DECISION SUPPORT SYSTEM FOR COST-BENEFIT ANALYSIS

IN SERVICE PROVISION

Emadoddin Livani

, Elham Paikari

and Günther Ruhe

1, 2

Department of Electrical and Computer Engineering, The University of Calgary, Alberta, Canada

Department of Computer Science, The University of Calgary, 2500 University Dr. NW Calgary, AB, Canada

Keywords: Service Engineering, Decision Support Systems, Cost-benefit Analysis, Bayesian Belief Networks.

Abstract: Cost-benefit analysis is an approach to relate effort and cost of an activity to the resulting benefit. In this

paper a novel decision support system for cost-benefit analysis in the context of service provision is

proposed. Four decision support scenarios are investigated: (i) analyzing the impact of the services on cost

and benefit, (ii) sensitivity analysis for the system variables, (iii) goal-seek analysis, and (iv) analyzing the

impact of the services on operational resources. The key engine of the analysis approach is a Bayesian

Belief Network (BBN). The BBN incorporates the key incoming, control and outgoing service parameters

as well as their probabilistic relationships. In the sense of a hierarchical system, the variation of some of the

parameters is guided by the results of optimizing operational resources being some of the BBN parameters.

We’ve evaluated the framework in a case study with the City of Calgary’s Waste and Recycling Services.

The results showed that using such a DSS facilitates the decision making process and improves the overall

cost-benefit ratio.

1 INTRODUCTION

Service Engineering (SE) is a technical discipline

concerned with the systematic development and

design of services using suitable models, methods

and tools. A service can be any kind of material,

energy, and information. Many studies have

investigated on SE, e.g. (Bullinger, 2003), (Sakao,

2007), (Kapitsaki, 2009), (Sundin, 2010).

Providing services needs resources, like time,

human, and budget. In the other hand, each service

has unique value for the service consumer, hence for

the service provider. As the resources are always

limited, selection of services is needed to increase

the value (benefit) of them. If we aggregate all the

resources as cost and present the value of them as

benefit, then the problem would be cost-benefit

trade-off analysis. This analysis needs to be

performed before an appropriate decision can be

made or a proper action can be taken (Liu, 2003).

There is no deterministic relationship between a

question and an answer in decision-making, as the

process normally involves a great deal of personal

experience and sophisticated reasoning. So, it’s

difficult to be modeled mathematically (Liu, 1999)

(Liu & Alderson, 1999). Probabilistic techniques

like Bayesian belief network (BBN) can be utilized

for this purpose. BBN has been used in the literature

as a decision making (and often decision support)

tool for representing and reasoning with uncertain

knowledge (Fenton, 2001) (Fenton, 1999) (Shirazi,

2009) (Heckerman, 1997) (Ibrahim, 2009) (Fineman,

2009).

Decisions are normally formulated by managers

as three levels: strategic, tactical, and operational.

The decision support systems in the literature

usually focus only on one type of decision and do

not consider the link between them (Liu, 2003)

(Zoric, 2011) (Nanazawa, 2009). So, there isn’t any

sophisticated decision support system for cost-

benefit analysis that evaluates both strategic and

tactical level decisions in one coherent solution.

In this paper a novel decision support system for

cost-benefit analysis of the services is proposed. It

addresses the above gap, by answering the following

questions:

• What’s the impact of a certain service on cost

and benefit?

• Which services dominate the others in terms

of cost and benefit?

Which system variables have the highest

198

Livani E., Paikari E. and Ruhe G..

DECISION SUPPORT SYSTEM FOR COST-BENEFIT ANALYSIS IN SERVICE PROVISION.

DOI: 10.5220/0003514101980203

In Proceedings of the 13th International Conference on Enterprise Information Systems (ICEIS-2011), pages 198-203

ISBN: 978-989-8425-54-6

 2011 SCITEPRESS (Science and Technology Publications, Lda.)

impact on cost and benefit?

