=Paper=
{{Paper
|id=Vol-1561/paper1
|storemode=property
|title=Early Experience with System Dynamics Modeling for Organizational Decision Making
|pdfUrl=https://ceur-ws.org/Vol-1561/paper1.pdf
|volume=Vol-1561
|dblpUrl=https://dblp.org/rec/conf/indiaSE/RajbhojS16
}}
==Early Experience with System Dynamics Modeling for Organizational Decision Making==
https://ceur-ws.org/Vol-1561/paper1.pdf
Early Experience with System Dynamics Modeling for
Organizational Decision Making
Asha Rajbhoj Krati Saxena
Tata Consultancy Services Research Tata Consultancy Services Research
Pune, India Pune, India
asha.rajbhoj@tcs.com krati.saxena@tcs.com
ABSTRACT operating condition is an effort-, time- and cost-intensive, and
Increased business dynamics mandates modern organizations to error prone activity. Also, cost of erroneous response is
proactively prepare their response to operating environment prohibitively high. Hence, it is necessary to evaluate strategies
changes. Typically, large size organizations consist of multiple before their adoption. Simulation based approaches supporting
interconnected departments. Individual department strategies need temporal and quantitative analysis are best suited for such needs.
to be balanced so as overall organization performance is System dynamics (SD) modeling [3, 4] supports modeling
improved. With increased number of interdependent strategic dynamic behavior of organizations and early validation of
parameters, complexity of their combinatorial evaluation strategies through temporal and quantitative analysis. Main
increases. System dynamics (SD) modeling supports modeling concepts in SD modeling approach are causality, feedback loop,
dynamic behavior of organizations and strategy validation through stock (accumulation of entities), flow (dispersal of entities) and
simulation. We used SD modeling techniques to model large size delays. Many SD tools [5, 6] are available that provide
organization decision making problem that involves dynamic as sophisticated simulation support to play out various what-if
well as combinatorial complexity. In this paper, we share our scenarios. SD modeling approach has also been proposed in
experience and learning from this endeavor. decision making of many domains [7, 8, 9]. These uses mainly
highlight dynamic complexity in decision making. We used SD
CCS Concepts modeling approach for large scale organizational decision making
• Organizational decision making➝System dynamic involving both dynamic and combinatorial complexity. We
modeling➝ combinatorial complexity • Enterprise modeling modeled fairly large IT services provisioning organization from
➝Simulation. practitioner perspective and simulated models for 5 years to
evaluate various strategies related to revenue and profit growth
Keywords under different operating environment conditions and changing
Organizational decision making; Enterprise modeling; System organization state. We share our early experience and learning
dynamics modeling; Simulation from this exercise in this paper. Though overall experience is
shared in a specific context, we believe, enterprise architects,
researchers and practitioner will find the takeaways from this
1. INTRODUCTION experience applicable even in a more general context.
Today's business environment is characterized by its dynamic The rest of the paper is organized as follows. Section 2 of the
nature. To survive and remain competitive, modern organizations paper presents a motivating example. Section 3 describes models
need to sense the environment changes and respond to them and analysis results of the example. Finally, we discuss our
proactively [1]. Typically large organizations have multiple inter- experience, learning and future work in section 4.
