Editor’s note: In this interview with FMA sister publication EHSLeaders, Cooper Briscoe, VP of marketing at Brivo, discusses how AI-powered video intelligence helps organizations get increased value out of existing camera systems.
(Edited for clarity)
Q: Why hasn’t the prevalence of cameras resulted in improved safety?
Briscoe: I think you have to look back at what cameras have mainly been intended to do over the course of the last few decades. As camera adoption has taken off, for example, I’m not talking to a lot of facilities managers today saying, “Are you considering adding cameras to a facility?” Most of them are prevalent, they exist, they’re out there today. But what’s really changing now is traditional camera security systems have been reactive. They have been there, nobody’s watching 24-7, those cameras, and they’re kind of passive. They’re sitting in the background, and then when something occurs, you want to go find out when did that thing occur, and who did the thing, right?
So what’s really changed or is changing is the ability to turn them into being a little more proactive with what they do. Cameras have always served the purpose on premise, which is, we need to find out what happened. It could be risk mitigation, it could be to avoid lawsuits, for example, and litigation, but it’s traditionally been “How do we react after something has occurred?” That’s the thing where you could potentially say cameras have been less effective in the past, and why they’re changing going forward.
Q: How can AI video agents help organizations spot hazards and risks?
Briscoe: Let me start with a little bit of the evolution of what’s started to happen as AI has become more introduced into the world of video. It started with—again, thinking retroactively—“I need to go find something.” The first improvement that we did here at Brivo is actually roll out something called Smart Video Search.
So if you’re a facilities manager and you’ve had to deal with something where you needed to look through video and something happened weeks ago, it’s a terrible experience. It takes hours and hours. It could take your entire day just to go back in time and look for video. Well, Smart Video Search changed the game a little bit and said, as opposed to going back in time and just rewinding and watching the video, why can’t we ignore down periods? If nothing’s happening on camera, simply ignore that stuff.
So we started to give facilities managers, supervisors, etc., the ability to more quickly find the things that mattered to them. Now I don’t have to scroll through and scrub that video anymore, I can quickly go, “When there was activity, let’s go find it.”
If that activity centered on a particular individual, let’s go find other examples of that individual. So that was maybe kind of step one in some of the evolution of this. For a long time, we had things that we could say were pixel-based motion. For example, a camera’s looking at a scene and if that scene changes, we want to know about it. Well, that’s great, except for if it’s an outdoor camera, and the leaves blow in the distance, or a cat or a raccoon goes by in the middle of the night. So that wasn’t a massive help.
Now we’re stepping past the retroactive forensic side of things, and saying, “What can AI do to actually monitor scenes, intelligently look at something, and then alert me when something occurs?”
Just to give you an example or two, Brivo’s launched a product called Eeva and that is the introduction of a video agent inside of the video management system. So now what facilities managers have the ability to do is say, “Let’s not let something happen and then go find out why it happened [and] who was involved in the past, let’s actually see what we can do to proactively alert us if a situation is going to occur. For example, I have a door, and you can’t prop things up by a door, so I don’t want boxes stacked by a door, it’s a safety hazard. Well, guess what? Now I can have a camera that is monitoring that scene at all times, and I could literally write in a natural language prompt that says, “Hey, if there’s ever anything blocking the door, I want to be alerted on it.”
Now, you can customize that and say if there’s something blocking the door for more than five minutes, more than 10 minutes, you can make that a thing to where as soon as it occurs, you get an alert and take action. You can also make it a little more passive to where you’re monitoring that situation, and then you get a report the next day of all the activity. So that’s one example that we’re seeing.
But where we are today is we’re at a point where the opportunities and the examples are limitless. So I can now take a camera, and I can say, “I want that camera pointed at this particular scene, because I want it to do X or Y or Z.”
The conversations are completely different today than they were a few years ago. So again, going from reactive to proactive, now I can have a conversation with an organization, and I can say, “What are some of the biggest problems?”
Well, maybe it’s that a forklift entered a restricted area. That would happen in the past, and maybe nobody found out about it unless they watched the video. Now I can just simply say, “If a forklift enters a restricted area, send me a notification immediately.” So I can take action upon that notification.
Or I can actually tie and trigger additional things to take place without my involvement. For example, if in that scenario a forklift entered a restricted area, yes, I could get a notification or an email on my phone, that’s great. But I can also trigger a warning, a siren. I can have lights flash, I can have audio talk down to say, “Hey, you’re in a restricted area, this is your first warning. Additional infractions are going to be met with escalating responses,” that type of thing.
So those are just a few examples of really what’s changing, but I think the summary is we’re going from looking at cameras being a security system where we react to things that occur to having those cameras actually being able to drive operational efficiency, being able to address things that could be problematic before they occur, and that’s really the benefit and the value of the system today, with AI introduced into these cameras.