• What’s the impact of a certain service on

tactical level variables?

In the next section, the architecture of the

proposed decision support system will be explained.

Then, the results of a case study evaluation of the

framework will be analyzed in Section 3. Finally,

Section 4 concludes the paper.

2 ARCHITECTURE OF THE

DECISION SUPPORT SYSTEM

The decision support system proposed in this paper,

as shown in

Figure 1, consists of three layers: user,

strategic, and tactical. These layers will be discussed

in depth in the following subsections.

Figure 1: Architecture of the DSS.

2.1 User Layer: Services and their

Mapping to the System Variables

Services are realized by unique combination of the

system variables and are associated with their own

cost and benefit. This layer, basically, maintains the

definitions of the services, system variables, and the

mapping between them. Mapping of a service to the

variables means determining the variables which are

affected by implementing it. This effect is measured

by changing in probability distributions of the input

variables in the BBN model (see the next

subsection).

2.2 Strategic Layer: Cost-benefit

Trade-off Analysis with BBN

At the strategic layer, Bayesian belief network

(BBN) is used to analyze the effect of the input

variables on the outputs of the model. A BBN is a

directed acyclic graph consisting of nodes and arcs

with a conditional probability distribution associated

with each node (Heckerman, 1997) (Fenton, 1999).

Nodes represent domain variables, and arcs

represent probabilistic dependencies between them.

Basically, in a BBN model there are three types

of variables: root, internal, and leaf. Root variables

are the inputs to the model so they don’t have any

incoming link from the other variables, as opposed

to the leaves which are output of the model and only

accept incoming links. Internal variables lie in the

middle connecting the former two types to each

other.

The BBN model in this research is used for three

well-known analysis types: user scenario, sensitivity,

and backward (goal-seek). A scenario can be created

by changing the probabilities of the input variables

or considering them as evidence i.e. setting them to

one of their possible values (by 100%). Each

scenario leads to different probabilities for the leaf

variables. The comparison of the scenarios means

comparison of the probabilities of the leaf variables.

We assume that in the BBN model there are two

output variables, one for cost and one for the benefit.

But the model is extensible to more outputs.

We define an abstract function F to map each

scenario (S) to a point in a 2-dimensional Cartesian

space. For each dimension, one for cost and one for

benefit, the function F is represented by Formulas 1

and 2. In Formula 1, c is the size of the states for the

‘Cost’. V



(k) is the probability of the ‘Cost’ being

state k. Similarly, in Formula 2, b is the size of the

states for the ‘Benefit’ and V



(k) is the

probability of ‘Benefit’ being k.



(

,

)







()|  = 1 …



→ℝ

(1)



(

,

)

:



()|  = 1 …  → ℝ (2)

This mapping function, which is abstract, plays

an important role in the scenario analysis. A simple

concrete form of it could be the difference of the

probabilities compared to the baseline situation

(baseline is the initial model without any evidences).

See Formulas 3 and 4 as examples. The goal is to

minimize the F(., Cost) and maximize the F(.,

Benefit).

The sensitivity analysis is pretty similar to the

scenario analysis, except the fact that the scenarios

are created, not by prior knowledge instead, by

setting each root variable (or internal) to its states

one by one and keeping other variables unchanged.

As a result,

∑

(



)





scenarios will be created for

Tactical:

Optimization

User:

Services

Variables which

needs to be calculated

at the tactical level

Impact on tactical

level variables

Cost-benefit

Trade-off Analysis

Variables associated

with each service

Strategic:

BBN

DECISION SUPPORT SYSTEM FOR COST-BENEFIT ANALYSIS IN SERVICE PROVISION

199

root variables, where r is the size of the root

variables and P(R

) is the size of the states for

variable R

. Similarly, for the internal variables

∑

(



)





scenarios are created, where i is the size

of the internal variables and P(I

) is the size of the

states for variable I

In any of the above analyses, user scenario and

sensitivity, the probability of the leaf variables (cost

and benefit) will be calculated for each scenario.

