New Times,
New Thinking.

Sponsored by WSP

Data science can help developers design future-proof infrastructure

Businesses can use digital tools to better consider the impacts of climate change, political turbulence and technology.

Designing any new infrastructure means planning for uncertainty. Whether it’s a bridge, a building, a reservoir, an offshore wind turbine, or a new broadband tower, such assets are costly and time-consuming to produce. They will remain in place for many years, so need to be fit for purpose throughout their long life-cycle.

The unknowns we face today are arguably more pronounced than ever before. As well as the risks that come with any big project, such as the cost-benefit ratio, planners must contend with political turbulence, shifts in population demand, and technological curveballs. Most destabilising of all is climate change, which can be hard to model with accuracy.

“Does it mean the location is going to be flooded regularly in the future, [or] that it’s going to be very dry and arid, [or] it’s going to have to deal with high winds?” says Dan Scott, chief data scientist at multidisciplinary professional services consultancy WSP. “Even when we’ve got reasonable historical data, the future is going to look quite different to the past.”

Data science can’t predict the road ahead with certainty but it can play a key role in the planning process. As Gianvito Lanzolla, professor of strategy at Bayes Business School, explains, digitising a physical asset greatly improves your understanding of it, not least when it comes to using predictive technology to maintain it.

“These are very complex models that break down complex problems into smaller ones, and then adapt to different solutions,” he says. “It gives you a view on how to change your choices vis-à-vis changes in the external parameters.”

It other words, it can help businesses assess possible eventualities and minimise their regrets over the long run. “It’s almost like you’ve got an infinite number of parallel universes in front of you, and the future unfolds in a slightly different way in each one,” says Scott. “What’s the best plan for each of those futures?”

In practice, this involves capturing what Scott calls “an upper and lower band of believability”, depending on the data that’s available and the organisation’s attitude to risk. While you may not know what’s coming next, you can probably imagine a best- and worst-case scenario. The idea is to test the whole range of cases in between, on the basis that the truth is in there somewhere.

Sometimes, this process may yield clear answers, while other times the answer might be to wait until more information becomes available. You might build in pre-determined “pivot points”, where dependent on circumstance, you pursue option A or option B. Either way, it’s important to locate a plan of action. So-called “analysis paralysis” – inaction in the face of a thorny problem – has been known to cost organisations dearly.

“There is no perfect choice,” says Lanzolla. “But there is one thing that we know very well in strategy, which is ‘no choice’ is always worse than making suboptimal choices. That’s not a trivial thing to be reminded of when you’re a CEO.”

While it’s early days for these kinds of analytical tools, they are already being applied across various infrastructure sectors. For instance, the UK water industry is using adaptive pathway approaches in its long-term planning process. This means answering questions about future supply and demand, and determining how much water will need to be extracted, desalinated or stored.

“It can easily be 15 years from the point of deciding you want a reservoir to actually having one,” says Scott. “It’s difficult to make a business case if you don’t know for certain that you need it. That’s where the adaptive planning piece comes in – you can test all these different futures and see in how many of them you need that reservoir. If it’s needed in more of those futures than not, it’s easier to make the decision to build it.”

Within the offshore wind sector, adaptive planning can help developers with decisions such as size or site (for example, what is the future price of electricity likely to be, and how likely is it that this stretch of ocean will be unsuitable for biodiversity conservation reasons?). In the environmental sector, these tools can also assess the need for extra flood barriers, while in the telecoms sector they can help determine the most optimal ways to lay down cable.

As well as weighing up different futures, analytical tools can be used to balance competing business objectives. That might mean establishing a priority order for different jobs or working out how to balance environmental impact against cost efficiency. “It’s about articulating your organisation’s objectives and then trying to find the mathematically best solution to the problem that you’re facing, allowing an algorithm to pull the levers the same way as the [executive] board would,” says Scott.

In doing so, you can eliminate some of the cognitive traps that come with human decision-making. Even with an asset that’s designed to last for 50 years, people tend to prioritise short-term needs. A more mathematical approach can weed out these biases, and it can offer flexibility to change the plan when unforeseen events arise. Scott estimates that, compared to traditional business processes, an algorithm-driven approach can cut costs by around 20 per cent.

Rather than “taking decisions away from the board”, predictive analysis can be a “decision support tool”, he says. Using predictive tools can help board members compare forecasts from their own decisions to the algorithm’s decisions, and show how future outcomes can look better or worse.

Despite these advantages, Lanzolla points out that deferring to an algorithm isn’t always an easy sell for businesses. “There are two issues – one is cognitive, another one is emotional,” he says. “People don’t necessarily trust the tools as the model is so complex, and they might feel uncomfortable even once they understand it. It’s important to help users build that comfort, while reassuring them that the final call doesn’t go to the algorithm but to the person who bears the risk.”

On top of that, he cautions that these models aren’t very well-equipped to deal with outlier events, which fall outside the parameters established. If each piece of infrastructure uses the same data in its modelling, then in the case of a black swan event – an event with negative outcomes that is impossible to predict – “all the bridges fall down at the same time”. He thinks it’s better to build resilience by using several planning systems with different parameters.

Yet another barrier to adoption, adds Scott, is the fact that businesses have been “flooded with well-meaning but ultimately flawed ideas” around artificial intelligence. Many of them have spent a lot of time and money rolling out systems that haven’t delivered on the hype. Consequently, they tend to shut down at any suggestion that data science might benefit them.

As Scott sees it, the reason businesses aren’t making better decisions isn’t because they didn’t have the data – it’s because there were other things standing in the way. Adaptive planning tools can help – they are not just about implementing the latest shiny technology, but about using the data you have more thoughtfully. 

 “AI can’t solve every issue affecting infrastructure but applied in the right way with the right leverage, it will have a massive impact,” says Scott.

[See also: We need long-term investment in critical national infrastructure]

Topics in this article : , , , ,