04 We believe algorithms drive outcomes

Reducing cost and improving value

Organisations from many different sectors of industry, including utilities, public transport and logistics, face similar challenges in allocation of resources.

These organisations typically have diverse portfolios of large, long-lived assets, high maintenance and development costs, and need to allocate capital to refurbishment or replacement of these assets.

The funding allocation for large physical assets


With constraints on capital and competition for resources, optimum decision making becomes increasingly critical to success. Adding further complexity to the decision as to which projects to undertake, are factors that may include:

  • Budget limits across multiple time periods and categories;
  • Availability of staff and equipment;
  • Financial return;
  • Non-financial benefits;
  • Capacity to meet future demand;
  • Strategic initiatives;
  • Uncertainties and risks; and
  • Mutually exclusive or dependent projects.

Under these constraints, asset managers inevitably need to tradeoff between competing projects to find a portfolio that provides the optimum level of value in the most efficient and prudent way. There are significant benefits to choosing the right optimisation methods:

  • Significantly lower expenses;
  • Significantly improved customer outcomes; and
  • Significantly better asset utilisation.


We are searching for the efficient frontier


The diagram below shows the efficient frontier – the cost and value of each portfolio that achieves the highest possible value within a given cost limit – for a typical set of candidate projects. A portfolio chosen by simple methods such as simple ranking will generally be suboptimal in the sense that it does not achieve the best possible value for a given cost

Identifying the efficient frontier

Choosing the right tool is crucial


Portfolio optimisation is an inherently difficult and computationally intensive undertaking. With only a modest number of projects there can be millions of potential portfolios to choose from, and the number of combinations doubles with every new project added into the mix. This, together with human behaviour and inevitable time pressures can often lead to the selection of sub-optimal portfolios.

Some common basic approaches to portfolio optimisation include:

  • Project-by-project approval – projects are approved individually on a firstin-first served basis. Without a holistic view of the portfolio, this can lead to many potential problems such as having too many small projects with low value and resources spread too thinly; and
  • Approval by ranking – projects are placed in order of priority, often using a spreadsheet-based tool, and approved according to whether they fit within the remaining budget. As discussed in the next sections, this method does not easily handle multiple constraints and can give suboptimal results even for very small sets of projects. Some other common approaches to portfolio optimisation are based around search methods and include:
  • Trial and error – a sample of random potential portfolios is examined;
  • Tweaking – a potential portfolio is modified slightly in the hope of obtaining an improvement. This method can get stuck on “local optima”, for example where removal of several large projects allows for addition of several small but high value projects;
  • Enumeration – all possible portfolios are tested one by one. This method becomes impractical even for modest numbers of projects; and
  • Evolution – pairs of potential portfolios are selected and combined to see if a better one is obtained. Genetic algorithms are very popular as optimisation tools, but suffer from the weaknesses that they require careful tuning and, once a potential solution is reached it cannot be guaranteed that no better one exists.


Our approach finds the efficient frontier


The tools that we developed at Readify are based on customised state-of-the-art optimisation algorithms and can provide accurate solutions in a fraction of the time required by other methods. Unlike randomised search-based methods, the tool uses robust mathematical techniques drawn from constrained optimisation theory that prove that the tool’s solution is indeed the optimal portfolio. Included with the extremely fast portfolio optimisation engine is a robust tool for quickly finding the efficient frontier. This not only allows asset managers to be confident in the results, but potentially enables them to perform other important analysis such as scenario planning and sensitivity testing.

Even for four projects the traditional methods fall short


Consider the following small example where there are four projects and a budget of $10 million. For this example there is no simple ranking method that obtains the optimal portfolio. 

Ranking falls short even in simple scenarios

Project A has the highest value but also the highest cost and worst value for money. Project D has the lowest value but also the lowest cost and best value for money. Selection by ranking produces sub optimal portfolios:

  • Ranking by value (highest first), first project A is selected and then project B, for a total of 9 points and all $10 million spent
  • Ranking by cost (lowest first), projects D, C, and B are selected for a total of 9 points and $7 million spent
  • Ranking by value for money, projects A, and B are selected for a total of 9 points and $10 million spent
  • The optimal portfolio consists of project A, C, and D for a total of 10 points for $9 million spent

In ranking by value, the budget was quickly spent on the most expensive projects before the small bargain value projects could be considered. In ranking by value-to-cost ratio, after the bargain value projects were all taken there was insufficient budget to include the largest project.

The optimal portfolio will generally be a mix of high value-to-cost, high value and low cost projects. Of course for this small example it is easy to reach the optimal portfolio by small adjustment of either of the suboptimal ranked portfolios, but in practice with larger portfolios the required adjustments are too many and too complex to perform without computer assistance. 

The more potential projects, the greater the advantage 


We compared the performance of several methods on a client example involving 490 projects and the results are displayed in the accompanying figure. The portfolio is subject to an additional constraint that 60% of the budget is spent on projects of a specific type. The efficient frontier was generated by Readify's tool which achieved up to double the value of the portfolio produced by the best of the three different ranking methods. The genetic algorithm performed better than the ranking methods in places but does not achieve the true optimum. Note that the simple random search failed to find any portfolios that satisfied the constraint.


Approaches to a 490 project portfolio

In closing


Project portfolio optimisation is a challenging problem  faced by many large organisations. For even modest-sized portfolios, the sheer number of combinations makes simple search strategies impractical or unreliable. In particular, ranking methods, which are popular for their relative simplicity, do not perform well with complex constraints and fall short even for small examples, while optimisation by tweaking can be time consuming and unreliable. Readify has developed advanced tools that can lead to significant improvements in cost savings, value and customer outcomes for minimal effort, and potentially free up staff time for more complex tasks such as sensitivity analysis and scenario planning.