Artificial intelligence and machine learning have well-known applications in modern machinery, but they have other uses as well. Urbint, a New York-based startup, applies AI to some of the oldest urban infrastructure there is — gas distribution lines. The company’s software enables utilities to prevent damage and make their pipe inspections more efficient.
The information combines data from building records to the latest weather information, said Corey Capasso, founder and CEO. Then it presents a plan that allows the utilities to improve the use of their resources as they manage sprawling networks built over decades.
“There are all these factors out there in the real world, and it’s immensely hard to understand them at scale,” Capasso said in an interview with BloombergNEF in early September. “One of our core competencies is creating that digital representation of a utility’s service territory where we’re understanding all those external factors that impact their assets. We call it decision intelligence, and it’s incorporated into the utility’s work flow.”
Q: How did Urbint get started?
A: We started the company a couple of years ago. My background over the past decade has been building companies in the enterprise software and data space. We’ve organically stumbled into what Urbint is; our mission is enabling utility companies and infrastructure operators to make cities safer and more resilient.
Q: How did you fall into that?
A: The idea started with the hypothesis that if we applied artificial intelligence to real-world data that we could find patterns and then make predictions on certain things, tied to infrastructure becoming distressed.
The idea was that AI could allow us to find patterns that weren’t obvious or realize that a sequence of events could lead to something you could predict. We wound up having a meeting with a very large utility company, and the discussion was that inside every building there’s a gas pipe, and sometimes the gas pipes leak or have issues. Gas pipes also go under the streets. It expanded outside the building to all infrastructure.
The world is changing quite drastically. We’re seeing unprecedented weather patterns, and there’s a residual impact of changing weather. We’re also seeing rapid urbanization in some areas, which makes general operations more complex.
Those two things, changing weather and rapid urbanization, are being exacerbated by an already risky situation of aging infrastructure and a maturing utility workforce. The reason we decided to pursue Urbint is that society is so reliant on critical infrastructure.
Q: That’s a long list of items to be evaluated through AI. How do you start on that, and how do you weight the factors you incorporate?
A: Our core customers today are gas utilities, and we’re focused on gas distribution. That’s where roughly 70% of all gas-related incidents happen, and those incidents are associated with 80% of fatalities.
One of the biggest challenges is aging infrastructure. Utilities are tasked with balancing safety, affordability and reliability. Due to lack of visibility into the real world where their assets are, they haven’t been able to quantify the tradeoffs. By leveraging artificial intelligence and a digital model of their service territories, we enable them to understand those tradeoffs better and make better decisions.
Q: What’s an example of that?
A: Whenever someone wants to dig in a service territory, they’re supposed to call 811 and say where they’re going to dig. The utility marks the line, and then excavators dig. Many times they hit the gas pipes anyway, and utilities are required to report to the Pipeline and Hazardous Materials Safety Administration what their third-party damage metrics are.
We create software solutions that predict which dig tickets are more likely to lead to third-party damages. That information lets the utilities take different types of interventions to reduce the number of times pipelines get hit.
Q: What does the AI look for?
A: It’s a combination of things. Who are the contractors, what’s the physical environment, is it on a slope, what are the soil conditions, what type of assets are involved. It’s really about understanding the sequence or combination of things that lead to an incident. For instance, a low- to medium-risk excavator may be a higher risk in a certain type of environment. The AI shrinks the list of tickets down to those most susceptible to damages. That’s where you apply your resources.
Q: What does a utility do with all this information?
A: We’re processing all these different data sets and it’s dynamic, changing in real time. The best example of that is weather. We’re not giving them an analysis of data, we’re actually telling them what decisions to make. These decisions are tied to programs.
Let me give you an example: methane leaks. Every utility has a leak survey program, by regulation. They have to survey all their pipes. They’ll start in one area and then go to another, so they can get to all of them in three years.
If Urbint can predict which lines are more likely to have a leak, the utilities can prioritize their inspection schedules to uncover higher-risk assets faster. The utilities are already inspecting properties. It’s just about figuring out which properties to inspect first. We call it decision intelligence, and it’s incorporated into the utility’s workflow.
Q: Who are some of your customers?
A: Urbint works with the majority of the top 25 gas utilities in the U.S.
Q: How do you measure success?
A: Every utility tracks damages per 1,000 tickets, so they have a baseline. By using Urbint, they will see a reduction in their damages per 1,000 tickets or find more leaks per 100 inspections, and they can actually quantify that reduction. With all of our clients, we go through a validation period where we actually validate the software. It allows them to understand what’s the incremental improvement that they get from our solutions.
Q: How is the data presented?
A: Our solution will tell you what decisions to make based on your resources. We say we enable utilities to make the right intervention at the right time with the resources you have, and that’s key to what we do.
Q: What’s an example of that?
A: One of our solutions is odor-call predictions. When customers call in to say they smell gas, utilities are measured by how many minutes it takes from the time the call comes in till the utility shows up. But call volume is inconsistent and affected by multiple factors, and response time is based on how you’re staffed that day and where your people are in the field. Our solution predicts how many calls are expected that day and where your people should be.
Q: AI for predictive maintenance takes current data — vibration, for instance — and connects that to breakdowns that occurred in other similar machines to make predictions. How important is historical data in Urbint’s calculations?
A: We’re seeing unprecedented changes in weather and urbanization, so historical data is only one factor, and there could be hundreds of factors. Maybe we see a certain type of vibration, but a certain type of vibration with a certain type of soil with a certain type of gas pipe is that much more risky than with a different combination. Perhaps the pipe is co-located with a sewer pipe that is weakening the soil.
Q: You’re not going out and putting sensors on gas pipes?
A: If a utility has sensors and wants to pass us that data, we can, but we’re not focused on sensor-based data. It’s not so easy to install sensors on an underground gas pipe.
Q: Do you have a favorite success story?
A: Southern Co. Gas is a top three gas utility in the country and they’ve used our damage-prevention product to reduce their third-party damages. They’ve spoken about it publicly. They got a safety award from the American Gas Association. Our AI was able to capture more than half of their damages in less than 5% of their dig tickets.
Q: You don’t have a Hollywood-style story of saving the day at the last minute?
A: I like to avoid the Hollywood blowouts. A lot of people position AI as helping you find the needle in the haystack. We believe a more effective approach is you’re making decisions every single day. If you’re going to leverage AI every single day, you’re accumulating risk reduction. You’re not finding a needle in a haystack, you’re just lowering risk every single day on things you’re already doing.
Q: Utilities tend to be heavily regulated. How does that affect their relationship with you?
A: A lot of regulation is pushing for more safety, and sometimes it asks utilities to do more with the same resources. Urbint evangelizes the use of technology for safety at the regulatory level, but what we’re really doing is empowering our utility customers to have a different conversation than they were having before. They can say, “Here’s our budget and here’s the most effective way we can spend it, and here’s what we can accomplish with a larger budget.”
Q: How far along are you toward getting into other areas, like electric and water and telecom?
A: That’s our longer-term goal. We do have some solutions in the electric vertical that we haven’t publicly announced. We are validating those with a select few public utilities, and then we’ll be releasing those next year.