dependent departments each focusing on specific organization
function. Each department has different, multiple strategies to 2. MOTIVATING EXAMPLE
optimize performance. Strategy adopted by individual department In this section we introduce a motivating example that sets the
keeping local optimization focus does not lead to overall context for the rest of the paper. Let us consider a large IT
organization optimal performance. Individual department services provisioning organization whose main business is
strategies need to be balanced so as overall best organization developing software projects as per customer needs. Overall high
performance is achieved [2]. This is possible only through con- level activities involved in this business are ‘bid project’ -> ‘win
joint evaluation of all related strategic parameters. As number of project’ -> ‘execute project and maintain good track record’ ->
such parameters increases their combinatorial evaluation 'receive payment from customers’. For executing these activities
complexity also increases. Thus, considering all inter- there are multiple departments involved as shown in Fig 1. Sales
dependencies and finding best possible response to changed department bids for projects by responding to requests for
proposals (RFPs). Resource management department recruits
"Copyright © 2016 for the individual papers by the papers' authors. people and allocates people for project execution. Delivery
Copying permitted for private and academic purposes. This volume is department executes projects. Account department keeps track of
published and copyrighted by its editors." finances. Each of these departments may choose different
strategies for achieving their goals as shown in Fig.1. For instance
to win more projects sales department may offer to reduce bid
price, promise early delivery etc. Delivery department may
consider improving employee productivity to maximize timely
2nd Modelling Symposium (ModSym 2016) - colocated with ISEC 2016, Goa, India, Feb 18, 2016
4
delivery. Resource management function may offer better pay
package to improve joining probability and so on.
These departments are interdependent on each other as shown in
Fig.1 and Fig. 2. Good delivery track record helps sales
department win bids. Delivery department cannot function unless
sales win projects and resource management department makes
appropriate number of people available. Unavailability of people
leads to delayed start of project that in turn results in delayed
delivery and penalty. Also, people cannot be recruited in excess as
more people on bench leads to poor employee utilization thus
impacting profitability of the organization. Functioning of these
departments is also influenced by external events on which
organization does not have any control e.g. RFP arrival rate,
supply of people etc – moreover they may vary over time. Factors
such as RFP win rate, employee attrition etc are also influenced
by size and quality of competition. Organization state also keeps
on changing over time. For instance, employee’s experience in
executing projects, senior-junior ratio, salary costs, organization
delivery track record etc. is dynamic. Thus, overall dynamics in
operating environment as well as changing organization state
complicates decision making process. Given this organization
dynamic context, organization has to analyze its performance and Fig 2 : Goals, measures, levers influence relationship
various actions in response to changes in different operating
environment conditions. As an example in this paper we
considered changing demand situation and evaluated various 3. MODEL BASED DECISION MAKING
strategies to analyze organization performance in terms of revenue We used system dynamic modeling iThink [5] tool to model
and profit growth over 5 years. organization behavior and analyze questions of interest. For
IT services provisioning organization also has to face creating models we have made certain assumptions about
continuously rising demand for reduced price and/or reduced organization settings. Typically, service provisioning organization
time-to-market delivery. To stay relevant in business, organization executes projects of different kinds, sizes, and complexities. We
has to continuously improve its operating efficiency. Hence, considered 2 kinds of projects namely J2EE and Mobile. Each of
organization is looking forward to develop end-to-end code these project kinds are further classified on size (Small/Large) and
generation tools and use them for project development. Before complexity (Simple/Complex) dimensions. Thus there are 4 types
implementing this strategic decision, it would like to first analyze: i.e small simple (SS), large simple (LS), small complex (SC),
what will be the overall investment to develop end-to-end code large complex (LC) of J2ee projects and similar 4 types of mobile
generator tools? , will this approach be profitable? how long will projects considered. We used COCOMO [10] equations for
it take to reach break-even point for profitability?. estimating the effort, time, team size. For each type of the
projects, arrival rate, winning rate, pricing etc. are given as input.
Typically, project execution resources have different kind of skills
and different number of years of experience. We considered two
kinds of workforce resources i.e. Junior (J) and Expert (E) having
different productivity and salary. Initial state of the organization is
specified by setting the initial values of all the lever variables,
other internal variables and stocks as shown in Table 1.