Using Formulas 1 and 2, the trade-off graph will be

created for all the scenarios. Figures 2 and 3 are

example results (they will be discussed in Section 3).

We used Pareto optimal solution (POS) to analyze

the trade-off graph.

Definition 1. Assume we have set P of points, each

point representing a scenario’s impact on Cost and

Benefit, measured by Formulas 1 and 2. Set P* ⊆ P

is called Pareto set if no point in P* is dominated by

a point in P (Nanazawa, 2009). We say point A

dominates B if it has lower cost but higher benefit.

For example, in Figure 2 the circled points are

Pareto points.

In the backward analysis, evidence is set for a

leaf variable instead of a root or internal one. The

model will then suggest new probability

distributions for the root and internal variables. This

analysis specifies the requirements of the model in

order to create the desired outputs. However, the

suggested probabilities for the root variables might

not always be feasible. So, an interaction with the

expert (user of the BBN model) is usually needed to

come up with an acceptable scenario.

2.3 Tactical Layer: Optimizing

Operational Resources

Although analyzing the services at the strategic layer

gives an insight on their cost and benefit, it can be

further supported by measuring their resource

consumption in the tactical layer. In this paper, the

resources are the vehicles; so the problem will then

be the vehicle routing (VR) optimization. However,

our approach is a bit different from the traditional

VR problem as we consider the intersections of the

roads as the nodes of the graph and the roads

between them as the edges. This will reduce the size

of the problem dramatically.

We introduced a customized solution to this

problem (named DCPP) by combining Chinese

Postman Problem (CPP) (Edmonds, 1973) and

Dijkstra shortest path algorithm (Cormen, 2009). H.

Thimbleby (Thimbleby, 2003) proposed a heuristics

for CPP in a connected directed graph. We extended

it in order to make it work in disconnected graphs as

well. Table 1 shows the pseudo code of DCPP

algorithm.

First (steps 1-2), the graph G´ is created by

removing the edges with weight 0 (no service point

on them) from G. Then (steps 3-5) the closest sub-

graph in G´ to the starting node is found. The closest

sub-graph is defined as the one which has a node

that is closest to the start node, based on Dijkstra

shortest path algorithm. In step 6 the CPP problem is

solved for this sub-graph. Then the next closest sub-

graph to the last visited node of the previous sub-

graph is found, again using Dijkstra. This process is

repeated until all the sub-graphs are visited. At the

end, the shortest path is taken to the starting node to

complete the circuit.

The optimized values will then be used for two

purposes:

1. As an additional support for selecting the

decision alternatives (services) by presenting the

actual effect on resource consumption;

Table 1: Pseudo code of the DCPP algorithm.

0 Algorithm DCPP (G: input graph, S: start node);

1 G´ Å Remove edges with weight 0 from G;

2 SG´ Å Set of sub-graphs in G´;

3 V Å S;

4 Route Å {};

5 SG Å Find the closest sub-graph in SG´ to node V based on

Dijkstra shortest path algorithm;

6 Route Å Solve the standard CPP problem for SG and append

to previous Route;

7 SG´ Å Remove SG from SG´;

8 V Å last node visited in SG;

9 Repeat steps 5-8 until SG´ goes empty

10 Find the shortest path from the last visited node to S

ICEIS 2011 - 13th International Conference on Enterprise Information Systems

200

2. To feed back the BBN model with,

potentially, the new probabilities (or even

evidences) for some of the input variables. Input

variables in the BBN model could be indirectly

related to each other, so the optimization will

suggest the optimized value for one, based on a

change in the other one.

3 CASE STUDY

The City of Calgary business unit for Waste &

Recycling Services (WRS) manages residential

waste and recycling collection for 300,000

residential homes and operates three landfills and

various community recycling depots. Examples of

the decisions made regularly at the WRS include:

• Strategic level decisions like the budget, type

and level of the services, training of the staff,

etc.; and

• Tactical (or operational) level decisions like

the number of humans and vehicles, routing

of the vehicles, etc.