I can search for people that have no hard hat on and get notifications. I can run reports on those if they’re not wearing a safety vest. We blocked a fire lane. All of those things now are at our fingertips where I can use natural language to type in a prompt and say, “This is what matters to me, and I want to be alerted or take this action when it occurs.” And that’s a fundamental change to the entire industry of security professionals. Really big deal.
Q: How does the video agent work with existing cameras and prompts?
Briscoe: At Brivo, we’re an open system, so the first thing with anybody that I talk to is not, “I want to go sell you cameras.”
One of the benefits of our solution is we’re building large language models in the cloud. These large language models can be hit by any camera out there. So since Brivo is an open company, we’ll work with thousands of camera manufacturers out there. And you don’t have to have the smarts and the intelligence in those cameras. So what does that mean to me as a facilities manager? It means I don’t need to come up with budget to do a big installation, I don’t need to replace all of the cameras, and I don’t need to worry about those cameras being out of date within six months or a year because of the pace of technology.
The benefit of us using agents to hit these large language models in the cloud is that it basically brings customization to anyone out there without them having to actually pay for a development project. And so it’s limitless today, using existing cameras, hitting the cloud, and the added benefit is that every individual end customer that is taking advantage of this is helping us train the models, helping us improve, and giving us more examples of opportunities where we actually solved a real-world pain point.
And that’s the thing that’s most exciting. Every week, there is a new example of an end customer that decided to try something, whether that was monitoring inventory levels on a shelf or monitoring whether kids were at a pool after hours. You go across the different industries, there’s an example for every single problem that exists out there today.
Q: Is there any limit to the number of prompts you can enter for things to watch out for?
Briscoe: No, you’re unlimited in terms of the number of prompts, and actually, one of the things that we’ve enhanced since we initially launched this was some people came to us and said, “I have one camera monitoring an area, but I actually want to set up different prompts on that one camera in the one area.” So, for example, I might care if someone has entered a restricted area, but I also want to know if a forklift drove by, or if a box was in that area. So slightly different prompts. We can now add multiple agents to the same camera to be able to monitor different aspects of a scene from that one camera.
Q: How long has this product been available?
Briscoe: Eeva as a solution was launched earlier this year, so we’ve been testing it for six-plus months prior, but it launched formally and officially at the ISC West show this year, so that was end of March, early April.
Q: What kind of feedback have you received from users?
Briscoe: It’s been tremendous. It’s very simple to use, but a little bit of understanding how to write the prompt is helpful. So people started to say, “I want to know about this” and we’ve slowly trained them to say, “Eeva, alert me when this occurs.” So that’s some of the biggest feedback.
But honestly. I think it’s the education of what’s possible today. So it’s twofold. One, the education of what’s possible, and two, dispelling myths around AI.
Let me start with that one first. A lot of people hear AI in a camera, and immediately, this is, “Oh my gosh, the government is tracking me, my data and information is getting to ICE, it’s getting to local law enforcement.” When you start to tie this to some of what we’re seeing, with these roadway cameras, with other companies that are out there, and a lot of this public pushback, it’s because you’re able to identify, in those scenarios, personally identifiable information. Tying a license plate to a human being is possible, and then when you don’t restrict access to who has the ability to see the system, it’s a really big problem.
I’ll give you this example: A university implemented a bunch of cameras around campus. They touted some AI capabilities in those cameras. Well, the response from the school community, parents, etc., was chaos, confusion. “Oh, this is terrible, they’re monitoring our children, they’re watching them everywhere they go, and anyone at any point in time can now find my child on campus.”
Not at all what was implemented. They may have implemented something that was utilizing AI, but it was really for more intelligent scene analysis, not for personal identification.
So number one is there’s a little bit of a fear out there when you start combining video and AI. That’s something [where] people jump to a conclusion. The first part of that is, it’s really the education of the end customer, and frankly, our channel of resellers—the partners that we work through—that work with end customers. [They] are used to being traditional security integrators, where they’re making a solution work, they’re used to those cameras being reactive, and we’re really trying to pull some of our channel forward to say, “Hey, the conversations you’re having today with end customers are completely different than you’ve ever had them before.”
You can now walk into any scenario, and not say, “What are the security concerns you have?” But you can say, “What are the biggest pain points you have operationally, and if you were to be able to go home and fall asleep, and someone else was looking for the things and taking action that you would have cared about had you been working? Can we implement that?”
And when you reframe it that way, and have a conversation with an end customer, and they can start asking questions, “Well, what about this? What about that?”
The possibilities become limitless, and it completely changes what we’re doing. Now we’re no longer trying to say, “If something occurs, wouldn’t you be happy to know that you had a camera system and could have seen or found out what happened?” To now saying, “What are the things that occur that bring pain? And let’s go talk through those different scenarios and see how this may or may not be able to help.
I’ll give you another example, and it’s a different solution, not called Eeva, but we have a face match solution. You notice I did not say facial recognition.
