Stepping onto the cavernous Elizabeth Line platform at Tottenham Court Road, it's easy to forget the five years of delay and £4 billion unanticipated expense that preceded Crossrail's completion. Startup nPlan doesn't forget it. In 2018, the Spitalfields-based firm claimed its AI technology – trained on a vast library of historical construction data – could forecast the project far more accurately and more quickly than the teams of consultants and experts tasked with the job.
"We forecast how the project would play out, using only the construction schedules from the project which were available up to 2012," co-founder Dev Amratia recalls. "When we checked back on Crossrail's completion, our prediction matched what actually happened - and was far quicker than traditional megaproject forecasting."
nPlan's analysis came too late to change anything for Crossrail. But it caught the eye of a senior transport delivery figure, who hired the firm to work on the Transpennine Route Upgrade, an £11 billion programme to electrify and modernise the rail route between Manchester, Leeds and York. Using nPlan's AI to assure track possessions is now written into TRU's contracts. That megaproject is currently on time and on budget.
That's unusual in infrastructure construction. Data suggests that six out of seven large-scale construction projects finish late, with one in ten overrunning by more than a year. The Project Management Institute estimates that for every billion dollars spent on projects worldwide, $127 million is wasted.
Prior to co-founding nPlan, Amratia saw the challenges firsthand in his nine years delivering complex construction projects for Shell across three continents, including a stint as project manager on a $25 billion gas plant in Ras Laffan, Qatar. "We would have happily paid a 30% premium to any contractor who could guarantee the build would be delivered on time and on budget," he says. "Nobody would take the deal. We were flying in the dark, burning cash on decisions based on intuition."
So Amratia teamed up with Alan Mosca, an AI engineer with a background in quantitative finance and high-performance computing, in 2017. Mosca was looking to work on problems with a more tangible impact. Following the collapse of contractor Carillion, he too concluded the risks in megaproject construction were too high. The two met at Entrepreneur First, and started building - becoming close friends in the process: Mosca was best man at Amratia's wedding.
The idea behind nPlan quickly gained traction. Within a month of founding, working from a desk at Entrepreneur First's office in Bermondsey, the duo had signed a paid pilot with HS2 and persuaded Costain and Kier to share historical project data. Deutsche Bahn, Shell, and ExxonMobil followed. Meanwhile, the company attracted angel investors including DeepMind founder Sir Demis Hassabis.
Persuading the industry that their AI actually worked was tougher. Construction was - and largely still is - an industry that runs on human judgment and experience. "The idea that a machine could meaningfully forecast the outcome of a project that had defeated experienced delivery teams wasn't trusted - we got some version of 'an AI can't possibly solve this'," Amratia says. By the end of 2019 Amratia had logged more than 100,000 air miles in pursuit of new business.
What changed was the launch of ChatGPT, which triggered a wholesale re-evaluation of what AI could do, even in industries that had spent decades resisting digital change, like construction. "Suddenly we weren't having to convince people that AI was worth taking seriously," Amratia says. "The conversation shifted from 'that will never work' to 'what can it do for us?'"
Last October, nPlan raised a $16 million Series B, led by French investor CapHorn, with Chevron Technology Ventures and Suffolk Technologies. Existing backers GV, Pentech and LocalGlobe also participated. Headcount now stands at 55; revenue doubled last year and is on track to hit £15 million in 2026. Last year, the firm signed deals with customers in five sectors, from transport to manufacturing, across the world. Major corporates including Chevron, Anglian Water, Sizewell C and National Grid use its AI, which is currently deployed on $500 billion of live projects worldwide. nPlan's dataset – the moat that makes it hard to copy – includes some 750,000 project schedules, representing over $2 trillion of capital spend across 15 countries. "Two and a half million years of project experience," Mosca says.
If there's a lesson Amratia takes from nearly a decade of building nPlan, it's about what it actually takes to apply AI to a serious industry problem. "The companies best positioned to win are the ones with deep expertise in their segment," he says. "They understand the complexity. They have the data. They know the market inside out. There aren't shortcuts to that."
