Modeling

1
Modeling Simulation of Knowledge Worker
Attention for Evaluation of Email Processing
Strategies

• Robert A. Greve

2
Agenda

• Introduction
• Literature Review
• Research Questions, Propositions, and Hypotheses
• Case Study Results
• Model Development
• The Simulation Model
• Results
• Summary and Conclusions

3
Introduction

• we have little published research about e-mail
  (Ron Weber, Editorial Comments, MISQ,
  September 2004)
• 68 check email more or less continuously and
  17 check email a few times per hour. (Osterman
  Research, DAntoni, 2004)
• Email Reactions - 70 within 6 seconds, 85
  within 2 minutes time to recover from email
  interrupts 64 seconds (Jackson, 2003)

4
Literature

• Interruptions
• Interrupted tasks require more time than
  uninterrupted tasks (Czerwinski, et al., 2000a)
• Effects are moderated by
• tasks complexity (Speier, 1999)
• the interrupts similarity (Speier, 1999
  Czerwinski, et al., 2000b),
• the phase of the task (Cutrell, et al., 2000
  Czerwinski, et al., 2000b Monk, et al. 2002)

5
Literature

• Interruptions
• Interruptions are preceded by an Interruption Lag
  and followed by a Resumption Lag (Trafton, 2003)
• Project Work ? Email Alert ? Interruption Lag
  ? Email Processing ? Resumption Lag ?
  Project Work
• On average it takes a worker 64 seconds to
  recover from an email interruption. The average
  time taken to react to an arriving email was only
  1 minute 44 seconds. Over 70 reacted to the
  email within six seconds of the email arrival
  (Jackson, 2003).

6
Literature

• Asynchronous Communication
• Asynchronous meetings require more time
  (Hightower and Sayeed, 1996)
• Groups working in the asynchronous mode of
  communication will report that the groups
  problem solving process is less efficient than
  will groups working in the face-to-face mode of
  communication (Dufner, et al., 2002)

7
Literature

• Email Overload Solutions
• Filtering
• (SPAM filters, Sharda, et al., 1999)
• Organizing
• (Marson, 2000 Rennie, 2000 Balter, 2000)
• Prioritizing
• (Balter, 2000 Losee, 1989 Horvitz, Jacobs, and
  Hovel, 1999)
• Timing
• (Jackson, 2003)

8
Experiments (Testing of Hypotheses)
Email Environment
Performance -Efficiency -Hours Worked -Email
Resolution Time
Email Processing Strategy
9
Research Objectives

• Answer questions concerning the impact of a
  knowledge workers email processing strategy on
  the knowledge workers performance
• Prescribe email processing strategy solutions for
  given environments and objectives

10
REMS (Research in Email Management Strategies)

• Gupta, A., Sharda, R., Greve, R., Kamath, M.,
  Chinnaswamy (2005) How often should we process
  email? Balancing interruptions and quick response
  times, to be presented at The 2005 Big 12 IS
  Research Symposium
• Key Differences
• Based on information and data collected from
  knowledge workers involved in long term projects,
  instead of tasks
• Capturing total hours worked and efficiency,
  instead of utilization
• Modeling of the knowledge workers attention as
  an entity, separate from the email entities
• Separation of urgent from non-urgent email
• Use of optimization tool

11
(No Transcript)
12
Research Questions

• Research Question 1 Where is the middle ground?
  How many controlled interruptions is enough to
  allow for appropriate resolution times?

13
Research Questions

• Proposition 1(a) Dividing non-priority email
  work into two specific time frames (holding email
  hours twice daily) will allow for successfully
  replying to all email within the 24 hour
  appropriate time frame.
• Proposition 1(b) Processing email in batches
  corresponding to 1/2 of an average daily email
  processing load will allow for successfully
  resolving all email within the 24 hour
  appropriate time frame.

14
Research Questions

• Research question 2 To what extent will fewer
  interruptions result in more efficient work
  completion?
• Hypothesis 2(a) Dividing non-priority email work
  into two specific time frames will result in
  significantly greater efficiency when compared to
  processing email continuously. (Speier, 1999
  2003 Trafton, 2003 and Jackson (2003)
• Hypothesis 2(b) Processing email in batches
  corresponding to 1/2 of an average daily email
  processing load will result in significantly
  greater efficiency when compared to processing
  email continuously. (Speier, 1999 2003
  Trafton, 2003 and Jackson (2003)

15
Research Questions

• Research question 3 To what extent will fewer
  interruptions lower information overload, as
  indicated by the numbers of hours worked daily?
• Hypothesis 3(a) Holding email hours twice daily
  will result in significantly fewer total hours
  worked daily when compared to processing email
  continuously. (Speier, 1999 2003 Trafton,
  2003 and Jackson (2003)
• Hypothesis 3(b) Processing email in batches
  corresponding to 1/2 of an average daily email
  processing load will result in significantly
  fewer total hours worked daily when compared to
  processing email continuously. (Speier, 1999
  2003 Trafton, 2003 and Jackson, 2003)

