Using AI to Combat
Construction Risk Management Bias

While risk is a part of our everyday lives, many studies have shown that humans are notoriously bad at estimating risk probabilities in isolation and even worse when those risks are linked to ourselves.

The recent pandemic brought the whole process of risk identification into sharp focus. When renowned statistician, Professor David Spiegelhalter, tried to communicate risks with the analysis that “the chance you have of dying from COVID if you catch it is roughly the same as your risk of dying over the coming 12 months anyway,”[1] he ran headlong into the other issue with risk analysis. This is namely that most of us aren’t statisticians and therefore don’t use the same language with the same nuances as those who are.

The result? Though he was actually trying to communicate that COVID would effectively double a person’s risk of dying within a single year, he came off sounding as if he were saying that COVID was nothing to worry about, thus the resulting blowback from readers and critics alike.

So, how do we, as project professionals with a deep regard for uncertainty and its impacts on construction, go about capturing construction risks for our projects in a way that is meaningful and realistic without expecting our stakeholders to have a doctorate in statistics? Artificial Intelligence (AI) can play a big part. Let’s examine how.


Understanding Our Own Perceptions of Risk

To do this we need to first understand what underpins the fact that most people can’t estimate risks. The UK Government’s Behavioural Insights team recently wrote that our perception of risk is skewed in three main ways.[2]

First, the extent to which the risk is uncertain or invisible, i.e., “certainty.”[3] It’s the reason that people tend to think that nuclear power is more dangerous than coal — many of us have experienced using coal for heat, while few of us have had direct experience with nuclear reactors.

Second, the “controllability” factor where risks are downplayed if you believe you control them. This is why people who are afraid of flying will happily drive long distances to avoid taking a plane despite driving being 75 times riskier in relation to injury than flying.

Finally, “availability,” which accounts for our tendency to judge events based on how easy it is to think of examples, which leads to overestimation of exceptional events such as shark attacks which attract sensationalist reporting. The truth? It’s vending machines you really want to be careful of as they’re 2.5 times more likely to kill you than a shark is.[4]


Gathering and Consuming Risk Data

This all presents a real challenge as these elements are central to being human. As such the answer must lie in how we communicate information on risks both as gatherers and consumers. Fortunately, the barriers to wide-scale risk identification and comparison are rapidly shrinking as digital tools become available to ease the process of collaboration and tap into the so-called “wisdom of crowds.” This is the idea that large groups of people are collectively smarter than individual experts when it comes to problem-solving, decision-making, innovating and predicting.

This type of crowd-sourced, digitalized, real-time thinking can help the construction risk manager without requiring them to manually update multiple spreadsheets and registers. I see this rapidly taking hold as more organizations take advantage of these tools as the risk profession transitions away from older, more “traditional” systems.

In my opinion, however, the bigger opportunity exists where we can take the data that we hold across our entire project portfolio (including historic actuals) to extend the wisdom to our entire corporate memory through what are called knowledge libraries. In this example, the software should be able to review the entire history of projects, their risks and impacts, to determine potential similarities and to provide a baseline set of standard construction risks for the activities in the new project.

This exciting development, driven by the use of AI and machine learning, provides information to risk practitioners and other stakeholders to properly frame the discussion with the information needed to make us more aware of our all-too-human biases. This will lead to a situation where systems pro-actively highlight activities and risks based on other occasions when similar work has been performed and the real-world impacts that have occurred.


Looking Ahead at Construction Risk Management

What might be the next step? An even deeper level of understanding might come from an AI’s ability to process small changes across large datasets, for example the use of time series data such as weather patterns. This might allow the system to review an entire schedule, cross-referenced with location, weather data and “downtime” costs. This could provide a dynamic view of the project where every activity that could be affected by extreme weather is highlighted and construction risk adjusted based on a truly dispassionate view of the data. In fact, our biggest challenge then might be the human pushback on what the data analysis is telling us!

Leveraging this information will not only improve the risk management process as it exists on our projects, but it will also inform its users, making the identification and classification of risks better as we challenge the bias demons of certainty, controllability and availability.

Ready to take a deeper dive? InEight can help get your projects where they need to go and help you create a solution or view that matches your needs while leveraging your teams’ existing strengths. Let us show you how.



[1] How Much “Normal” Risk Does COVID Represent? WintonCentre, March 21, 2020.

[2] Improving People’s Risk Perception of Coronavirus, The Behavioural Insights Team, September, 2021.

[3] What Makes Risk Acceptable? Revisiting The 1978 Psychological Dimensions of Perceptions of Technological Risks, Journal of Mathematical Psychology, Science Direct, December, 2016.

[4] Availability: A Heuristic for Judging Frequency and Probability, Cognitive Psychology, Science Direct, September, 1973.


Blog Tags