capacity management in call · pdf filecapacity management in call centers ... capacity...
TRANSCRIPT
Capacity Management in Call Centers
Basic Models and Links to Current Research
from a review article authored with
Ger Koole and Avishai Mandelbaum
Outline:
• Tutorial
– background on how calls are handled
– basic model of call-center operations
• Research overview
– system primitives
– basic model
– routing and networking
– further directions for research
1
Tutorial, background...
Typical call-center environments:
Top figure from Larreche et al. (1997).
2
Tutorial, background...
Representative infrastructure for asingle location:
customers PABX
agents
IVR/VRU
ACD
CTIserverPSTN customer
data servertrunk lines
3
Tutorial, background...
Queueing model associated with asingle location:
retrials
arrivals
abandon
queue
busy
lost calls
retrials
lost calls returns
N = 3 CSR-servers
5 = (k – N) places in queue
w = 5 work stations
k = 8 trunk lines (not visible)
Call-center hardware Queueing model parameters
4
Tutorial, basic model...
Traditional model used in capacitymanagement:
• Work arriving over time
• Short-term capacity requirements
• Medium-term shift scheduling and rostering
• Long-term hiring and training
5
Tutorial, basic model...
Number of calls arriving over severaltime horizons:
…each month of the year …each day of the month
…each hour of the day …each minute of the hour
Number of calls arriving…
6
Tutorial, basic model...
Work modelled as arriving in discretetime buckets:
0
5
10
15
20
25
30
8 9 10 11 12 1 2 3 4 5
½-hour interval i
Ri = λi/µ = work arriving during i
7
Tutorial, basic model...
Waiting-time standards drivecapacity requirements:
• ASA = average speed of answer
E[wait]
• TSF = telephone service factor (or service level)
P{wait ≤ T}
• Fraction of calls abandoned before served
P{abandon}
• Determining capacity requirements for interval i:
e.g. Ni = min{N | E[wait] ≤ ASA∗}
8
Tutorial, basic model...
A report with 1/2-hour intervals:
9
Tutorial, basic model...
Performance estimate usesM/M/N/∞ model:
x xx
xxx
x• no blocking, abandonment, or retrials
• fixed arrival and service rates
• exponential interarrival and service times
• measures of stationary performance
10
Tutorial, basic model...
Estimates require only λi, µ, N :
• P{wait > 0}
= 1−∑N−1
m=0 Rim/m!∑N−1
m=0 Rim/m! + (Ri
N/N !) / (1− Ri/N)
(Erlang-C formula)
• P{wait ≤ T}
= 1 − P{wait > 0} · P{wait > T |wait > 0}
= 1 − P{wait > 0} · e−(Nµ−λi)
• E[wait]
= P{wait > 0} · E[wait|wait > 0]
= P{wait > 0} · (Nµ − λi)−1
11
Tutorial, basic model...
Example with 30-second ASA:
0
5
10
15
20
25
30
8 9 10 11 12 1 2 3 4 5
½-hour interval i
offered load idleness for ASA
Ni = number of CSRs working in i
12
Tutorial, basic model...
Determining staffing requirementsfor a day or week:
• Data
Ni = num. CSRs required for interval i
cj = cost of putting a CSR on schedule j
aij ={
1 if sched. j take calls during int. i,0 otherwise.
• Decision variables
xj = num. CSRs to work on schedule j
• Math program to choose schedules
min{c′x |Ax ≥ N ;x ≥ 0;x integer}
13
Tutorial, basic model...
Example with 4-hour shifts on the1/2-hour:
total Ni1 13 121 1 21 211 1 1 21 191 1 1 1 21 191 1 1 1 1 24 201 1 1 1 1 1 24 211 1 1 1 1 1 1 28 201 1 1 1 1 1 1 1 28 19
1 1 1 1 1 1 1 1 17 171 1 1 1 1 1 1 1 14 13
1 1 1 1 1 1 1 1 14 131 1 1 1 1 1 1 1 19 19
1 1 1 1 1 1 1 1 19 191 1 1 1 1 1 1 19 17
1 1 1 1 1 1 15 111 1 1 1 1 15 15
1 1 1 1 13 131 1 1 8 8
1 1 8 8
8:00-8:298:30-8:599:00-9:299:30-9:5910:00-10:2910:30-10:5911:00-11:2911:30-11:5912:00-12:2912:30-12:591:00-1:291:30-1:592:00-2:292:30-2:593:00-3:293:30-3:594:00-4:294:30-4:595:00-5:295:30-5:59 1 3 3
Shifts = xj8 8:30 9 9:30 10 10:30 11 11:30 12 12:30 1 1:30 2
43 13 8 0 0 3 0 4 0 2 5 2E-10 5 3
14
Tutorial, basic model...
Excess capacity resulting from thescheduling IP:
0
5
10
15
20
25
30
8 9 10 11 12 1 2 3 4 5
½-hour interval i
head
coun
t
offered load idleness for ASA slack from shifts
15
Tutorial, basic model...
More on the scheduling process:
• Scheduling IP recommends schedules
• Rostering assigns CSRs to schedules
(shift-bidding)
• Scheduling IP also recommends a number of
CSRs, 1′x
• Target headcount 1′x typically inflated to
accommodate “shrinkage”
• Target headcount 1′x can be used in a
longer-term hiring problem
16
Tutorial, basic model...
