Abstract
During the planning process, project time management includes activity definition, activity sequencing, resource estimating, duration estimating and schedule development. Existing tools and techniques for activity duration estimating include expert judgment, analogous estimating, parametric estimating, as well as three-point estimates. Duration estimating databases that capture the organisation's knowledge are an invaluable input for activity duration estimating. Repetitive projects, such as those appearing in construction or aerospace industry may benefit the most from the existence of efficient duration estimating databases. This paper looks at the processes necessary to both gather the required information and utilise it to assure the efficiency of the time project management process. The first part of the paper presents the available tools and techniques for project time management, while the second part discusses a procedure for activity duration estimating and knowledge management, in repetitive projects.
Introduction
While success in most of the projects is measured, mainly, against cost, provided that time, scope and quality are not extensively violated, there are plenty occurrences where the main concern in project management is time management. Such occurrences constitute projects related to a specific deadline that cannot be loosen under any circumstances. For instance Olympic Games, a Satellite launch, Elections for Government or a millennium project (Atkinson, 1999) have a specific date set in advance that cannot be overrun. Even if we have a serious limitation in scope or we have to suffer a budget overrun, we cannot accept a delay in the starting ceremony of the Olympic Games. Thus, in such projects that are strongly time-constrained, time management is of huge importance. Furthermore, even if a project is not, mainly, time-constrained, completing projects on time, generally, is an indicator of an efficient industry (Chan & Kumaraswamy, 1997).
The literature concerning project time management is extensive, containing many different approaches. However, in almost any case the steps within the proposed approaches are quite similar presented with a different terminology. Early as 1990, Karshenas and Haber (1990) divide project time management (the planning part) in programming, scheduling and resource allocation. Programming was supposed to include the determination of the activities, their duration and their sequence. Scheduling should define the calendar time limits for each activities and the last phase was to assign resources to each activity. Quite later, Gelinas (1999) sets project planning (time management) as the definition of the activities and sub-activities to be realised, determination of resources required and lengths of the activities and sub-activities and sequencing of the activities and distribution of the resources. More systematically, but following the same concept, Burke (2002) recognises that the work breakdown structure is really the initiation of project time management since it is the source of project activities. Having defined the project activities one should move on determining logical connections (dependencies) and then define the duration of each activity. Right afterwards a scheduling tool (like CPM) should be used and a schedule will be provided after taking into account calendar restrictions. Resources are separately handled and are calculated according to schedule needs. The differentiation to this concept comes with Verzuh (2003) who presents a six-step detailed planning process that stresses the important role of resources in scheduling. Verzuh's process consists of developing a work breakdown structure, sequencing the tasks, estimating work packages, calculating an initial schedule, assigning and levelling resources and developing a budget. Maylor (2003) goes a small step further suggesting the concurrent calculation of activities durations and resources needs, which is probably had exactly happens in practice where managers have to balance duration restrictions with resources restrictions.
It has to be stated here that resource constrained scheduling was a topic that has been concerning the scientific community since 1960s. However, the different approaches differ in whether we first assign resources and then calculate durations or we first assign durations and then calculate resources. All the aforementioned suggestions have been taken into account and finally PMI (2004) proposed that project time management should include activity definition, activity sequencing, activity resource estimating, activity duration estimating, schedule development and schedule control.
Apart from the Project Management Institute (PMI®) (2004), which gives certain importance to organisational process assets, the rest of approaches are mainly focusing on planning tools. Alquier, Cagno, Caron, Leopoulos and Ridao (2000) have stressed the importance of the existence of a well maintained historical information database, which they name corporate memory, within the concept of time management, especially during the early stages of the project's life cycle. They define this corporate memory as knowledge-constructive and knowledge-emergent. Knowledge-constructive as it supports adjustment to new cases when reusing knowledge, ontology restructuring and cases addition, and knowledge-emergent as it allows tacit knowledge capture from senior managers, learning from experience and supports best practises.