16
Research Questions

• Research Question 4 To what extent will email
  arrival patterns influence the success of given
  email processing strategies?
• Proposition 4(a) Email hours scheduled during
  peaks in arrival patterns will have significantly
  shorter resolution times when compared to email
  hours not scheduled during peaks in arrival
  patters.
• Proposition 4(b) Email processed in batches will
  have significantly shorter resolution times when
  compared to email hours not scheduled during
  peaks in arrival patters.
• Proposition 4(c) Email processed in batches will
  not have significantly different resolution times
  when compared to email hours scheduled during
  peaks in arrival patterns.

17
Research Questions

• Research Question 5 Can an optimization tool be
  used in conjunction with simulation to automate
  the analysis of email processing strategies in
  finding an optimal email processing strategy for
  specific performance objects and constraints?
• Proposition 5 Optquest, coupled with the Arena
  simulation tool will produce results consistent
  with those obtained through analysis of the Arena
  simulations output.

18
Research Approach

• Case Study Assessment
• Modeling of the Knowledge Work Environment
• Simulation (ARENA)
• Experiments (Testing of Propositions
  Hypotheses)
• Optimization through Simulation (Optquest)

19
Case Study Results

• Project managers interviewed
• Email is an essential tool
• Email is monitored continuously
• Email is intrusive
• Email overload is a real problem
• Processing all for the few
• Never caught up
• Go home when a milestone is reached

20
A Typical Day of Email
21
Mathematical Model
22
Mathematical Model
23
Mathematical Model

• Wqjs emails wait in the queue (time spent
  waiting for the knowledge workers attention)
  for email of urgency j, having sequence
  number s
• Wsjs emails wait in the system (email
  resolution time) for email of urgency j
  having sequence number s
• Wsjs Wqjs Pkds
• __
• Wsj mean email resolution time for email of of
  urgency j
• __
• Wsj ?s Wsjs / S

24
Mathematical Model

• Yd total email processing occurring on day d
• Yd ?k ?s Pkds
• Zd total amount of primary work completed on day
  d
• Zd gt Qd - Yd
• Gd total lag time occurring on day d
• Gd ?s Lds ?s Rds

25
Mathematical Model

• Hd total hours worked by the knowledge worker on
  day d
• Hd Yd Zd Gd
• __
• H mean hours worked by the knowledge worker
• ?d Hd / D
• Ed knowledge worker efficiency occurring on day
  d
• Ed (Yd Zd) / Hd
• __
• E mean knowledge worker efficiency
• ?d Ed / D

26
Arena Simulation
27
Email Flow, Statistics, Disposal Submodel
28
The Continuous Email Processing Strategy Submodel
29
Non-Continuous Email Processing Submodels
30
Simulation Implementation

• Warm Up 30 days
• Run Length 90 days
• Replications 40
• Types of Simulations 13
• Observations (n) 520

31
Results

• MANOVA Model Results
• __ __ __ __ __ __
• H E Ws1 Ws2 Ws3 Ws4 X
• The email processing strategy (X) had a
  significant main effect. A statistically
  significant difference was found between groups
  (a 0.001).

32
(No Transcript)
33
Results of All Strategies
34
Results of All Strategies
35
Results of All Strategies
36
Optquest

• Find policy X from set of policies X, such that
  objective function Z is optimized, subject to
  constraint C.
• Example _
• Max E
• __
• S.T. Ws2 lt 3

37
Optquest Implementation
38
Results

• Proposition 5 Optquest, coupled with the Arena
  simulation tool will produce results consistent
  with those obtained through analysis of the Arena
  simulations output.
• SUPPORTED (Results were consistent with those
  obtained through brute force search)

39
Results of OptQuest
40
Contributions to Knowledge

• One more piece of the puzzle (incremental
  contribution from Gupta, et al. (2005))
• Overlooked solution
• Timing of email processing
• Separation of urgent from non-urgent email
• Consideration of high-level knowledge workers
• New understanding of knowledge workers email
  challenges
• Processing strategies had little to do with
  efficiency.
• Unique analytical approaches
• Simulation of Knowledge Worker Attention
• Optquest Optimization through Simulation

41
Contributions to Knowledge

• Demonstration of 3 gain in efficiency without
  sacrificing email processing success
• Demonstration of need for policies specifying
  urgency of email in need of processing

42
Future Research

• Expanded modeling of email environments
• Expanded modeling of email processing strategies
• The use of Optquest in creation of an ESDSS
• Testing in real environment

43
Questions?