Longer-term hiring and training:
• nt = 1′xt
target headcount for scheduling period t
• τ
lead time required to hire and train a new CSR
• Then in scheduling period t hirent+τ −t+τ−1∑
j=t
yj(1− β)τ−(j−t)
+
where
nt+τ = target headcount for hiring today
yj = CSRs that start working in period j
β = 1-period attrition rate
17
Tutorial, basic model...
ACD reporting system drivescapacity planning:
18
Research, model primitives...
The arrival process:
• Time-inhomogeneous process
– Andrews and Cunningham (1995)
– Massey et al. (1996)
– Brown et al. (2002)
– Henderson (2003)
– Avramidis et al. (2004)
– Huang and Shen (2004)
• Uncertain rates
– Whitt (1999)
– Jongbloed and Koole (2001)
– Brown et al. (2002)
– Avramidis et al. (2004)
• Correlation among rates
– Andrews and Cunningham (1995)
– Whitt (1999)
– Brown et al. (2002)
– Avramidis et al. (2004)
19
Research, model primitives...
Service times:
• Exponential distribution
– Kort (1983)
– Harris et al. (1987)
• Gamma distribution
– Chlebus (1997)
• Lognormal distribution
– Bolotin (1994)
– Mandelbaum et al. (2001)
– Brown et al. (2002)
20
Research, model primitives...
Abandonment:
• Earlier work on impatience and abandonment
– Palm (1943)
– Kort (1983)
– Roberts (1979)
• Abandonment behavior in a call center
– Mandelbaum et al. (2001)
– Zohar et al. (2002)
– Brown et al. (2002)
21
Research, basic model...
System performance and limits:
• Many-server limits for M/M/N model
– Halfin and Whitt (1982)
– Borst et al. (2004)
• Systems with abandonment
– Baccelli and Hebuterne (1981)
– Brandt and Brandt (1999a,b)
– Garnett et al. (2002)
– Mandelbaum and Zeltyn (2004)
– Whitt (2004a,b,c,d)
• Also blocking, retrials, dependent λ...
– Jagerman (1974)
– Massey and Wallace (2002)
– Whitt (2002,2003)
– Aguir et al. (2004)
22
Research, basic model...
Staffing for a more complex arrivalprocess:
• Time-varying rates – numerical methods
– Yoo (1996)
– Ingolfsson et al. (2002)
• Time-varying rates – approximate methods
– PSA – Green and Kolesar (1991)
– fluids – Mandelbaum et al. (1995...2000)
– Mt/G/∞ systems
Eick et al. (1993a,b)
Jennings et al. (1996)
Massey and Whitt (1997)
Whitt (1999)
• Uncertain arrival rates (and absenteeism)
– Ross (2001)
– Chen and Henderson (2001)
– Jongbloed and Koole (2001)
– Whitt (2004e)
– Harrison and Zeevi (2005)
23
Research, basic model...
Higher-level problems:
• Solutions to scheduling and rostering problems
– break placement: Segal (1974)
– column restriction
Henderson and Berry (1976)
– alternative formulations
Aykin (1996), Brusco and Jacobs (2000)
– rostering: Thompson (1997)
• Joint staffing and scheduling
– Yoo (1996), Ingolfsson at al. (2000,2002)
– Atlason et al. (2002)
– Koole and van der Sluis (2003)
• Long-term hiring and training
– Aksin (2002)
– Gans and Zhou (2002)
24
Research, routing and networking...
Skills-based routing:
• The general routing problem
1 2 3 4 5
1 2 3 4 5 6types of calls
pools of CSRs
feasible routings
• Special routing structures
“I” “N” “X” “W” “M”“V”
1 2 2 2 2 11 1 1 1 3 2
1 2 2 2 11 1 1 1 32
25
Research, routing and networking...
Non-scaled analysis:
• Exact analysis of special designs
– Bhulai and Koole (2000)
– Gans and Zhou (2003,2004)
• Heuristic policies for special designs
– Brandt and Brandt (1997)
– Stanford and Grassmann (2000)
– Shumsky (2004)
– Stolletz and Helber (2004)
– Chevalier et al. (2004)
• Heuristic policies for more general designs
– Koole and Talim (2000)
– Borst and Seri (2000)
– Wallace and Whitt (2004)
26
Research, routing and networking...
Asymptotic analysis:
• “ED” regime
– batching policies –Harrison and Lopez (1999)
– thresholds, N-design – Bell and Williams (2001)
– Gcµ rules – Mandelbaum and Stolyar (2003)
• “QED” regime
– Armony and Maglaras (2004a,b)
– Armony and Mandelbaum (2004)
– Harrison and Zeevi (2004)
– Atar et al. (2004)
• Other scalings
– Harrison and Zeevi (2005)
– Bassamboo, Harrison and Zeevi (2004)
27
Research, routing and networking...
Networking:
• Network structures
network ACD
network
call centers
load balancing overflow load balancingand overflow
• Some recent telecom research
– Borst et al. (1996)
– Kogan et al. (1997)
– Servi and Humair (1999)
• SBR methods may also apply
28
On the horizon...
Examples of further directions forresearch:
• A broader view of the service process
– Srinivasan and Talim (2004)
– Aksin and Harker (2001,2003)
– de Vericourt and Zhou (2003)
• Better understanding of customer and CSR
behavior
– Carmon and Kahneman (2002)
– Shumsky and Pinker (2003)
• Multiple levels of equilibria
– Whitt (2003)
– Armony and Maglaras (2004a,b)
• Customer relationship management by the
numbers
– Ariely et al. (2002)
– Gunes and Aksin (2003)
29