Repetitive projects, such as those appearing in construction or aerospace industry may benefit the most from the existence of such, efficient, duration estimating databases. This paper addresses this issue by discussing a procedure for activity duration estimating and overall project duration estimating, in repetitive projects based on activity duration estimating through knowledge management techniques (Leopoulos, Kirytopoulos & Bellos. 2002).
The procedure aims to feed appropriate data to a Monte Carlo Simulation model for schedule development, thus incorporating risk and uncertainty in schedule development.
Project Time Management
Project Time Management Processes
Project time management includes the processes required to accomplish timely completion of the project (PMI, 2004 p. 123). The core problem of project time management concerns activity definition, activity sequencing, activity resource estimating, activity duration estimating, schedule development and schedule control. The outcome of project time management is a project schedule that includes start and finish dates of each activity. Even if no risks appear, the project is unlikely to finish on schedule if either the duration of activities or the definition of dependencies is not realistic (Diamantas, Kirytopoulos & Leopoulos., 2006).
Project Schedule Development
Project schedule development determines planned start and finish dates for project activities. Schedule development can require that duration estimates and resource estimates are reviewed and revised to create an approved project schedule that can serve as a baseline against which progress can be tracked (PMI, 2004 p. 143).
The most commonly used tools for schedule development are the network analysis (Burke 2002), the Monte Carlo simulation – MCS (van Slyke, 1963), and the Schedule Model (PMI, 2004).
Schedule network analysis is a technique that generates the project schedule. It calculates the early start and finish, the late start and finish and the float of work elements in a project given their duration and logical dependency (Turner, 1999). Network techniques include the CPM, the Program Evaluation and Review Technique – PERT, and the Graphical Evaluation and Review Technique – GERT. Network techniques allow the management of essential time management problems such as scheduling, cost planning and resource allocation (Pagnoni, 1990).
CPM is a mathematically based algorithm for scheduling a set of project activities. It was developed in the 1950's in a joint venture between DuPont Corporation and Remington Rand Corporation for managing plant maintenance projects (Maylor, 2003). It can be applied for the scheduling of any project with interdependent activities, whenever the project schedules can be assumed to be free of choices and cycles, and when activities' duration can be deterministically estimated.
PERT is a well known method with proven value in managing complex projects. It was developed in 1958 by the US Navy Special Projects Office as part of the Polaris mobile submarine launched ballistic missile project (Malcolm, Rooseboom, Clark, & Fazer 1959). The main difference between CPM and PERT is the fact that PERT takes the uncertainty associated with activity duration estimating explicitly into account (Pagnoni, 1990). PERT assumes all activity durations to be independent random variables belonging to the same distribution, usually the Beta distribution, and requires for each of them only three duration estimates: the most optimistic, the most likely and the most pessimistic.
GERT was first introduced by Pritsker, in the middle sixties, for planning and analysing terminal count-down of an Apollo space system. GERT allows choices and cycles within the project schedule, as well as the stochastic estimation of activity duration. The analytical investigation of GERT network is not practicable as a rule, thus GERT is naturally oriented toward simulation (Pagnoni, 1990).
Van Slyke (1963) was, probably, the first to suggest the use of MCS to find the cumulative distribution function of project network completion times. He also introduced the use of non-beta duration distributions for modelling activities' duration. MCS involves the random sampling of each probability distribution within the model to produce hundreds or even thousands of iterations. Each probability distribution, modelling activity duration, is sampled in a manner that reproduces the activity's distribution's shape. Therefore, the results of all the iterations form an overall distribution for the model that, in this case reflects the distribution of the overall project duration (Vose, 2000). MCS practitioners often use the three-point duration estimation approach to establish a statistical distribution describing the duration of each activity.
The schedule model tool and the supporting schedule model data are used in conjunction with manual methods of project management software to perform schedule network analysis to generate the project schedule (PMI, 2004).
Activity Duration Estimating – Tools and Techniques
Activity duration estimating is the process during which the duration for each project activity is estimated based on information from the activity scope of work, the required resources types, the estimated resource quantities and resource calendars (PMI, 2004). It is a critical process within time management since an unrealistic estimation of activities' duration will feed with inaccurate data the project model. This may result in wrong decisions based on inaccurate information, or create a false safety feeling to the project manager.
