(Community Matters 196) How Using AI Helps Achieve Quicker Emergency Response in Calhoun County

Michael Armitage with the Calhoun County Consolidated Dispatch Authority talks about “Callie,” a virtual assistant now answering Calhoun County’s non-emergency dispatch lines so telecommunicators can stay focused on urgent calls. He breaks down how "Callie" detects emergencies, gets monitored and tuned so public safety outcomes improve without replacing people.


Episode Resources

Calhoun County Consolidated Dispatch Authority Website

ABOUT COMMUNITY MATTERS
Former WBCK Morning Show host Richard Piet (2014-2017) returns to host Community Matters, an interview program focused on community leaders and newsmakers in and around Battle Creek. Community Matters is heard Saturdays, 8:00 AM Eastern on WBCK-FM (95.3) and anytime at battlecreekpodcast.com.

Community Matters is sponsored by Lakeview Ford Lincoln and produced by Livemic Communications.

Transcript

Welcome And What’s New At 911

Richard Piet 0:01

Community Matters, that's us on Saturdays at 95.3 and Battle Creek Podcast.com any day including Saturday. Find us there along with these episodes and more. A service of Lakeview Ford Lincoln. Michael Armitage is back with us from the Calhoun County Consolidated Dispatch Authority. This time we're talking about a new innovation that's happened at the Calhoun County 911, and it involves a little bit of AI. Hello, Michael.

Michael Armitage 0:32

Hi, Richard. Good to be with you.

Richard Piet 0:34

This is really you, not the AI you. That is really me. Okay, just checking. Same here. It's really me. So we're making light of this, of course, but we know that the virtual assistant world is alive and well, and you found a way to make that work and thus make your team

The Non-Emergency Call Flood

Richard Piet 0:58

more productive. What is it?

Michael Armitage 1:00

Yeah, so we've been looking at different options over the last couple of years when it comes to non-emergency call handling, because about half of the 200,000 calls on average we take a year are on our 10-digit lines, mostly non-emergency. So, what tools are available to allow the telecommunicators to really focus on emergencies and have a more manageable uh workload?

Richard Piet 1:28

About half of your calls are non-emergency.

Michael Armitage 1:32

Yeah, a lot of them they range from you know maybe parking complaints to I need to call back an officer, uh, leave a message for an officer. So there's a lot of that type of phone activity that happens every single day in the dispatch center outside of the 911 calls.

Richard Piet 1:50

Yeah, well, not to uh certainly discount those calls, emergency calls for service, but I suppose half is better than three-quarters or 90% or something like that. So that's good. So there are actual virtual assistant options out there that can handle the non-emergency calls.

Picking A Virtual Assistant Vendor

Michael Armitage 2:12

Yeah, that's right. And we tested several different vendors that have products like this, and you know, we actually are not on the leading edge of this. There's about five uh centers already in Michigan alone that have already started using a virtual assistant over the last year or so, including our neighbors over in Kalamazoo. They were, I think, the first in Michigan to do it. So we've been watching their experiences as well as you know, for ourselves going out and looking at the different vendors that offer programs like this.

Richard Piet 2:43

So you got to put this through its paces.

Michael Armitage 2:46

Yeah, absolutely. We want to make sure this is something that we know is important. We want to make sure that we get it right, and we get a product that's both intuitive and that's able to spot those emergencies. So if somebody does call a 10-digit line and they do truly have an emergency, that bot is smart enough to know, okay, without any delay, that needs to go to a dispatcher.

Richard Piet 3:09

Yeah, so you need a really uh I was trying to find the right word, honed supplier that has a uh solution like this that behaves as well as a human.

Michael Armitage 3:22

That's right. And it's you know, that was a component of it too, is how fast it was able to process and and converse. It's it's you know, an element, of course, an important element is the accuracy of it, of course, but also you know, there are some comparable vendors with when it comes to accuracy, but maybe their bot took a lot longer to process, so it look took a lot longer to be able to get through that call, just as an example.

Richard Piet 3:49

Yeah, yeah. So, how did you test it? I mean, what were the kinds of things you wanted to make sure it could do?

Michael Armitage 3:56

Beyond being able to you know talk to people that used it, you know, we would run it through demos where internally where you know we would call in maybe a traffic complaint to the bot and just really see how it handles those scenarios. So we did a lot of you know, just that kind of testing of what we experience every day. How

Testing Real World Call Scenarios

Michael Armitage 4:18

does it handle those types of calls in a scenario situation?

