An Analysis of Email Response Policies under Different Arrival Patterns By Ashish Gupta Doctoral Student, Department of Management Science

1
An Analysis of Email Response Policies under
Different Arrival PatternsBy Ashish Gupta
Doctoral Student, Department of Management
Science Information Systems, Oklahoma State
University, Stillwater. Ramesh Sharda Regents
Professor of Management Science Information
Systems, Director, Institute for Research in
Information Systems, Oklahoma State University,
Stillwater.
2
Objective of the study

• To improve individual knowledge worker
  performance by identifying policies that will -
• To model email work environment by considering
  various email characteristics.
• Improve response time of emails and primary task
  completion time
• Reduce number of interruptions
• Validate the results of prior research.

3
Problem significance

• 2004 AMA Research on workplace E-Mail
  Productivity
• On a typical workday, time is spent on e-mail is
  ?????
• 059 minutes 77.9
• 90 minutes2 hours 18
• 23 hours 2
• 34 hours 2.5
• Osterman Research- How often do you
• check your E-mail for new messages
• when at work?

4
Problem significance

• E-Policy Institute (2004)
• Annual Email growth rate 66
• Corporate Research
• IBM, Microsoft, Xerox, Ferris, Radicati, etc.
• Need for more research in MS/IS that
• Looks at the problem of information overload and
  interruptions simultaneously.

5
Extant Research

• Overload due to emails-
• First reported by Peter Denning (1982).
• Most recently reported by Ron Weber (MISQ,
  Editor-in-Chief 2004)
• Interruptions due to emails-
• Reported by some- Speier,et.al.1999, Jackson,
  et.al., 2003, 2002, 2001), Venolia et.al. (2003)
  

6
Extant Research

• The nature of managerial work, Mintzberg (1976)
• Managerial communication pattern, Ray Panko
  (1992)
• Email as a medium of managerial choice, M.
  Markus (1994)
• You have got (Lots and Lots) of mail in The
  Attention Economy by Davenport (2001)
• The Time Famine Towards a Sociology of Work
  Time, Leslie Perlow (1999)

7
Phenomenon of Interruption
Interruptions- According to distraction theory,
interruption is an externally generated,
randomly occurring, discrete event that breaks
continuity of cognitive focus on a primary task
(Corragio, 1990 Tétard F. 2000).
8
Previous Research Model
Only high dependency on email communication (3
hrs) with exponential email arrivals was studied
9
Detailed Research model
Processing time is based on email category
10
Email types

• Emails differentiated on the basis of its
  content or the action required by the user

Notation Email type Discrete arrival percentage
1 Priority email 5
2 Spam 5
3 Informative email 20
4 Email with non-diminishing service time 55
5 Email with diminishing service time 15
11
Email Policies
Dependency on Email Communication Dependency on Email Communication Dependency on Email Communication Dependency on Email Communication
Policy type Very Low (1 hr) Low (2 hrs) High (3 hrs) Very High (4 hrs) Notation of Email hour- slots
Triage 8am-9am 8am-10am 8am-11am 8am -12 noon C1 1
Schedule 8am-830am 430pm- 5pm 8am-9am 4pm-5pm 8am-930am 330 am to 500 pm 8am-10am 3pm- 5pm C2 2
Schedule 8am-815am, 11am-1115am 1pm-115pm 445pm- 5pm 8am-830am, 11am-1130am 1pm-130pm 430pm- 5pm 8am-845 am, 11am-1145am, 1 pm - 145 pm, 415 pm - 500 pm 8am-9am 11am - 12 1pm- 2pm 4pm- 5pm C4 4
Schedule 8am-808am 9- 908am and so on 8-815am 9-915am 10-1015am and so on 8-823am 9-923am 10-1023am and so on 8- 830am 9- 930pm 10- 1030pm and so on C8 8
Flow Processed as soon as emails arrive Processed as soon as emails arrive Processed as soon as emails arrive Processed as soon as emails arrive C Not Applicable
12
Methodology

• Discrete event simulation using Arena 8.01
• Model Run length 500 days
• Model Warm-up time 50 days
• No. of replications of each model 20
• 16 scenarios evaluated for 5 different policies.
• Thus, Total number of simulations models 16 x 5
  80
• Total number of data points generated
• 80 x 20 1600