Table 1 : Variable values
Fig 1: Organization Structure
2nd Modelling Symposium (ModSym 2016) - colocated with ISEC 2016, Goa, India, Feb 18, 2016
5
Fig. 3: System Dynamic Model
2nd Modelling Symposium (ModSym 2016) - colocated with ISEC 2016, Goa, India, Feb 18, 2016
6
3.1 Analyzing organization growth under strategy to increase pay for junior to 1.2K. With this change,
delayed projects are reduced to zero , but increase in revenue was
dynamic operating conditions marginal and profit was reduced to great extent. Thus, this
In the interest of space, we discuss only a subset of the model. analysis hinted that strategy to reduce time and to improve
Fig. 3 shows systems dynamic model describing behavior of sales, employee productivity through training augurs well for growth.
delivery, resource management and accounts department. ‘Sales’
sector covers project bidding process where ‘PM arrival rate’ and Using reduced time and improved productivity option, we
‘Sales team size’ variables control the project ‘Target market’ analyzed next scenario of changing demand situation. We
inflow. RFP response time is modeled using ‘RFP delay’ stock considered slow decline of traditional J2EE project demand and
and using delay function on outflow ‘RFP responded’. Similarly steep rise in of mobile project demand as shown in Fig. 4. X axis
bid processing delay is modeled using a separate stock and delay shows time duration of 5 years. Y axis shows arrival rate change
function. The stock ‘Missed’ represents missed RFP opportunities. with maximum value being 11 projects per month. The project
This measure helps in deciding the sales team size. Bid win rate is arrival rate change is considered non-linear and different for
modeled as ‘WinRate’ variable which is computed using track different types of project as shown in the Fig.4. With demand
record, rate deviation, price deviation variables. Delivery sector change, we observered that number of delayed mobile projects
covers project execution process wherein ‘pipeline’ stock have increased and mobile J, E bench gone to low peak. As J2EE
represents bids won. ‘Project to start’ variable determines the project demand reduced, it’s J, E bench was increased. Hence,
flow of project that can start execution. Project people allocation J2EE people can be reskilled to Mobile. To arrive at appropriate
is decided using this variable. For each J and E bench separate reskill rate, we played out with different value as shown in Table
stocks are used and project allocation is modeled using outflow 2.
and de-allocation is modeled using inflow. Allocation, de-
allocation part covers people allocation to projects considering
project size, projects in pipeline, bench strength and project
execution priorities. People are de-allocated and moved to bench
on project completion. On promotions J people are moved from J
stock to E stock. ‘Delivery’ sector also shows delayed projects.
We only considered the delay due to unavailability of people. To
model different types of projects, we used array abstraction [11]
in sales and delivery sector. ‘Resource management’ sector shows
recruitment of different type of people and delays. Joining delay
of people for all type resources is modeled using conveyor stock.
Accounts sector shows profit, revenue, expenses computations.
Salary, training, project related expense are accumulated in
‘Expenses’ stock. Earnings from projects are accumulated in
‘Revenue’ stock. Profit is computed using values of these stocks.
Competition influence to the organization impacts win rate,
employee attrition rate, employee joining probability parameters.
It is modeled by setting appropriate initial value of these
parameters
3.1.1 Results
Once all the necessary behavioral aspects are captured we played
out different what-if scenarios. Prior to analysis of dynamic
demand, we analyzed whether organization is operating in
comfort zone and is there any scope for improving employee
utilization. With given initial settings, 5-year simulation showed
significant value for ‘J bench’ and ‘E bench’ stocks. Thus, it can
be inferred that organization can target more number of projects
say by increasing bid win rate. For achieving this possible
strategies are: To reduce time by 10%, to reduce price by 10% or
reduce time and price both. We observered reduce time gave
better revenue and profit growth as compared with reduce price.
Use of both strategies together increased the winning rate even
more. Undesirable effect was significant increase in number of
projects witnessing delayed start due to unavailability of resources
– value of J bench and E bench stocks. The delayed start resulted
in delayed delivery and subsequent penalties. To reduce delays
we used strategy to improve employee productivity through
training. With this lever change, project delay got reduced and
revenue and profit increased. We observed bench measures again
to check further possibility of improvement. Bench of J people
indicated that its going quite low as compare to E people. To keep Fig 4: Demand rate change
balance of senior:junior ratio, possible solution was to increase J
people salary to increase the joining probability. Hence, we used
2nd Modelling Symposium (ModSym 2016) - colocated with ISEC 2016, Goa, India, Feb 18, 2016
7
Table 2 : Re-skill rate change scenario playing
For the given organization state, E reskill rate of 2 per month and
J reskill of 3 per month gave better result. Delayed mobile project
were also reduced. For further possibility of improvement when
we observed mobile ‘Missed’ opportunities measure, it indicated
that there is still more market that can be targeted to utilize J2EE Fig. 5 : Break even analysis model
J,E bench strength. Hence, we increased mobile sales people by
increasing their joining probability through increased payment.
