staffing process was not consistent or optimal
o They used Erlang-based over/under charts and eyeballed when and where
to hire
o It took a lot of time to produce a hiring plan!
o Their spreadsheets were not designed to answer the particular question
[Aside: I had a laptop with an enterprise planning system and so (over chowder
and beer at the local airport) we created the following chart…]
Varying Team Size at a Large Financial Firm
Varying Team Size at a Large Financial Firm
24
0
20
40
60
80
100
120
140
160
180
0 10 20 30 40 50 60
Calls H
an
dle
d P
er
Ag
en
t
Team Size
Calls Handled Per Agent in Order to Maintain Service Goal
Assumptions:Service Goal: 85% of all calls handled in 20 secondsTeams do not share calls significantly13,024 calls per weekHandle Time = 425 secondsShrinkage (vacation, sick, etc…) = 30%
Note: With no teams, calls handled per agent is 226
More “Specialization” and “personal” service
Effic
iency
Answer: It is VERY expensive to specialize your agents
Once I get over ~75,000 calls, I am out of economies of scale!
Varying Team Size at a Large Financial Firm
During “interesting times” strategic planning is critical: Strategic what-if questions
are everywhere. Solid strategic analysis will greatly improve your operational and financial
performance.
Planning and forecasting is about decision-making. Everything else is noise.
Don’t fear the math: Mathematical modeling techniques like integer programming and
simulation may look difficult, but it is worth doing the extra homework (or buying the
services of an expert)
Speed matters: Automating the strategic planning process will, not only allow you to see
your family during budget season, but it will also provide significant value to your
companies. Risk analyses help tell us what to do!
Final Thoughts
Decisions is a long-term contact center strategic planning and what-if
analysis system. It helps to:
Because it is fast and accurate:
o Perform risk and sensitivity analysis of your contact center
o Evaluate center what-ifs: investments, consolidation, and growth
opportunities
Decisions complements traditional workforce management software by
focusing on strategic decision making and long-term planning
What is Interaction Decisions?
ForecastRequirements
SimulationStaff & Capacity
Plan OptimizationBudget
Contact Us!
Ric Kosiba
410-224-9883
… if you would like a copy of the slides or to see a
quick Decisions demonstration
Also! We have white papers all about planning and analysis at
www.inin.com (look for Strategic Planning)
Also, also! Stop by our booth next week at Society for Workforce
Planning Professionals conference!