The most commonly used techniques for activity duration estimating are the expert judgment, the analogous estimating, the parametric estimating and the three-point estimates.
Expert judgment uses personal intuition and awareness. The output of expert judgement is either a duration estimate, or a recommended minimum and maximum activity duration based on past experience with similar projects. According to Vose (2000) the uncertainty in such subjective estimates has two components: the inherent uncertainty of the variable itself and the uncertainty arising from the expert's lack of knowledge of the parameters that describe that variability. These uncertainties may or may not be distinguished during the activity duration estimating phase but both types of uncertainty should at least be accounted for. Therefore, the guidance of a historical information database should be used whenever possible.
Analogous duration estimating uses historical information and expert judgement as the basis for estimating the duration of a future activity. The reliability of the estimates produced by analogous duration estimating depends on how similar the future activity is with the past activities, whose actual durations have been recorded (PMI, 2004). Analogous estimating is mainly used in the aerospace industry.
Parametric estimating is the estimating of the activity durations by dividing the quantity of work to be performed by the productivity rate. Parametric estimating is, frequently, used in the construction industry. For example, an activity of the project is the plastering of 100m2 walls. The productivity rate of the plastering team is 20m2 per day. Thus, the activity duration is estimated to be 5 days.
The three-point estimate is a technique for activity duration estimating taking into account the uncertainty in the original estimation. Instead of a single estimate the project team determines an optimistic, a most likely and a pessimistic estimate. The optimistic and pessimistic estimates correspond to the best case and worst case scenario, respectively. The most likely estimate is a realistic expectation of the activity duration given the resources likely to be assigned, their productivity and risks that may affect activity duration.
Project Risk Management
Project risk management is seen as a valuable process built-in to project management. It deals, mostly, with risks that may affect the objectives of the project, namely time, cost, quality, and scope. Risk management is usually described in literature (Turner, 1999; Kirytopoulos, Diamantas, & Leopoulos, 2005) as a four-step process. A thorough literature research indicates that the references as far as the steps of project risk management are concerned are very extensive, although each one is relatively similar to the other. An explicit convergence of opinions can be noticed, as the various approaches differ mainly at the degree of detail or the naming of the process stages. A generic process of risk management consists of risk identification, risk analysis (assessment), risk response (handling), and risk follow-up (monitoring). In the context of this paper the authors focus on risks with impact on project duration, and how that impact should be incorporated within the project schedule.
Turner (1999) suggested the categorisation of risks according to their impact. He recognised two types of risks; business risks and insurable risks. Business risks, sometimes called uncertainty, are the risks inherent in all estimates. They are two-sided risks. Sometimes their impact to the project will turn out to be good (opportunity), while other times it will turn out to be adverse (threat). Insurable risks are risks that can only harm the project (threats). The way from ones home to his work is a very good example of this categorisation. Traffic and traffic lights are considered business risks. They can be a threat (slowing the drive) or an opportunity (speeding up the drive). Measuring and controlling such risks is not always possible, while when it is possible to measure and control them it is, usually, either insignificant or not cost-effective. On the other hand, an accident, such as a car crash or a flat tyre, is an insurable risk (it can only delay us).
Inspired by Turner's categorisation the authors suggest the categorisation of risks into embedded uncertainty and discrete risks (events). Embedded uncertainty is similar to business risks in that it is inherent to any estimate and might have a positive or negative result. However, this uncertainty is very hard to measure or control and its impact to the deviation from a mean value is relatively short. Discrete risks are risks that have a random chance to happen and an impact on project objectives, as well. However, discrete risks, almost always, have a significantly bigger impact than embedded uncertainty and thus they should be controlled. A major difference between discrete risks and Turner's insurable risks is that discrete risks can be a threat or an opportunity instead of a threat only while the difference between Turner's business risks and embedded uncertainty is that embedded uncertainty will have a relatively short impact. Using the same example as above, traffic lights and usual traffic differentiation are categorised as embedded uncertainty while an accident or a tyre blow-out is regarded as a discrete risk.
In project scheduling the three point estimates should depict the embedded uncertainty while discrete risks should be modelled separately.
