Bentley goes Blyncsy for AI
(Sorry – I’m truly terrible at headlines.)
Do you remember // did you listen to Bentley’s Q4 earnings call? CEO Greg Bentley was very excited about the potential of the company’s asset analytics offerings, specifically its 2023 acquisition of Blyncsy. You may remember that Blyncsy crowdsources images of roads, bridges, guardrails, lane markers, and other roadway elements and applies computer vision and artificial intelligence to identify maintenance issues.
Mr. Bentley said that Blyncsy “monetizes a mile of roadway, and what we charge depends upon how frequently the AI is run on fresh imagery” and that “Blyncsy is now also pursuing procurements that can exceed seven figures of ARR.”
That’s a lot of ARR that will eventually turn into a lot of revenue, so I asked Bentley’s PR team to connect me to someone who could tell me more, and last week, I spoke with Mark Pittman, Director of Transportation AI at Bentley Systems —and, not coincidentally, founder of Blyncsy.
While researching Blyncsy, I discovered that it works with some very cool customers. For example, the City of New York used Blyncsy tech to discover which crosswalk pavement markings need repainting. The City manages thousands of these, and it’s a bigger deal than I had imagined: I pictured a truck with a schedule: “It’s April, so we must be on 42nd Street”. Nope. Since last year, NYC’s DoT has been using Blyncsy to run real-time, AI-powered crosswalk detection and paint line scoring, which they then use to prioritize work. They’re also using this data to develop a historical model to track changes.
Mr. Pittman told me that Blyncsy works with a “fleet of over 800,000 vehicles [of course, not all in New York City]. These are long-haul trucking drivers and last-mile delivery vehicles—all of which have cameras on them, and we can purchase that imagery.”
Back to NYC. The crosswalks are painted with water-soluble paint (per EPA rules) and degrade quickly — but not evenly, since more traffic means faster degradation, intersecting crosswalks may have thicker paint, and so on. Early results are promising, and the City is adding additional cameras to city vehicles to capture more data.
Another example is highway guard rails. Many Departments of Transportation put an inspector in a truck and have them drive the roads, looking for issues and scoring them according to urgency. That’s not cheap: it requires someone with expertise to do the scoring, a truck, and gas—and they can’t be everywhere all at once.
Mr. Pittman added, “That’s the core of the technology value proposition from our perspective: every single day, there are fewer government employees to do the work, and the system is getting larger and more complex. We’re looking at things that are safety or operationally critical.” In other words, the DoT has to do this; there’s no choice, so it’s time to look at viable alternatives.
And now we get to the revenue potential. If it costs hundreds of dollars per day or section of roadway to identify potential problems (and keep in mind, there may be none — and that’s important to know, too), what’s a cheaper alternative?
Mr.Pittman: Our “cost model depends on scale. Customers might be buying millions of miles of data from us that will [be at a] lower cost [per mile], but a general range is that it’s $10 per mile to inspect the road. We compare this to human inspection, typically somewhere between $80 and $100 to $200 per centerline mile* to send someone out. LIDAR inspections are typically between $200 and $400 per centerline mile”. Bentley typically tells prospects that “we can inspect all of your roads every single week for the same cost as one LIDAR inspection and give you change detection on top of that right.” From a cost perspective, image gathering / AI processing is an economical alternative to human inspection. And it’s likely more accurate, too, since images are constantly sourced and don’t depend on the driver’s schedule.
But it gets better, if you’re Blyncsy. Mr. Pittman told me that a new federal highway guideline will require annual inspections of road striping on all roadways of 35 miles per hour or higher starting in 2026. Doesn’t that already happen? Nope. States today typically inspect just 1% to 3% of their paint lines yearly to get a baseline and then extrapolate from that for the rest of their system. The Feds felt this wasn’t good enough, and so now, everyone at all levels is trying to figure out how to identify the quality of their paint lines and (likely) come up with a plan to fix it.”
Add in the pressures put on road systems by the makers and drivers of sensor-enabled and self-driving cars (for whom lane markers are a big deal), and we can see Blyncsy’s potential in just this one application area.
Image processing and AI fit well into Bentley’s overall strategy for roadway infrastructure clients. Blyncsy contributes data about what might need fixing, a human identifies how urgent that is, and then Bentley’s engineering (OpenRoads, for example) can begin serving up details of the roadway itself. The very definition of engineering and operations working together.
* I also didn’t know what a centerine mile is. It’s the unit of measure the government uses to determine the total length of a roadway section. It ignores how many lanes the road may have, so it isn’t the same as the road’s carrying capacity, for example.
The title image is from Blyncsy, with thanks.
