May 2, 2020
Clear Roads, Green Lights, Can't Lose: Our Origin Story
This our origin story: how we decided to build a technology that could give back time to everyone, how we put it on a major city roadway and reduced travel times, how we showed that we could give people up to two and a half days back each year.

Solving a critical traffic problem

Traffic can be immensely frustrating, whether it’s how long it takes to drive to and from work every day; or getting delayed by a sudden traffic jam on the way to meeting up with friends; or hitting red light after red light after red light on an empty road. You lose a lot of your time to traffic — about 10 days each year. That’s 10 days each year you could’ve been out enjoying life. But instead, you’re sitting in a car, on a road, just waiting for the journey to be over.

In 2017 alone, traffic congestion cost the US economy $305bn and the global economy over $1tn. It is a problem which impacts everyone across world — limiting economic opportunities, damaging the environment, causing individual frustration and unhappiness — and it is only going to get worse. Smart signals can alleviate the problem, reducing travel times by up to 25% and reducing vehicle emissions by up to 22%.

Traffic signals and timing plans

Traffic signals are the biggest lever by which a city can manage and control the flow of vehicles: they help manage vehicle volumes, regulate speeds, and prevent accidents. All of this is done by changing the timings of signals (i.e. how the green lights are sequenced, how long they run for).

In a given road network, for a given traffic pattern, there is an optimal signal timing. As traffic patterns change, the signal timings should change with them. How these timings change during the course of the day and week are determined by a signal timing plan.

Traffic signal timing — a difficult and costly problem to solve

Getting traffic signal timing plans correct is an incredibly complex problem which involves balancing multiple competing objectives including:
  • Network efficiency (i.e. how to get people from A to B quicker)
  • Road-user experience (i.e. how to reduce the number of road-user complaints)
  • Road-user safety (i.e. how to avoid collisions at intersections)
Whilst trying to do it for large numbers of intersections, whilst traffic patterns constantly change, using limited tools and only partial datasets. It  is like shooting multiple moving targets, while wearing a blindfold, with a gun that only shoots blanks.

And then this has to be done cost-and-time efficiently. Today, the best way to do it is manually which can take up to 70 hours per intersection and requires a small army of people including Traffic Signal Engineers, Traffic Signal Technicians, and Signal Timing Consultants .
It often costs thousands of dollars and weeks, if not months, to update timing plans for even a small road network. Even then, there isn’t any guarantee that the signal timing plans are as good as they could be. Given the time and financial cost to conduct, signal timing plans can rarely be updated more frequently than every 3 to 5 years. Yes, you read that correctly, every 3 to 5 years. Think about how much the world has changed in the last 5 years. You could’ve gotten married, had kids, and those kids could already have finished pre-school. In that time, we could’ve upgraded our iPhones four times and we’ve even sent software updates to cars that enable them to drive themselves. And yet there’s a good chance that in that time, nobody has updated your traffic signals. As traffic increases and the number of qualified traffic engineers in the US declines, you’re only going to be waiting longer.

The impact of intelligent transportation systems

Solving this problem could make the world look very different. According to the Federal Highways Administration (FHWA), with better signal timing we’d see reductions in travel time up to 25% and reductions in vehicle emissions up to 22%. That means two and a half days each year that you get back to spend however you want. That means cutting down carbon emissions for the whole of the US by 5%! It’s a world where roads are clearer, the air is cleaner and people are much much happier — and smart signals can get us there.

How smart are “smart signals” today?

At Flow Labs, when we talk about using AI to update traffic signal timings, the most common question we hear is “Don’t they already do that automatically like in ‘The Italian Job’?”. Our answer is (1) that was a terrible reboot of a British classic and (2) no, the vast majority of signals are retimed manually.
The second most common question we get asked is “Why don’t you just make them smart by putting lots of sensors at the intersection and letting the hardware retime the lights automatically?”. The short answer is that it’s being done already, and it doesn’t work that well. These real-time systems are commonly known as Adaptive Traffic Control Systems (ATCS) — they are what people first think of when they hear the term “smart signals”. Despite the technology being available for over 30 years, of the 350,000+ intersections in the US today, ATCSs have been installed at less than 3% of them.
With up to $230k per intersection deployment costs and additional overhead to maintain, ATCSs are an expensive investment for the average transport agency. Where agencies do buy them, they’re applicable to less than 20% of signals. They struggle in city grid networks, multi-modal environments, and highly congested conditions — basically everywhere that really matters. Even once installed, performance studies of ATCS have shown mixed results and in practice, many of the systems that are installed are subsequently switched off because of erratic, unpredictable behavior or poor performance.
“It’s been an all or nothing thing, you can’t turn the switch on and go adaptive, find out that it doesn’t work and go back to your traditional actuated, normal-time-of-day operation. You put yourself out on a limb.”
That has changed, however. Little added he’s been interested in bringing technology to Ada county Idaho since 2004, but waited until now to make sure there was enough variety in the marketplace. "The ability to shut off adaptive control is one of the features the county will place a high priority on when deciding which system to select." If the most demanded feature for your product is the ability to turn it off, you’ve messed up somewhere.

