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

Ric.Kosiba@InIn.com

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)