What construction can learn from weather forecasts: Improving how we communicate risk
In this post, we explore techniques to explain risk and uncertainty in construction projects to enable better decision making.
At nPlan, we are solving the momentous problem that infrastructure projects are consistently delayed and over budget. We use machine learning to gather insights from historic schedules, which enables us to understand the complexity of a project, learn about the tendency of activities to overrun and extrapolate information to predict outcomes of future projects. The task of learning from hundreds of thousands of highly complex infrastructure schedules is far from trivial.
For a more extensive background on nPlan and why we are excited about the profound implications of the problem we solve, please see Dev’s previous blog.
There is another challenge we face, which poses a very different question — How can we represent risk, uncertainty and future events? How can we help our users understand predictions which are inherently probabilistic?
“We are a species that is uniformly probability-blind”
Piattelli-Palmarini (1)
The planned duration of an activity, by the premise of project management, is highly uncertain. There is a chance, or a certain probability, that an activity will finish on time, a probability that it will finish early, and a probability it might finish late. Communicated correctly, the probability that an activity will finish on time is a simple concept, yet, as has been shown across multiple domains — from medical statistics to political outcomes — probabilities are misinterpreted more often than not.
Framing probabilistic outcomes
Consider the following statements:
1) Of the last one thousand pillars we have built, one was delayed.2) There is a 99.9% probability that no pillars will be delayed.
They imply similar information, but with very different connotations. One is framed in the context of the present and makes us feel optimistic, the other makes an assessment derived from historic data and has negative connotations. As has been frequently shown by pioneers of behavioural economics, Tversky & Kahneman (2), we are more likely to trust a contractor who communicates the latter. The cognitive process of translating a probability into something comprehensible is, for most, more effortful than the information is worth. Given the vast scale of mega projects — tens-to-hundreds of thousands of activities — the consequences of making this error might be colossal. Fortunately, we are surrounded by naturally occurring numbers in our daily lives (3), and if probability is presented in a way which looks familiar to us, we can grasp the information it is conveying very quickly.
In short, when communicating probabilities, we must make sure that our audience is able to interpret the information in an unbiased and effortless way.
Drawing attention to colour
In a world where attention-grabbing content fuels both the global economy and fills our days, our brains have limited ability to store, process and recall information. This problem is only exacerbated when we have to think about numbers with decimal points or percentage symbols. There are a few notable ways to help us better capture a user’s attention, but these provide limited value in helping us communicate probability. One technique is the use of colour. The use of variation in colour, for example using a palette scale, can represent a lot of complex information in a mentally intuitively way. Using colour helps us to remember information.
For centuries map-makers, and more recently weather forecasters, have known that one of the best ways to demonstrate likelihood is through colour. A cold weather front will become more blue depending on both the intensity and likelihood of coldness it will cause. In hurricane forecasting, an audience’s ability to quickly understand forecasts can make the difference between life or death. Predicting the likelihood of a project’s activity finishing on time can, of course, be communicated in a similar way. When appropriately coloured, a Gantt chart is transformed from a temporal device to a powerful probabilistic communication tool.
Spatial Visualisation
We often have to represent information in more dimensions beyond the two axes of a traditional graph. Spatial visualisation of information is becoming an increasingly popular way to communicate information, whether used to tell stories, or to convey news. Different shapes can be used to represent different pieces of information, and this becomes an even more powerful tool when attempting to present a complex piece of information such as probability. There are few techniques on how to effectively communicate data with shapes and visualisations, so the best approach is to search: experiment with different shapes, dimensions, charts. There will be specific ways that both engage an audience, and communicate a piece of information in the most effective way.
Machine learning is not easy, but it’s a tractable solution. Our goal at nPlan is to make the results of phenomenally powerful algorithms accessible to all. By putting into practice the above techniques in our product thinking and execution, we aim to deliver a platform which radically improves the way risk and uncertainty are understood in the execution of construction projects.
We’re working with some of the most influential organisations in construction worldwide. If you’re interested in learning more about improving your understanding of risk and uncertainty, please feel free to email me at nick@nplan.io
With thanks to; Prof. Yael Grushka-Cockayne at Harvard Business School for her comments and resources, and my colleagues Sarah, Toby, Dev, Alan, João and Gary for their input.
- Piattelli-Palmarini, M. (1994). Ever since language and learning: Afterthoughts on the Piaget-Chomsky debate. Cognition, 50(1–3), 315–346.
- Tversky, A & Kahneman, D. (1981). The Framing of Decisions and the Psychology of Choice
- Gigerenzer, G (2003) Simple tools for understanding risks: from innumeracy to insight
