Embed Size (px)
Citation preview
WHAT-IF, COST VERSUS SERVICE, AND
VARIANCE ANALYSIS
Ric Kosiba
Your Seminar Leader
Ric Kosiba is vice president and founder of the Interaction Decisions Group at Interactive Intelligence. He founded the company, Bay Bridge, in 2000 that was acquired by Interactive Intelligence in August of 2012.
Ric has been involved in the call center industry 20+ years and enjoys working with contact center analysts and helping companies maximize profitability through math modeling. He holds a B.S.C.E., M.S.C.E., and Ph.D. in Operations Research and Engineering from Purdue University (go Boilers!).
Ric is responsible for the development and enhancement of our contact center capacity planning and analysis product line. He thoroughly enjoys working with our brilliant development and operations research team, which helped Interaction DecisionsTM become the leading U.S. supplier of long-term forecasting and planning solutions.
Ric resides in Maryland with his wife and four children. He loves being a dad and enjoys coaching his kid’s football, basketball and lacrosse.
Purpose: Make Informed Decisions!
• Variance Analysis: What is changing, and will it mess up our
operation?
• Sensitivity Analysis: How much does a (single) change from a
performance driver affect my performance?
• What-if Analysis: We need to evaluate one or several business
scenarios (operational performance, finances, business risk)
4
What are common methods for performing what-if
analyses?• Ad-Hoc
– Back of the envelope (VERY common!)
– Data analysis: conducting data merges and averages
– Ad-hoc Erlang analysis
– “Peak week” analysis
• Computer Modeling/Capacity Planning Methods– Erlang capacity planning spreadsheets
– Regression models/spreadsheets
– Workload models/spreadsheets
– Simulation capacity planning spreadsheets/systems
5
Staff Available Vs. Service Level
0.00
0.20
0.40
0.60
0.80
1.00
1.20
0.00 100.00 200.00 300.00 400.00 500.00 600.00 700.00 800.00 900.00
Staff Available (FTE's)
Ser
vice
Lev
el
Why do we need a computer model? (Why doesn’t data analysis work?)
This is real-world data plotting
staffing against service level.
Why is it so ugly?
Let’s look at call center data:
Each data point represents a different economies of scale, different handle times, different agent efficiency, different schedule efficiency and adherence, different distributions of behavior…
6
Isolating for Call Volume
Staff Available Vs. Service Level
0.00
0.20
0.40
0.60
0.80
1.00
1.20
300.00 350.00 400.00 450.00 500.00 550.00 600.00 650.00 700.00 750.00 800.00
Staff Available (FTE's)
Serv
ice
Leve
lEven isolating for call
volume yields a pretty ugly
graph: Averaging these
values, or using raw
“history” is not an accurate
method for predicting
performance
Many analysts would try to
lean on regression models,
would build a “best fit”
line…
So how can you do these sorts of analyses?
• When data is hard to interpret or is messy– analysts rely on
models– Oldest is the Erlang family of equations
– Staffing equations (like workload equations)
– Regression models
– Simulation
• You need a few things– Data and forecasts
– A model, and proof that your model is accurate (validation)
– Financials
– Scenarios to evaluate
Simulation is used because it is accurate
0.00%
10.00%
20.00%
30.00%
40.00%
50.00%
60.00%
70.00%
80.00%
90.00%
100.00%
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29
SER
VIC
E LE
VEL
(%
)
DAY #
Preferred Service: Service Level Comparison Simulation vs. Erlang-C vs. Actuals (Weekly Summary)
SL Actual SL Sim SL Erlang
Call Type
Simulation Service Level Prediction Erlang-C Service Level Prediction
Avg. Err Avg. Abs. Err
Std. Dev. Abs.
ErrAvg. Err Avg. Abs. Err
Std.
Dev.
Abs.
Err
Loans 0.59% 0.78% 0.65% 22.42% 22.42% 4.61%
Member Services 0.24% 1.19% 1.20% 25.97% 25.97% 5.56%
Preferred Services -1.27% 2.21% 1.75% 24.11% 24.11% 4.67%
Retail 0.86% 2.66% 1.39% 3.36% 4.03% 4.85%
Credit Card 0.31% 1.01% 1.08% 9.99% 9.99% 3.41%
Auto Insurance 1.20% 2.45% 2.27% -3.46% 3.46% 1.59%
Average 0.32% 1.72% 1.39% 13.73% 15.00% 4.12%
Tip: Validation of your analytic process breeds confidence in both your analyses, and you! Make validation a regular part of your planning meetings
Discrete-Event SimulationWhen the QB passes:• Is he being rushed?• Is he running?• Is it a short pass?• A long pass?• Is the receiver covered?• Does the receiver have
good hands?
What are the odds that the pass will be completed? Or Intercepted? Or dropped?
Inbound (Chat or Call) Simulation
Simulation evaluates service and abandons associated with a scenario by tracking
simulated call by call performance
Abandon
< SL Threshold
> SL Threshold
When a call comes in:• What is the statistical
distribution of patience?
• Of handle times?• Is an agent available? If
not, when will they be available?
• How much time left on the (sampled) patience?
What are the odds that the call will be completed? Or Abandoned? How busy are my agents (occupancy)? What is my longest queue length?
What is my tallied service level? Abandon rates?
Each caller has a sampled patience factor (from the distribution of real patience), and a sampled handle time
Hooking simulation into a capacity plan to drive what-if analyses
Change staffing (or volume or AHT or sick or…) and see what happens to service level, abandons, speed of answer, CustExscores, occupancy, costs, revenues…
Change service level desired (or ASA or abandons desired) and see the staffing required
Simulation drives staffing requirements
Sensitivity Analysis
Performance Driver
Perf
orm
an
ce
Sensitivity Analysis:
In your simulation model, vary your performance driver, holding every other variable constant, to see the affect on performance!