In the following subsections, the results of the

case study evaluation for each component of the

decision support system, presented in Section 0, will

be discussed.

3.1 User Layer

Examples of the services at WRS are: collection of

the residential waste and recycling, commercial

waste and recycling, and Christmas tree collection.

These services are unique as they need their own

planning, budget, resources, and income. The cost of

the services is measured by the actual resource

consumption, which are mainly the human/vehicle

used. The income (benefit) is a bit different though.

For the commercial collection, the benefit is simply

the charge, but for the residential units it’s measured

as the quality of the services (QoS). The WRS runs a

survey each year to measure the customer

satisfaction and interprets it as the QoS. So, if WRS

increases the collection days per week for the

residential waste, the QoS will increase but at the

same time the cost will increase too. Therefore, the

trade-off between cost and QoS is always pursued.

3.2 Strategic Layer

In this experiment we focused on the residential

waste collection. To elicit the system variables, we

used a tool named Very Best Choice Light

(VBC)

(Ruhe, 2010). VBC is a collaborative DSS for

eliciting and ranking system variables, requirements,

or features. Stakeholders are defined in VBC to rank

the variables. Consulting with the WRS experts, 20

stakeholders (from WRS and some external ones)

and 20 initial variables were defined and devised as

5 groups: human, vehicle, routing, quality of service,

and logistics. The stakeholders were asked to:

• Revise the variables, introduce new ones, or

remove existing ones

• Rank the variables based on their impact on

the cost

We selected the top 15 variables and built the

BBN model using knowledge of the domain experts

at WRS. The model and its variables are accessible

on (Livani, 2011). SamIam (SamIam, 2011) was

used to analyze the BBN model.

The objective function of Formulas 1 and 2 has

been instantiated as Formulas 3 and 4, for Cost and

QoS respectively.

(



,,



)=





− 







+(





− 





)

(3)

(



,,



)=





−





+(





−





)

(4)

In Formula 3, 





is the probability of Cost

being ‘High’ for scenario S

and 





for the

baseline scenario (S

). 





is the probability of

Cost being ‘Low’ for scenario S

and 





for the

baseline. The objective function is defined similarly

for QoS as Formula 4. So if the probability of being

‘High’ is increasing by a change in the inputs, it

means the Cost (or QoS) is increasing in that

scenario. But if the probability of being ‘Low’ is

increasing, it means the Cost (or QoS) is decreasing.

We’ve ignored the ‘Medium’ category for now

because increasing (or decreasing) of it doesn’t

affect the Cost or QoS directly.

The baseline scenario resulted in probabilities for

Cost and QoS respectively being as (46%, 49%, 5%)

and (67%, 18%, 15%) for (High, Medium, Low)

categories.

The next step is the sensitivity analysis. We

created 75 scenarios by setting each variable (root

and internal variables) to one of its possible states at

a time (as an evidence), while keeping the other

variables unchanged. The scenarios can be found on

(Livani, 2011). Two graphs have been created, one

for the root (Figure 2) and one for the internal

variables (Figure 3). The circled points in each graph

show the Pareto points, which dominate the other

points in both Cost and QoS aspects.

DECISION SUPPORT SYSTEM FOR COST-BENEFIT ANALYSIS IN SERVICE PROVISION

201

Figure 2: Cost vs. QoS trade-off for input variables.

Figure 3: Cost vs. QoS trade-off for internal variables.

There is a difference between Figures 2 and 3.

Pareto points in Figure 2 are related to the input

variables, so they should be possible to achieve

because they are the user inputs. But in Figure 3 the

Pareto points are related to the internal variables

which will then create new probabilities for the input

variables. These new values might not be always

achievable due to the restrictions in the inputs. So,

the interaction with the user is needed to adjust the

probabilities.

