On a frigid winter morning in 2004, I stood on the tarmac at Kabul International Airport in Afghanistan. The sun was rising, and my crews were already getting to work repairing a giant hole in one of the runways blown by a rocket-propelled grenade. I had one thought in my mind: How can I fix this runway faster? At that time, the airport was the artery for critical aid, supplies and transportation. Overland convoys could take weeks and were subject to constant attacks. My blown runway was holding up nearly 30% of all airport traffic. The entire country — the tens of millions of civilians, the U.S. and NATO militaries, the aid workers — all depended on my work.
I grew up in the construction business, the son of a civil engineer. My first job was as a draftsman. I later managed projects large and small, from Beirut to Dubai to Afghanistan to Athens. On every job, I had heard the same questions. How can we build faster? What can we do to speed up the job? Why are my crews waiting?
That morning in Afghanistan was the first time I vowed to do something about this problem. That vow led me to found Alice Technologies.
Construction projects are intricate dances requiring devilishly complex sequences of tasks, many of which are entirely or partially dependent on one another. A simple home improvement project might require 100 different tasks. A complex construction project — an office building, a manufacturing plant, a bridge — might require 6,000. In building a new addition to a house, you can’t put up the roof before the studs are placed and framed. In building a bridge, you can’t hang the cables before the towers are built and you can’t build the towers before the pilings are sunk.
For this reason, the construction industry has relied on “schedulers”, a guild of savants who take 3D project design plans and reduce them to Gantt Charts and a sequence of steps. A good scheduling company might need two months to create a master schedule for a large project. From that schedule, the general contractor will plan all activities — when material shipments arrive at the job site, the sequence of sub-contractors when welders and electricians show up.
Yet all construction projects can be broken down into finite patterns and “recipes” of activities. Those activities must take place in a well-defined three-dimensional grid — the job site. From my math studies, I knew that computers can manipulate patterns to optimize for outcomes — speed or cost, for example. Optimization is something the construction industry has sorely lacked. According to a 2017 report by McKinsey & Company, the construction industry has consistently lagged other sectors of the economy in productivity growth. Out of an annual global spend of $10 trillion, McKinsey estimated there might be room for $1.6 trillion in savings. I intuitively knew this. Heck, everyone in construction knows this. But for me, it wasn’t academic. It was personal. I wanted to repair that runway faster.
I went to Stanford to get a Ph.D. in Construction Management, and to learn more about the math and the planning. But I wanted to build something real and useful. I alternated between work and studies at six-month intervals, which allowed me to develop the concept for ALICE. At Stanford, I gained better insights into how modern computer technology and artificial intelligence could be applied to the construction industry.
My second “Aha!” moment came in the Netherlands in 2009. Two Dutch guys were screaming at each other. If you know the Dutch, you know this is rare. Dutch people don’t usually display public anger. The source of their conflict? A construction project delay. Structural steel for the project was going to arrive six weeks late. Each day of delay would cost 50,000 Euros. “My guys can’t go any faster,” one of the screaming Dutch construction executives said.
I looked out the window at the vast job site. It was mostly empty except for six workers. I thought, “How much space is wasted on the average construction site?” I flashed back in my mind through the many jobs I had managed and worked. I replayed a mental reel of empty spaces everywhere. At that moment, I realized that if we could do a better job utilizing the space, and planning activities to fit in the space in parallel, we would finish jobs much faster.
Moving around objects in three-dimensional space was something that computers had been doing for a long time. If I could figure out a better way to use this space, to sequence these patterns, I could automate and radically improve construction project planning. I took out my camera and snapped a picture of the stalled Dutch site. I took that picture back to Stanford as a guide for our first prototype. As soon as it was done, I returned to the Netherlands to show them what I had built.
It was simple. I input a bunch of tasks as data, including time required, space required, and dependencies on previous and subsequent tasks. The algorithm took the tasks and ran simulations. I had to create the datasets by hand to train the machine learning software because the construction industry does not generate reliable structured data on job site tasks and sequences.
The algorithm had a basic front-end showing colored boxes moving around in a two-dimensional space to represent the sequence of tasks. Those same screaming Dutchmen said “Yes, this solves our problem. Show us future versions.” Over the next four years, I won a fellowship and hired assistants to improve the prototype. We entered ALICE in a prestigious startup competition at Stanford and beat out more than 100 other companies. A lawyer offered to incorporate the company for free. I woke up in a dorm room as the CEO of that company. The venture capitalists funded us immediately. And the rest is history.
But not really. Today ALICE Technologies is working with some of the largest construction companies in the world to help them build faster and more cheaply. We integrate directly with leading Building Information Management systems and generate sophisticated 4D (with time as the fourth dimension) planning routines and recipes that make it drop-dead simple to optimize the most complex projects. As a customer of ours who’s a 35-year veteran scheduler likes to say, anyone can become an ALICE expert in half a day.
We found the sweet spot of collaboration between man and machine. Leveraging the knowledge of a project scheduler, ALICE can calculate in less than 10 minutes roughly 600 million simulations of sequences of the thousands of discrete tasks needed to complete. A human being would need 13.5 years to complete the same calculations, assuming they worked non-stop and at perfect efficiency. ALICE can run simulations in ways in ways no human could such as: determining what happens when tasks are run in parallel; what happens when additional resources — cranes and crews, for example — are added; or how costs might change if different materials are used. Like Google’s latest AlphaGo machine learning system, ALICE learns from scratch without needing massive data inputs. This helps our customers optimize their schedules in a matter of days. You can see what today’s version of ALICE is capable of in the video below.