Problem Statement
CPM is a simple and straightforward scheduling tool; however it fails to, adequately, model the stochastic nature of projects' duration. PERT (Malcolm et al., 1959) was developed in order to take into account the stochastic nature of projects' duration, but, still, suffers from major disadvantages well documented in previous literature (Diamantas et al., 2006). An alternative to PERT is the MCS which seems to be able to handle more complex project scheduling problems and solves most of the limitations that PERT faces. One of the most important advantages of MCS according to the literature is the ability to define the most applicable distribution for each activity's duration instead of defining the most convenient distribution for the whole project (van Slyke, 1963). The definition of the right distributions for the activities' duration is the most critical issue in the application of MCS.
The most commonly used distributions, for the modelling of activity's duration, are the Beta distribution, the Triangular distribution, the Uniform distribution and the Normal distribution (Kirytopoulos, Leopoulos, & Malandrakis, 2001). Each distribution allows a different approach in handling uncertainty. The Beta and Triangular distributions are asymmetrical, while the Normal and Uniform distributions are symmetrical. The definition of the right distributions requires accurate historical information.
Any activity duration estimating technique, described in the previous section, can be applied to provide the data that will be fed to the MCS. However, the authors argue that the transformation of the historical information to usable data is the best approach, although not extensively described in the literature. Thus the authors focus on expert knowledge and propose a knowledge management process to aid the expert in his decision. Furthermore, the authors provide a framework that allows the integration of discrete risks in the project schedule.
Suggested Approach
The suggested approach is divided in three phases. Processes executed before project scheduling, while project scheduling (planning project phase) and after project scheduling (close project phase). In the first phase the organisation has to set up a Knowledge Repository. The second phase uses the knowledge captured in the Knowledge Repository (use of Organisational Process Assets) in order to provide a project schedule. Finally, the last phase tracks the actual durations of each activity during the close project process. During this phase, discrete risks' time impacts are excluded from the actual duration of each activity. Thus, the time impact of risks is recorder at a risk register (part of the Knowledge Repository) and the whole information is enhancing the Knowledge Repository (phase 1) and increases its value to the company (for more on value management refer to Thiry, 1997). The three phases of the suggested approach consist of five steps (Exhibit 1).
The first step (phase 1) is the creation enhancement and maintenance of an accurate historical information database. Whenever a project ends the actual duration of each activity and risk information is recorded in the historical information database, thus improving its maturity level (PMI, 2003).
The second step (phase 2) is the development of the project schedule. The project activities are defined based on the project scope, developing the project's work breakdown structure (WBS). Then, the historical data regarding each activity duration and possible risks are exported from the Knowledge Repository in order to be fed into the fitting process. It should be noted here that, since discrete risks' impacts have been excluded from activities' actual durations (refer to step 5), the fitting of the actual durations for a specific activity will offer the embedded uncertainty.
Vose (2000) defines fitting as a technique to interpret observed data for a variable in order to derive a distribution that realistically models its true variability and the uncertainty about that true variability.
The fitting process begins with the analysis of the properties of the observed data. Vose (2000) suggests considering the following points before attempting a fit:
- Is the variable to be modelled discrete or continuous?
- Do I really need to fit a mathematical (parametric) distribution to my data?
- Does the theoretical range of the variable match that of the fitted distribution?
- Is the variable independent of other variables in the model?
- Does a theoretical distribution exist that fits the mathematics of this variable?
- Does a theoretical distribution exist that is well known to fit this type of variable?
In the context of this paper activity duration is considered as continuous, and the fit to a mathematical distribution necessary. Furthermore, there are well known distributions in the literature that fit activity duration, such as the Beta distribution, the Triangular distribution, the Uniform distribution and the Normal distribution (Kirytopoulos et al., 2001). Vose (2000) describes, in detail, methods of finding either a theoretical (parametric) or an empirical (non-parametric) distribution that best fits the observed data.