3.3 Tactical Layer

We applied the DCCP algorithm to a part of the road

network of the City of Calgary. Each part is named a

‘beat’ and is defined as an area of the city which can

be services by one vehicle in one day. The data,

provided by the WRS, contained the roads and

intersections between them, length and direction of

the roads.

The optimized routes, created by the DCPP

algorithm, showed 20% improvement in the total

length of the routes taken by the trucks, compared to

the actual routes taken by the city vehicles. We also

integrated our results with ArcGIS (ESRI, 2011) to

visualize the routes, available at (Livani, 2011).

3.4 Interaction between Strategic and

Tactical Layers

The goal of the tactical layer is not just optimizing

the operational resources. The results of the tactical

layer are fed back to the strategic layer to re-analyze

the model. One of the strategic variables in the BBN

model is the ‘KM Travelled per day’. This variable

is directly affected by the beat design, which is

usually unique for each service (waste, recycling,

etc.). So, any change at the strategic layer which has

an impact on the beat designs, needs to be further

evaluated at the tactical layer by the optimization

component. New values for this variables leads to

new probabilities for the system variables. Therefore,

the BBN model must be re-run. Another impact of

the optimized routes will be decreasing the

productivity of the collectors (human resources)

every time that new routes are created. Therefore,

again, the model needs to be re-run and new Pareto

points will be generated.

4 CONCLUSIONS & FUTURE

WORK

In this paper a novel decision support system for

cost-benefit analysis in service provision has been

proposed. It consists of three layers: user, strategic,

and tactical. Services and their mapping to system

variables are defined at the user layer. At the

strategic layer, Bayesian belief network (BBN) is

used to analyze the effect of the input variables on

the outputs (here cost and benefit). Results are

presented in the form of trade-off between cost and

benefit; using Pareto optimal solution.

The strategic decisions will be evaluated further

at the tactical layer through resource optimization.

We evaluated our DSS in a case study with the

Waste and Recycling Services (WRS) unit of the

City of Calgary, Canada. Results showed that

analyzing a service at the strategic level and

implementing it at the tactical level is not enough.

Instead, the optimization results must be analyzed to

see which variables are impacted by the new values.

Then the BBN must be re-run to create new Pareto

points. This will lead to an iterative process for

evaluating and composing the new services.

In this paper the initial (whilst recent) evaluation

-40%

-20%

20%

-20% 0% 20%

F(., QoS, S

)

F(., Cost, S

)

-40%

-20%

20%

40%

-70% -20% 30%

F(., QoS, S

)

F(., Cost, S

)

ICEIS 2011 - 13th International Conference on Enterprise Information Systems

202

of an ongoing work towards creating a DSS for

service engineering has been presented. Further

analyses and investigations are needed to increase

the accuracy and acceptance of the results. This can

be done through more discussions with the domain

experts and also mining the data available at the

WRS. Using the Multi-Criteria Decision Analysis to

further analyze and compare the Pareto points is also

among our future works.

ACKNOWLEDGEMENTS

We would like to thank Natural Sciences and

Engineering Research Council (NSERC) of Canada

for partly supporting this research; project no.

386808-09. We also would like to appreciate the

Waste and Recycling Services of the City of Calgary

for their support and assistance when running the

case study.

REFERENCES

Bullinger, H., Fähnrich, K. & Meiren, T., 2003. Service

engineering--methodical development of new service

products. International Journal of Production

Economics, 85(3), pp.275-287.

Cormen, T. H. et al., 2009. Introduction to Algorithms,

Third Edition 3rd ed., The MIT Press.

Edmonds, J. & Johnson, E.L., 1973. Matching, Euler tours

and the Chinese postman. Mathematical Programming,

5(1), pp.88-124.

ESRI, 2011, http://www.esri.com.

Fenton, N. E. & Neil, M., 1999. Software metrics:

successes, failures and new directions. Journal of

Systems and Software, 47(2-3), pp.149-157.