We played out various reskill rates again as earlier. E reskill rate
of 13 per month and J reskill of 6 per month and allocating 65 %
trainee to Mobile projects gave better revenue and profit results.
Thus, through simulation we could arrive at reskill rate for given
changing demand rate scenario.
3.2 Analyzing code generator platform
development investment
For analyzing whether investing in developing end to end code
generator [12] tool be profitable and by when break-even point of
profitability will be reached we extended the system dynamic
model shown in Fig.3. We added code generation based project
Fig. 6: Employee allocations for tool dev
execution in delivery model. Generation based project
development time is computed using COCOMO estimation. We
further extended model to capture tool development effort and
3.2.1 Results
delays as shown in Fig. 5. ‘Tool use %’ indicated percentage of
For developing end to end code generation tools we considered
total won projects that can be targeted using generative approach.
initial 30 J employees and 15 E employees with reducing strength
Use of generative approach also introduced dynamics in
over 5 year as shown in Fig. 6. X axis shows time duration of 5
organization operation. Use of generative approach also
years. Y axis shows employee allocations for the tool
influenced targeted projects. Due to productivity gain, number of
development. We arrived at this tool team allocations considering
people required for project execution were reduced; as a result
following assumptions: 1) Initial tool development time as 6
more people were available for project execution; hence more
months. 2) Post development small % of team participates in
projects could be targeted which in turn increased tool usage. This
initial tool deployments and large % of team continues with tool
created positive re-enforcing cycle for organization operation.
enhancements for betterment. 3) Over 3 years there will be
We assumed that there is no direct revenue earning through continuous increase in tool use and tool enhancement effort will
generator tools sell and gains are primarily due to indirect profit be slowly reduced. 4) Towards end major effort goes in tools
earning due to productivity gain and competitive advantage for deployments, support and consultancy. With these assumptions
project win. To analyze expenditures, salary of the all employee overall team size is considered reducing over 5 years.
allocated for the tool development is accumulated in ‘Tool dev
With these settings, when we ran the model for 5 years we
expenses’ stock. To analyze gains, employee effort saving for
observered that break-even point is reached after 45 months as
execution of projects using code generation approach is
shown in Fig.7 and thereafter profit gain was exponential. Thus,
transformed to indirect money saved and accumulated in ‘Tools
developing end-to-end code generation tools and using them for
indirect revenue’ stock. ‘Project completed using tools’ is
project development validated as a viable option for the given
dynamic entity in delivery model. It is used for computing indirect
settings.
savings per project.
2nd Modelling Symposium (ModSym 2016) - colocated with ISEC 2016, Goa, India, Feb 18, 2016
8
example covered comparatively few set of levers. Typically, large
organizations play with many more number of strategic levers to
analyze operating environment dynamics. Hence, we think both
dynamic and combinatorial complexities are important concerns
for organizational decision making. SD modeling only addressed
dynamic complexity in decision making of IT services
provisioning organization. Our initial study indicates, hybrid
approaches that can combine other technique with SD modeling
technique [14, 15] may help in addressing combinatorial
complexity also. We planned to explore further in this direction.
5. REFERENCES
[1] Soroosh Nalchigar, Eric S. K. Yu, Steve M. Easterbrook.