There is no international database. There is no government database. But if I’m running a facility, for example, and the global president of the company were going to arrive at my facility, I would have the ability to type in, or to put in a VIP from my company that says, “When this person arrives, we want to make sure that everybody’s ready, the factory tour is on time, we’ve got a welcome committee,” and so you can get a notification when that person were to arrive.
Now, the difference here between these other scary models is that is you as a video management system administrator. So you have one location. That one location matters to you. You have the ability to say, “This person has trespassed in the past, I have a video image of them. If that person ever arrives back on the property, I want to know about it.” So think about person of interest, and what AI is doing to help us there. I want to identify a VIP and give them some special treatment, or I want to identify someone that is a previous offender. We have a lot of interest in the world of retail, as you can imagine.
If somebody’s been caught in the past, they’ve trespassed, they’ve stolen from you, that’s you at your store adding that person in to say, “If this person ever comes on property, I want to know about it, because they’re a known thief.” Very different world than what people’s perceptions take them to, which is “The government’s always looking and anybody could find me anywhere I go now.”
Q: Are there similar products to Eeva that are out right now or are you in a unique space?
Briscoe: We have been first in class on this one. You are seeing people start to adopt large language models, and utilizing that, and you will see more, but we were first to market on this one. And it is still something that requires a tremendous amount of education, understanding, and I think is going to be that way for probably the next couple of years. It is one of the biggest fundamental shifts or chasm crossings that we’ve done in the industry in a tremendous amount of time.
The more we can talk this through, the more it opens up people’s eyes to what’s possible. So when we’re talking to HOAs, for example, they care about illegal dumping. They care about a kid in a pool after hours. You know, they care about catalytic converter theft in their parking lot.
We now have capabilities using AI to say, “I don’t want to look at pixel-based motion in a parking lot that has high activity.” If a car goes through, like at a car dealership, ignore a car driving by, because cars drive by. The minute someone steps out of that car, I’ve now flagged it that this is a potential threat. So I ignored a pixel-based motion, I ignored the cat going by, I ignored cars running by, but now a person’s come out of a car in a parking lot full of cars after hours.
That’s a threat and I want to know about it. I can have that go to a professional monitored central station, or I could do it in a proprietary manner on my own. But what I’ve really done is I’ve said, I used to pay, for example, $8,000 a month for a guard that was, no offense, ineffective at times, or mostly walked a perimeter, wasn’t necessarily on top of it. And now I’ve replaced that with a camera that exists that’s just simply looking for that one thing.
We have a customer in the Middle East. They have a bunch of shawarma shops and the cone angle impacts the flavor of the meat off of that shawarma. Their goal as an organization is to be the best shawarma provider that exists. What they have ultimately found is their individual franchisees might want to get as much squeeze of that cone as possible to the detriment of their brand. So they’ve literally implemented Eeva to monitor the shape of the shawarma cone.
And they know the optimal spot, and so what they’re doing is they’re using it as a quality assurance measure to make sure that everything’s going smoothly.
We’ve got customers in Japan that have very stringent rules in terms of when the trash cans go out. It has to go out on a particular day. They’re now monitoring these trash cans because they can find the residents, and they can ultimately avoid fees and fines that are going to be the responsibility of the property owner when they can identify those individuals from their location. So if somebody is dumping something or throwing it away on a day where they shouldn’t, they can now get a report and an activity.
Q: What’s the potential for this technology in terms of safety going forward?
Briscoe: You’re going to see more specific models and prompts. For example, one of the things that we evolved to is if you started out early on, you were coming up with what those prompts were. We’re now building a library to say, “Hey, here are some other prompts that people have used.”
That does a couple things. One, it gets people’s minds kind of going, “Oh, wow, there are limitless possibilities here.” And two, it makes it a little bit easier for the layman that’s starting out. We didn’t want to make this a programming language. We didn’t want to make this something that people couldn’t figure out. It had to be built and architected in a way to where I could type and talk into it. “Hey, give me a list of activities and events that happened from 10 p.m. to 6 a.m. while I was at home sleeping.”
I can see people interacting with this where it simply becomes part of how they’re doing business, and for me to go home at night, to be able to arrive and say—especially for 24/7 facilities— give me a review of everything that occurred overnight is really a big deal. We’ve got people looking for rodents after hours. “I want to be notified if we have a rodent problem, see where they’re coming from.”
Q: Are there certain industries or verticals that you’re particularly focusing on now?
Briscoe: We do a lot with facilities management, we do a lot with warehouses, with data centers. More specifically, from a Brivo standpoint, we do a massive amount with education, and not to go on a tangent there, but we have a really good solution as it relates to our access control system. We have a 911 camera sharing solution for schools. We have a gun detection solution for schools. We also have vape detection. So you start to pull those things together. We really hit the story there. Maybe a little bit less with Eeva that we’ve been talking about.
We’re doing more in retail as these capabilities have come up. A lot of co-working spaces have become popular, but those are some of the top ones that we’ve been working with.