If you had to employ someone and they tell you that they can only do 20% of the jobs that you want them to do — and not the 20% you really, really care about — and then when they actually do a job, they do it well some days, they do it badly other days, and some days they act so crazily that you simply have to send them home — would you really say they were smart? More importantly, would you hire them? Would you pay them a large sum of money up-front? That is the dilemma for transport agencies considering ATCS. For the vast majority of agencies today, it is still far more effective to use large teams of people to retime signals.
Insanity: doing the same thing over and over again and expecting different results"
Albert Einstein
Smart signals are a great solution to congestion in major cities, but all of the latest solutions being developed fall into the same traps as those of the past. Smart signals of the future need to be better: they need to be cost-effective; they need to be applicable to signals in any type of road environment; and they need to provide measurable and predictable performance improvements. And that’s what we’re building at Flow Labs.

Building the Future of Intelligent Transportation Systems

In 2019 we developed the first version of our smart signal technology, AI Traffic Control Software (AI-TCS), and we decided to see how well it would do in the real world.

Going up against the best

Utah Department of Transportation (UDOT) is best-in-class in the US when it comes to traffic signal operations — whether it is their hiring of world-class talent, innovative business processes, leadership in signal performance analytics, or use of the latest and greatest technologies (including ATCS) — but even they still struggle to keep up with maintaining their signal timings. At Flow Labs, we’re never content with taking the easy road, so we decided to put our technology up against the best.

On a major corridor

We took control of the US-89 corridor — a major commuter corridor north of Salt Lake City serving up to 40,000 vehicles per day, half of which come during the peak hours. This is an efficiently operated corridor: it is managed by a Statewide Signal Engineer and a team of up to 10 signal timing consultants and it was only recently re-timed. We were tasked with updating their signal timings during their peak hours, the most congested times of the day. significantly improved traffic flow

Over a 2 week period, our AI was able to achieve substantial improvements in peak time traffic flow:
  • Fewer stops at red lights: Think about the satisfaction you feel when you hit multiple green lights in a row. Our technology increased arrivals on green from 71% to 75%. For road users this means that you’re hitting 14% fewer red lights, and you can ride the “green wave” more often.
  • Faster queue clearance: Ever felt the frustration of being at the back of a long queue waiting for a green light to come, and then when it finally comes, it turns red again before you can get through? This is known as a split failure — and we reduced these by up to 37%. In highly congested roadways, queues are inevitable, but with our technology, road users won’t be waiting in them for long.
  • Significantly lower travel times: Most importantly, we saw reductions in travel time of up to 24%. If you had that across your entire journey, that would be 2.4 days you’d get back each year! In congested environments, it’s difficult to hit free-flow speeds (i.e. the speed limit) but our technology enables road-users to get closer to it than ever. As we continue to develop our technology we’re looking to close that gap even more.
All of this amounts to over $1.4m in benefits to the Utah economy each year in travel time savings alone for just one 7-mile stretch of road, not to mention additional benefits from reductions in fuel consumption and reductions in vehicle emissions. All this with just a little bit of software, and we haven’t yet gotten close to its full potential. Like we said, UDOT is best-in-class in the US — our technology can achieve much more for cities which don’t have UDOT’s capabilities or resources.

So what could this mean for you and your city?

Our technology was able to achieve significant travel time reductions, at peak time, on an already efficiently operated corridor. For you this means clearer roads, cleaner air and shorter journey times.

Our technology doesn’t require any new hardware or infrastructure, it can be set-up within minutes on the whole corridor and it achieves best-in-class performance at dramatically lower cost than existing smart signal solutions. For cities this means that it can be deployed across the entire city in months, not years, and won’t require significant costs to taxpayers. As a citizen this means you get to experience all the benefits, as soon as possible, without having to pay anything extra for it.

We’re excited to live in a world where people aren’t waiting at red lights, where city roads are clear of congestion, where people can spend more time at their destination than on their journeys and we hope you are too.

We’re writing the next chapter of our story — want to be a part of it?

Anything worth doing is worth overdoing
Mick Jagger
Not content with just one 7-mile stretch of road in one city, we want to make a much bigger impact — we want to put our AI Traffic Control Software (AI-TCS) in every major city in the US within the next 5 years and we’re building a team that will get us there.

If you want to be a part of it, get in touch at
Flow Labs technologies like Predictive Traffic Control (PTC) enable you to spend less, deploy faster and see results quicker. Contact us to schedule a demo today.
Thank you!
Your submission has been received!
Oops! Something went wrong while submitting the form. Please try again.