Towards Actionable Business Intelligence: Can System
Dynamics Help? PoEM 2014: 246-260
[2] Suman Roychoudhury, Asha Rajbhoj, Vinay Kulkarni,
Fig. 7 : Break-even analysis result Deepali Kholkar. Models to Aid Decision Making in
Enterprises. ICEIS (3)) 2014: 465-471
[3] Jay Wright Forrester. Industrial dynamics, volume 2. MIT
4. EXPERIENCE, LEARNING AND press Cambridge, MA, 1961.
FUTURE WORK [4] John Sterman. Business dynamics. Irwin-McGraw-Hill,
Using system dynamic modeling we could model behavior of 2000.
various departments of large IT services provisioning
organization. Quantitative nature of SD models helped in [5] http://www.iseesystems.com/Softwares/Business/ithinkSoftw
specifying operating environment conditions in terms of RFP are.aspx
arrival rate, supply of people, and competition influence and so [6] Vensim, November 2013. http://www.vensim.com.
on. We could also model dynamic organization state in terms of [7] Francisco Campuzano and Josefa Mula. Supply Chain
people bench, different skilled people, expertise level (J / E), and Simulation: A System Dynamics Approach for Improving
organization track record and so on. SD models are typically Performance. London: Springer, 2011, vii, 106 pages. ISBN
meant for aggregated and generalized view of systems. With 978-0-85729-718-1
people and project type use, we tried to make model more specific
and bit closer to real life situation. These types can be expanded [8] Hirsch G, Homer J. Modeling the dynamics of health care
further for improving accuracy of the results; however as it services for improved chronic illness management. In: 22nd
hampered simulation speed and also increased further International System Dynamics Conference. Oxford,
combinatorial complexity we restricted it to few types. England: System Dynamics Society; 2004.
Prior to SD modeling we identified various levers, measures, [9] Pruyt, E. (2006a). System dynamics and decision-making in
goals and their interrelationship. This activity was solely manual the context of dynamically complex multi-dimensional
and required necessary expertise about the domain of the societal issues. In Proceedings of the 2006 Conference of the
organization to be modelled. This representation helped in the SystemDynamics Society, Nijmegen. System Dynamics
problem space definition clarity and in creating SD model. It also Society. 7, 12
guided about which measures to be observed and next strategic [10] Barry W Boehm. Software Engineering Economics. Prentice
value / option to be chosen during simulation. Hall, 1981 ISBN:0138221227
Evaluation of organization goal attainment under different [11] Myrtveit, Mangne. Powerful Modeling Using Array
operating conditions was possible by playing out different Variables. 12th International Conference of the System
strategic parameters con-jointly. However, we observered that Dynamics Society
playing out multiple levers to arrive at suitable strategic option
[12] Vinay Kulkarni, R. Venkatesh, Sreedhar Reddy: Generating
was very time consuming activity. Our observation is in line with
Enterprise Applications from Models. OOIS Workshops
Homer [13] that says “The more one extends the scope of a model
2002: 270-279.
in an attempt to make it more useful or complete; expanding it to
include more concepts and variables of interest, the more effort [13] Homer J. (2014), Levels of evidence in system dynamics
will be required to achieve a desired level of evidence”. Each modeling, Syst. Dyn. Rev., 30: 75–80. DOI:
simulation run involved changing lever values -> observing 10.1002/sdr.1514
measures -> getting hints about what should be value of same [14] Duggan, Jim. 2007. Using system dynamics and multi
lever and / or next lever change . This loop had to be repeated objective optimization to support policy analysis for complex
multiple times till best possible lever values are arrived at. systems. In Understanding Complex Systems, 59–81.
Numbers of change levers were more, and most of the levers were Springer Berlin/Heidelberg.
of integer value types which increased the simulation search space
to large extent and resulted in increasing combinatorial [15] Alborzi, M., 2008. Augmenting system dynamics with
complexity. Moreover, changing lever values and observing genetic algorithm and TOPSIS multivariate ranking module
simulation result was completely manual process hence it look for multi-criteria optimization, Proc. of Int. Conf. of the
considerable amount of time and effort. We would say our System Dynamics Society.
2nd Modelling Symposium (ModSym 2016) - colocated with ISEC 2016, Goa, India, Feb 18, 2016
9