Blog

Planning a building — projecting timelines, manpower needs and costs — is among the most complex and uncertain tasks in the construction industry.

A major development such as a skyscraper can take months to estimate and cost millions of dollars to design and engineer. Even at the bidding stage, engineering firms invest significant time and money, knowing they’re likely to win the work less than 10% of the time.

A New Haven technology startup says it can shrink that painstaking process from months to about seven minutes using artificial intelligence.

“LeanCon is building the first pre-construction engineering team created by artificial intelligence,” is the way that co-founder Ziv Levi pitches his product.

Investors have taken interest. The company recently announced a $6 million seed funding round, double its initial target.

Levi and his co-founder Sapir Tubul are both construction engineers by trade. They grew up in Israel and began working on construction sites with their engineer fathers. They later met as civil engineering undergraduates at Technion, the Israeli Institute of Technology.

After graduation, the two took different paths. Levi earned a master’s degree in civil engineering and later completed an MBA at Yale. Tubul earned a master’s degree in computer science, specializing in machine learning and artificial intelligence.

Levi said the idea for LeanCon grew out of his experience working as a construction engineer, where he dealt firsthand with the complexities of planning large projects.

“I had to make hundreds of complex decisions, crucial decisions for the project. The complexity in this process is to understand how everything would fit together,” he said.

That complexity also often results in estimates that are far off from reality — with most construction projects going over budget, falling behind schedule or both — and Levi said that’s not surprising.

“There is not actually a clear method for how to plan projects. As a construction professional, we rely mostly on our personal intuition and experience,” he said.

Size of the prize

LeanCon aims to move the process from intuition to data, and it’s developing software that uses AI to generate detailed projections on costs, schedules, logistics and construction methods.

Within the last 18 months, the company has been working with one of the country’s largest privately held construction and real estate development firms to continue refining its models in real-world situations.

The surprising thing, says one of their investors, is that it works at all.

“Artificial intelligence models in their current incarnation are really lousy at doing several things: spatial reasoning, logic and the complex interaction of those two things,” said Phil Bernstein, professor at the Yale School of Architecture and a former vice president at Autodesk, which makes 3D design and engineering software.

It’s not for lack of effort in the industry though. There are currently hundreds of companies attempting to implement AI in different aspects of construction engineering. One database lists more than 1,700 AI apps for the architecture, engineering and construction industry.

The reason so many are willing to try is the size of the prize. Construction is a massive global industry — often estimated at about $11 trillion annually — and it’s ripe for a data-based approach.

“Somewhere between 30% and 40% of all the money on construction is wasted because construction is very inefficient,” Bernstein said.

Yale skeptic becomes a backer

Bernstein first met Levi when he was an MBA student at Yale. Levi sought the professor’s opinion on his big idea.

“I told him I thought it was not going to work,” Bernstein said. “I told him I thought he was barking up the wrong tree.”

Levi was not put off though. He kept refining the idea and coming back for more feedback, until Bernstein finally had to admit he might have something. Bernstein contributed angel funding in LeanCon’s recent seed round.

Bernstein said he has been most impressed by Levi and Tubul’s focus and energy in tackling the problem, as well as their ability to bring in real engineering projects to test the software.

“The construction industry suffers from very poorly organized information that we don’t like to share with one another, and those are not good characteristics for training AIs,” Bernstein said. “So, you have to pick a limited problem, you have to understand what the logic is and you have to be able to train, train, train.”

“We’re collecting data all the time,” said Tubul, LeanCon’s co-founder who previously worked as a data scientist at Intel and a software engineer at an Israeli defense contractor.

In addition to the founders, the company now has eight employees. With the seed funding, the company aims to grow to 12 to 14 employees. But hiring is not straightforward.

Most current employees are software engineers with backgrounds in AI and data science. But they also need a strong understanding of the construction industry. Tubul said he recently interviewed about 100 candidates to fill two positions.

As they advance their product and become better known in the industry, Levi says the challenge is to stay focused. Their model is based on vertical construction of high-rise buildings. They were recently approached by a U.S. company building a resort in Mexico based on a series of small villas — but they had to turn the offer down.

“We need to prove that the business aspect of the company is there, it’s sustainable,” Levi said. “We are on the chase of product-market fit, creating technology that our clients really like. They will not remember the way of doing this process without LeanCon — that is the target.”

Levi says the company’s New Haven base has proven useful — not only with the support that Yale has been able to provide, but also for other connections. New Haven-based Newman Architects is now an investor, as is Connecticut Innovations, the state’s quasi-public venture fund.

Both joined the seed round, which was led by Denver-based Ibex Investors.

“We definitely owe a lot to this place,” Levi said.