As far as the non parametric distributions are concerned it is, usually, sufficient to use a cumulative frequency plot of the data points themselves to define its probability distribution. The available techniques for fitting a parametric distribution include the maximum likehood estimators (MLEs), minimising goodness of fit statistics such as the Chi Squared (X2), the Kolmogorov-Smirnoff (K-S), or the Anderson-Darling (A-D), and the use of goodness of fit plots (Vose, 2000). Further discussion regarding the process of fitting a distribution to historical data is beyond the scope of this paper. The authors use a spreadsheet add-on in order to conduct distribution fitting.
The third step (phase 2) is the quantitative risk analysis based on the information recorded in the risk register. This step consists of the calculation of the stochastic characteristics of each risk, probability of appearance and impact on duration, which will be fed to the schedule model. The probability of appearance is calculated through counting the appearances in the past projects and dividing this number with the total number of projects registered in the historical information. The impact of duration is defined through the calculation of a best fit distribution based on the data available in the Knowledge Repository (what were the impacts in the past) plus a fine tuning from an expert if certain situations demand it (i.e. specific conditions environment of the new project). The fitting process follows the methodology described in step 2.
Exhibit 1: Suggested Approach
The fourth step (phase 2) is the performance of the quantitative analysis for schedule development. In order to perform this analysis the project schedule is developed. It integrates all the information produced in earlier steps, including activity duration estimates, risks' probability of appearance and duration impacts estimates. Finally, a MCS is run and the results are being recorded, analysed and disseminated to the individuals that need this information in order to take decisions.
The fifth step (phase 3) happens after project execution where the outcome is known and concerns the separation of discrete risks' time impact from the total activity duration. The produced duration, free of discrete risks' effects, is stored in the Knowledge Repository in order to be fed, when needed, to the activity duration estimating of step 2. The effects of discrete risks (impacts on duration) are also stored in the Knowledge Repository in order to be taken into account during the quantitative risk analysis of step 3.
Exhibit 2 represents how the final project duration distribution is going to be concluded through the proposed approach.
Exhibit 2: How to reach overall project duration distribution
Discussion - Conclusions
Once having such a process set, the project management team may have an educated guess of the duration of each activity and based on it perform a MCS that will lead to a realistic estimation of the distribution that describes overall project duration (Exhibit 2). However, the structure of the Knowledge Repository should be defined in such a way that will really permit the use of knowledge acquired. For instance, if we just keep track of the information considering just actual duration we will not be able to use the info in the future. Apart from that piece of information we have to keep data such as the number of resources involved, any particular circumstances and any condition that may be of use when we want to exploit that information in order to make duration estimation for this activity in a next project. In other words, we may have to consider organising knowledge in a more efficient way based for instance on a parameter (i.e. excavation: duration parameter m3/h per excavator in rainy environment). Such organisation of the Knowledge Repository can be time consuming, difficult to achieve but worth-trying for future activity duration estimating. The organisation of the Knowledge Repository may be an outcome of a forum discussion.
Uncertainty is going to be incorporated in the process in a structured way that will distinguish embedded uncertainty from discrete risks. This approach will help us define a typical distribution for the duration of each activity (including only embedded uncertainty) and depending on the particular conditions of each project identify specific risks and use them at the project model that will be used in the MCS. The Knowledge Repository may well serve as a risk identification tool (retrieve risks that have emerged or identified for an activity in the past), among other risk identification processes (refer to Hillson, 2004) but may also serve as an analysis tool since the impact of those risks in the past has been recorded. However, specific attention should be given in the similarity of the conditions under which the impact occurred to the existing conditions of the forthcoming project.
The objective of this paper was to present a procedure for activity duration estimating, in repetitive projects (projects that usually have same kind of activities). This procedure aims to feed appropriate data to a Monte Carlo Simulation model for schedule development. The proposed approach is based on knowledge management. Duration estimating databases that capture the organisations' knowledge have always been seen as an invaluable input for activity duration estimating. This paper offers a process of how to strategically use such a database (here referred to as a Knowledge Repository) in order to capture the existing information and use it effectively in activities' duration estimating, thus enhancing organisational process assets.
A forum discussion is needed in order to define any weaknesses of the proposed approach and suggest alternatives or improvements.