13
Scenarios
Scenarios Email (E) dependency E Arrival pattern E processing time
1 Very low Time stationary Expo Small
2 Very low Time stationary Expo Large
3 Very low Non-Stationary Expo Small
4 Very low Non-Stationary Expo Large
5 Low Time stationary Expo Small
6 Low Time stationary Expo Large
7 Low Non-Stationary Expo Small
8 Low Non-Stationary Expo Large
9 High Time stationary Expo Small
10 High Time stationary Expo Large
11 High Non-Stationary Expo Small
12 High Non-Stationary Expo Large
13 Very High Time stationary Expo Small
14 Very High Time stationary Expo Large
15 Very High Non-Stationary Expo Small
16 Very High Non-Stationary Expo Large
14
Parameters
S Type 4 email (E) Processing time (PT) Type 5 E PT (min) Total Email PT per day Avg. Email Arrival Rate Primary Task (P) Arrival Rate /day E Util P Util Min (EP) Util
1 5 5 1 12 62 0.125 0.775 0.9
2 15 15 1 5 62 0.125 0.775 0.9
3 5 5 1 12 62 0.125 0.775 0.9
4 15 15 1 5 62 0.125 0.775 0.9
5 5 5 2 24 52 0.25 0.65 0.9
6 15 15 2 10 52 0.25 0.65 0.9
7 5 5 2 24 52 0.25 0.65 0.9
8 15 15 2 10 52 0.25 0.65 0.9
9 5 5 3 36 42 0.375 0.525 0.9
10 15 15 3 15 42 0.375 0.525 0.9
11 5 5 3 36 42 0.375 0.525 0.9
12 15 15 3 15 42 0.375 0.525 0.9
13 5 5 4 48 32 0.5 0.4 0.9
14 15 15 4 20 32 0.5 0.4 0.9
15 5 5 4 48 32 0.5 0.4 0.9
16 15 15 4 20 32 0.5 0.4 0.9
Processing time of (a) Type 1 email-
Expo(10 min) (b) Type 2 email- Expo (0.5
min) (c) Type 3 email- Expo (5 min) (d)
Primary task- Expo(6 min)
15
Birds Eye view of Entire model built using Arena
Zoom in follows.
16
Arena Email flow Snapshot
1
2
3
Emails created based on different schedules that
determines whether it is Expo or Non-Stationary
Expo and at what rate
1
Preempts the KW when an email of type 1 arrives
during email hrs . Stores remaining processing
time in an attribute RT
2
To record output statistics of each email type
separately
Releases emails of type 2,3,4 on the basis of
policy
3
Checks if email has been in system for gt or lt
than 24 hrs
17
Arena Primary Task Snapshot
Checks to see if RTgt0. If yes, RL and
IL are added If no, Primary task is
sent next processing stage
Attribute RT is reset to 0 to erase the memory.
This makes the attribute RT reusable for
recording remaining time interrupted primary task
in future.
18
(No Transcript)
19
Model Logic

• New email arrival Ei occurs at time T0, for all i
  n n 1 . . 5
• If i 1,
• Step1. Email released at T0.
• Step2. If STATE (KW) IDLE E1.WIP0
• KW seized
• Than, Set RT Ta 0
• IL 0, RL 0
• Process E1
• Release KW
• If STATE (KW) BUSY E1.WIP0
• Seize KW
• Than, Set RT Ta
• Record IL Tria (a, b, c), Tb
• Process E1
• Release KW
• Calculate
• ? Tb /( Ta Tb) for all 0 ? 1
• Calculate
• RL RT ? ( K-1) (1- ?) ( L-1
  ) / Beta (K,L)

20
Model Logic

• For K 2, L 1
• Calculate
• T1 IL Tb RL
• Seize KW for time T1
• Process Pi
• Set RT0
• Release KW
• If i 2 3 4 5,
• Step.3 Release Ei, if
• (STATE(dummy) IDLE_RES
• Process email 1234.WIP 0
• email 5 in 1.WIP 0
• email 5 in 2.WIP 0 )
• ( STATE(anti dummy) IDLE_RES
• Primary.WIP 0
• NQ(Hold primary.Queue) 0
• IL Primary .WIP0
• RL primary.WIP 0 ) TRUE
• Else Hold

21
Model logic- comments

• If New arrival Pn
• Step4. Release if,
• STATE(kW) IDLE_RES
• Else Hold
• //
• Tb- Value added time spent on the task Before
  interruption
• Ta- Value added time spent on the task After
  interruption
• ? - Fraction of task completed before
  interruption occurred
• IL Interruption Lag
• RL Resumption Lag
• Pi interrupted primary task
• Dummy resource- implements email hours
• Anti-dummy resource implements non- email hours
• //
• Stop
• Stop
• Stop

22
Results

• (a) Percent Increase in Utilization

23
Results

• (b) Additional Time (min) spent per day due to
  interruptions

24
Response time results

• Avg. Email Response Time
• Avg. Email processing time (Value added)
• Avg. Email wait (Queue) time fig. c
• Avg. Primary Task (PT) Completion Time fig. d.3
• Avg. PT value added processing time
• Avg. PT non-value added processing time
  due recalling switching fig. d.1
• Avg. PT wait (Queue) time fig. d.2

25
Results
(c) Email Wait time i.e. inbox queue and holdup
time
26
Results
(d.1) Avg. Additional time spent (wasted) in
recalling and switching for processing one
primary task
27
Results

• (d.2) Average Primary Task Wait Time

28
Results
(d.3) Average Primary Task Completion Time
29
Optimal Policy ??

• Previous research found C4 as the optimal policy
  (no consideration was given to email arrival
  pattern and characteristics).
• Current Research found under varying email
  arrival characteristics-
• Optimal policy for primary task completion time -
  C1 C2 closely followed by C4.
• Optimal policy for email response time C
• Optimal policy for reducing interruptions- C1
  C4 closely followed by C2

30
Limitations of the model

• Assumptions of the model are its limitations
• Knowledge worker works strictly according from 8
  to 12 and then from 1 to 5pm. Need for relaxing
  the work-hrs.
• Knowledge worker is busy only 90 of the time in
  a given workday.
• KW is working on interruptible primary task. In
  reality, not all primary tasks are interruptible.
  For e.g. group meetings
• Primary task modeled is interruptible only 3
  times.
• Emails are not interrupted.

31
Limitations future research

• Perform the study in field or experimental
  settings.
• Modeling utility/ life of an email.
• Modeling group knowledge network and at
  organizational level.
• Modeling by incorporating more doses of reality.
  Considering other communication media along with
  email.
• http//iris.okstate.edu/rems/
• Suggestions or comments or Questions????