Sensitivity Analysis
Plotting call volume against service level, holding everything else constant
A quick aside… cost of shrinkage
Getting shrink correct is as important as getting volumes right!
16
Let’s do a what-if: Decrease ASA goal?
• A consultant has determined that to be “The Best” your company
should have very high Average Speed of Answer. New target: 3
second ASA.
• You’d like to determine the true optimal value for your sales function
Example: Is the ASA goal that our consultant
suggests optimal?
17
Staffing Vs. Average Speed of Answer
Our current goal, a 10 second ASA seems too low to the
consultant.
What if we decrease ASA to 3 seconds? Our staffing
would need to increase by 23 FTE.
What would happen to agent occupancy?
23 FTE
18
Staffing Vs. Occupancy
If we increase staffing by 23 FTE’s,
our occupancy drop to 73%.
What about Abandons?
23 FTE
7% decrease
19
Staffing Vs. Abandons
Increasing staff by 23 FTE would decrease our
abandon rate by 0.75%. What about our sales?
0.75%
20
Staffing Vs. Sales
$30,000/Mo.We could estimate that sales would increase by $30,000 for
the month. But what about our costs? Our center
“profitability”?
21
Staffing Vs. Profit (the coolest sales graph ever)
Staff increase reduces profit by $26,000/mo.
Considering employee costs, the profit-
maximizing staffing is 189 FTE. Staffing
costs swamp the expected improvement
in sales by $26,000 per month. The
consultant is wrong
22
Determining Your Goals Using Sensitivity Analysis
You can analyze your service goals by
determining your goal versus your costs.
Note: this requires both operational and
financial analyses
23
What-if: What is the best balance of hiring versus overtime?
Also must consider site
capacity, ability to hire,
classroom size, etc…
Need to look across
the entire network
and over seasonal
peaks and valleys in
order to meet
service goals
You can use optimization
technologies, such as integer
programming to determine your
optimal hiring and overtime
plans.
24
Long-Term Staffing Tradeoffs
Total Fixed and Variable Cost for a Sample Call Center
Given Different Staffing Strategies
5700000
5720000
5740000
5760000
5780000
5800000
5820000
5840000
5860000
5880000
5900000
40% 45% 50% 55% 60% 65% 70% 75% 80% 85% 90%
Hire to Percent Answered in 20 Seconds
Cos
t (D
olla
rs) Staff to 40% Service Level
Overtime to Hit Goal
Staff to 70% Service Level
Overtime to Hit Goal
Staff to 80% Service Level
Undertime to Reduce Costs
During Slack Weeks
Staff to 90% Service Level
Undertime to Reduce Costs
During Slack Weeks
Staff to 60% Service Level
Overtime to Hit Goal
What-if: Have you ever tried to tame marketing?
• Marketing seems to be the operation’s bane– Uncoordinated mail drops– Experimental offerings– Catching the call center
unaware with heavier volumes or more complex phone calls
• … and they never, ever call us first! How Operations Views
a Marketer
There are times to market, and times to not…
But, this is a terrific time for a marketing event!
Given seasonality, call center demand fluctuates. This is a terrible time for a marketing event
A little coordination goes a long way
This company involved, in their regular, monthly (best practice) planning and decision making meeting, marketing (also, best practice)
– The operation alerted marketing when they had little capacity for events
– They also alerted marketing when they had excess capacity to be used for marketing events
– AND, marketing used this venue to alert the operation to their plans
Marketer, after they call us
With a validated model, what-ifs are unlimited
• “ What if we increase or decrease the hours of Operation.” How could we get the tool to tell us how
much more staff we would need for this.
• Determining service standards for sales centers
• Evaluating the importance of your shrinkage forecast
• Planning for uncertain events
• Analyzing sit-withs and buddy training programs
• Hiring versus overtime policies
• What service standard should we choose for each contact type?
• What happens if we invest in new technologies that decrease our handle times?
• What happens if we add a new cross-sell program that will increase handle time and increase
revenues?
• What if we change our service goals?
• What if our new marketing program generates more calls than expected?
29
Variance Analysis
Note! Wages are growing relative to the forecast!
Variance analyses are our canary in the coalmine
Why is there variance?
1. The forecast may be off– a math or process error
2. Something external may have changed the mix of contacts or agent performance
3. Something internal may have changed the mix of contacts or agent performance
4. It could be a random occurrence (a blip)
A very appropriate role of your forecasting process is to be an early warning signal. And variance is often the impetus for many what-ifs!
Is it error or is it change?
Error
• Can I tweak my model to make it
better?
• Hindsight: Was it my modeling choices
that made the difference in accuracy?
• Nobody understands what’s different
this month
Change
• Or do I have to throw away my
model for something radically
different?
• Hindsight: Did I have any way of
forecasting this right?
• We are hearing about the change
(from the agents, from the news,
etc…)
What many management teams view as error is really change! The best management teams view error as change!
How can you provide great analyses?
You need a capacity planning process that is both quick
and accurate!
Automate, optimize, and validate!
What is Interaction Decisions?
• Decisions is a long-term contact center strategic planning and what-if analysis system. It helps to:
• Because it is fast and accurate:– Perform risk and sensitivity analysis of your contact center
– 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
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 Decisionsdemonstration
Also! We have white papers all about planning and analysis at www.inin.com
(look for Strategic Planning)
