AI special: Can artificial intelligence help avoid another Carillion?

Cost models have been developed that give a clearer picture of the different costs for various elements of a project. It means that different approaches can be compared easily to establish the best way forward from a financial perspective.– Gari Nickson, GenieBelt

Gari Nickson, continues our focus on artificial intelligence in construction with a look at how it can help control costs.

As Carillion’s collapse has reminded the industry, construction is an extremely precarious business. A huge amount of risk is involved, from a financial perspective through to the practicalities of carrying out a project.

Part of the reason for the spectacular demise of Carillion, the UK’s second-largest contractor, was the fact the firm was hit by some big jobs that went wrong – two hospitals and a £550m road project in Scotland – which meant it was forced to take a hefty financial hit.

Could Carillion have been helped by some of the new techniques that utilise AI to help estimate costs on construction jobs?

As we know, controlling costs on a job is often difficult and estimating the price of work is one the most important initial aspects of any project.

“The risk in a project is always probabilistic and the human mind is not good at doing risk-based probability management, especially when we’re combining many different probabilities,” Aptage CEO John Heintz has told He refers to “hope-based planning”.

“It’s natural. We’re to some degree all optimistic. We all see the positive path forward, the way this could work, and we don’t have evidence to prove it can’t work, so we hope it’s going to go the way we want it to,” he says.

Early planning is crucial to the success of a job and quite rightly more emphasis is being placed on preparation before shovels go in the ground.

As the old adage goes: fail to prepare, prepare to fail. In construction this could never be truer.

Factors at play

A whole host of factors are at play in estimating the cost of a job – and can even depend on the amount of experience of the team giving the estimate, through to environmental factors.

In the case of Carillion’s problem Scottish roads job, it would have been great to have a computer to estimate the amount of time that would have been lost to weather. Although perhaps it doesn’t take a genius to work out that working on a roads job in the middle of winter in north Scotland will bring problems…

How does AI in construction management help?

Cost models have been developed that give a clearer picture of the different costs for various elements of a project.

It means that different approaches can be compared easily to establish the best way forward from a financial perspective. It allows a company to work out how much resource to dedicate to each aspect of a job and means that the final cost of the job is much more likely to match that of the original estimate.

The project cost estimation methods fall into five groups, according to a paper in the Journal of Engineering. These are machine-learning, knowledge-based systems, evolutionary systems, agent-based systems and hybrid systems.

Machine learning

Let’s deal with machine learning. ML, as it is known, has a number of advantages and disadvantages. Among its strengths are the ability to deal with uncertainty, the ability to work with incomplete data, and the ability to judge new cases based on acquired experiences from similar cases.

Some researchers have even developed a method for calculating the time and cost on earthworks. Interestingly, it was the failure to finish earthworks before the winter that was reportedly a key problem for Carillion on the Aberdeen roads job.

But machine learning is not perfect. One of its weaknesses is a lack of technical justification, which means the causes beyond the decision are not made clear, known as a black box decision.

Knowledge-based systems

This is defined as any technique that uses logical rules for working out the end result. Two advantages of this system are the ability to justify an outcome and relatively straightforward methods, meaning that it is simple to develop. However, one of the drawbacks is the problem of self-learning.

Evolutionary systems

This approach is a group of intelligent systems centred around continuous optimisation with heuristics (rules, learned or hard-coded by evolutionary processes).

Evolutionary systems are used primarily as optimisation tools where there are many solutions. The ability for it to solve complicated and uncertain problems is one of the main reasons to use this approach.

Agent-based systems

These are regarded as one of the main elements of AI, as they simulate the action and interactions of autonomous agents. Agent-based systems are seen as a way to estimate direct and indirect costs for construction projects.

One group of researchers developed a system which simulated the negotiation process between contractor and client looking at risks and the potential for cost overruns. This system was developed by interviewing eight construction professionals.

Hybrid systems

This approach is a group of techniques that are combined, normally to overcome the limitations of each individual system. The latest methods using AI are focused on hybrid systems.

One drawback of this approach is the lack of computer tools that could support its implementation.

AI for the future

AI remains in its early stages in terms of helping construction professionals. But as can be seen by these techniques, there is plenty to leverage to help humans hopefully avoid some of the problems that crop up on a job.

Meanwhile, construction professionals are becoming more and more aware of the role AI can play. An Institution of Civil Engineers (ICE) survey last year found that 73% of built environment professionals believed AI will have a “significant impact” on their organisations. And nearly eight in 10 (78%) believed AI will have a positive impact on the sector.

As the Carillion crisis has highlighted, at least when it comes to managing construction projects, the more tools at your disposal to estimate risk, the better. In that respect, AI can surely only help in being able to plan projects in a smarter, more risk-free way. 

Gari Nickson is an expert in the application of artificial intelligence in construction. He’s an entrepreneur, co-founder of, and adviser to Contractor Freedom

Story for BIM+? Get in touch via email: [email protected]


  1. Great theory, but is this relevant to practice? How many companies win contracts on the basis of how much they think the job will cost to get done as opposed to how low they think they need to bid to win the job in the first place? I would suggest that for a number of reasons the latter takes presidence and hence companies survive on very low margins and cashflow manipulation. Perhaps it would be better to equip clients with the type of analysis outlined in this article so that they can assess the viability of bids and the risk of a contractor going bust because a bid is too low.

  2. Paul – very true – one of the things we are trying to push more at GenieBelt is clients taking ownership and I love your idea as it really strikes a chord with what I am hearing from the industry

  3. Gari Nickson,
    AI cannot help to identify and play with human psychology and behavior, I mean cost models with managing people and lead changes. Still, leaders in the industry are continuously failing to understand by using AI at different levels. However, your paper is good to read for a research student. The main problem is the relationship gap between the client and the main contractor, apart from the cost factor and risk factor.
    Raja Pillai

  4. AI can help manage a process, but the process is dictated by policy and policy is written by humans/MDs aka the Richard Howsons.
    Carillion making the decision to bid for more work than their own internal infrastructure would allow couldnt have been avoided using AI. Businesses gamble, if the gamble works then money is made, in Carillions case the gamble to grow and bid for huge projects where they new risks were high didnt pay off. AI cant predict these risks nor can it manage them. Especially when AI needs all the information available to make its decisions which snr management often omits as its too sensitive.

Comments are closed.

Latest articles in Analysis