Fenton, N. & Neil, M., 2001. Making decisions: using

Bayesian nets and MCDA. Knowledge-Based Systems,

14(7), pp.307-325.

Fineman, M., Fenton, N. & Radlinski, L., 2009. Modelling

Project Trade-Off Using Bayesian Networks. In

Computational Intelligence and Software Engineering,

2009. CiSE 2009. International Conference on.

Computational Intelligence and Software Engineering,

2009. CiSE 2009. International Conference on. pp. 1-4.

Available at: 10.1109/CISE.2009.5364789 [Accessed

October 15, 2010].

Heckerman, D., 1997. Bayesian Networks for Data

Mining. Data Mining and Knowledge Discovery, 1,

pp.79–119.

Ibrahim, H., Far, B. & Eberlein, A., 2009. Tradeoff and

Sensitivity Analysis of a Hybrid Model for Ranking

Commercial Off-the-Shelf Products. In Engineering of

Computer Based Systems, 2009. ECBS 2009. 16th

Annual IEEE International Conference and Workshop

on the. pp. 119-127. Available at:

10.1109/ECBS.2009.47 [Accessed September 22,

2010].

Kapitsaki, G. M. et al., 2009. Context-aware service

engineering: A survey. Journal of Systems and

Software, 82(8), pp.1285-1297.

Livani, E, http://www.ucalgary.ca/~elivani/DSSCBA.xls.

Liu, K. & Ong, T., 1999. A Modelling Approach for

Handling Business Rules and Exceptions. The

Computer Journal, 42(3), pp.221 -231.

Liu, K., Alderson, A. & Qureshi, Z., 1999. Requirements

recovery from legacy systems by analysing and

modelling behaviour. In Software Maintenance, 1999.

(ICSM '99) Proceedings. IEEE International

Conference on. Software Maintenance, 1999. (ICSM

'99) Proceedings. IEEE International Conference on.

pp. 3-12.

Liu, Y. et al., 2003. A Cost-Benefit Evaluation Server for

decision support in e-business. Decision Support

Systems, 36(1), pp.81-97.

Nanazawa, Y., Suito, H. & Kawarada, H., 2009.

Mathematical study of trade-off relations in logistics

systems. Journal of Computational and Applied

Mathematics, 232(1), pp.122-126.

Ruhe, G., 2010. Product Release Planning: Methods,

Tools and Applications 1st ed., Auerbach Publications.

Sakao, T. & Shimomura, Y., 2007. Service Engineering: a

novel engineering discipline for producers to increase

value combining service and product. Journal of

Cleaner Production, 15(6), pp.590-604.

SamIam, 2011, http://reasoning.cs.ucla.edu/samiam.

Shirazi, A. et al., 2009. A framework for combined

bayesian analysis and optimization for services

delivery. In Service Operations, Logistics and

Informatics, 2009. SOLI '09. IEEE/INFORMS

International Conference on. Service Operations,

Logistics and Informatics, 2009. SOLI '09.

IEEE/INFORMS International Conference on. pp.

382-387. Available at: 10.1109/SOLI.2009.5203963

[Accessed September 14, 2010].

Sundin, E., Öhrwall Rönnbäck, A. & Sakao, T., 2010.

From component to system solution supplier: Strategic

warranty management as a key to efficient integrated

product/service engineering. CIRP Journal of

Manufacturing Science and Technology, 2(3), pp.183-

191.

Thimbleby, H., 2003. The directed Chinese Postman

Problem. Software: Practice and Experience, 33(11),

pp.1081-1096.

Zoric, J., 2011. Connecting business models with service

platform designs: Exploiting potential of scenario

modeling. Telematics and Informatics, 28(1), pp.40-54.

DECISION SUPPORT SYSTEM FOR COST-BENEFIT ANALYSIS IN SERVICE PROVISION

203