Richard Piet 4:21

So somebody calls and says, uh, you know, Capitol Avenue 994, there's uh you know a hubcap in the road. I'm making this up. Uh but they're trying to report that to you, obviously a non-emergency thing, but something fairly important. And so how would it handle it?

Michael Armitage 4:39

Yeah, and actually in that situation, it might actually send it to a dispatcher because it's picking up that there's an object in the roadway. So, what it does is it, for instance, if there's a tree down, we go through the storm, we've experienced it already with one storm event. If it's just a tree down, it's not uh it just needs the road department to come out, you know, during storm events. We'll we'll set it a little bit looser where even if that tree's in the roadway, right? It the AI will handle the call and give it to and process it. But normal everyday scenarios, yeah, if there's a hazard that could be caused by it, it's going to escalate it. So even something like that, it probably will send a dispatcher.

Richard Piet 5:21

Yeah. And if somebody just called and said, I want to talk to Michael, uh, it knows then to transfer the call to you.

Michael Armitage 5:30

Yes, and you know, one of the probably the biggest success rates, too, that we've had so far with it is processing alarm calls because the it can take all the information from the alarm company, and then the nice feature about it is it puts it right into our dispatch system. So it pops, so the alarm company calls, and this can be for any call that it answers, right? But I'm using right now the example of alarm company, it could be a parking complaint or anything else. It will answer that call, process that call, and then automatically put it in our computer-aided dispatch. So it's right in front of our dispatchers. So it's just like one of their coworkers took the call, right? It's right there for the dispatcher to send to a police fire EMS.

Richard Piet 6:14

I would imagine too, some of these calls are also VAs, right? Or automated messages, it knows how to handle that.

Michael Armitage 6:21

Yeah, we get a lot of uh automated messages that come through those those lines for you know, maybe a carrier's experiencing a 911 outage. Those are calls that we'd had to before pick up, listen to the message, find out what the problem is. AI now can listen to that message for us and just send us a summary of hey, ATT or Comcast or whatever vendor it is is experiencing some sort of issue in their 911 network. Wow.

Richard Piet 6:51

Now,

Callie’s Script And Constant Learning

Richard Piet 6:52

will people be told that this is a virtual assistant, or is it so human interactive that you don't have to do that?

Michael Armitage 7:01

So, uh, right when you when you place the call, it introduces itself as Callie, and then that's what our staff we actually went out and asked uh what they'd like to have it named. We named it Callie, uh kind of uh spin-off of Calhoun County, right? So Callie will uh she introduces herself and then I mean ask you to describe what you're calling about, and then you know goes through the questioning process. And what's really cool, and the average person calling in won't recognize this, but we've done so much testing with it, is it's always learning. So we put in the parameters, we put in we spent months building kind of its knowledge base, right? But it's still constantly learning. So every time you call, you can call with the same situation twice in a row, and the questions that ask are be just slightly different because it has that conversational element to it.

Richard Piet 7:58

Interesting. Yeah, all right. So, of course, the uh point needs to be restated. This is non-emergency calls, right? So if someone dials 911, it's going to a human.

Michael Armitage 8:11

Yeah, absolutely. And the whole premise behind this again is to be able to free up, both free up our telecommunicators to handle those emergency calls, but also, you know, sometimes in in the afternoon, our ring times and non-emergencies, just because of our call volume, can be over a minute long. So if Callie can answer that call and determine, oh, maybe somebody is calling with an emergency, that gets bumped to the top of the queue. So it really serves two functions. It's being able to process you know a lot of those non-emergency calls, but also to quickly identify if any of those calls that are coming in

New Triage Rules And Call Priorities

Michael Armitage 8:50

on that 10-digit line are an emergency.

Richard Piet 8:52

So I'm trying to imagine the old way. 50% of the calls that come in are non-emergency. But if your staff is handling emergency calls at that moment, someone would have had to uh direct their attention to the non-emergency line. You probably staffed it accordingly, but walk us through the way it used to be.

Michael Armitage 9:12

Yeah, so yeah, we have an answering priority in dispatch, right? Our 911 lines are, of course, first priority, and then it was administrative lines. So, you know, our if everybody's tied up on a call, even if you're on a 911 call, let's say it's a lesser priority 911 call, they would have to put people on hold to be able to answer that ringing 10-digit line. And what this does is we actually added three tiers to our call answering system. We have 911, and then the second level of urgency now is if the AI detects an emergency on the 10-digit line, that gets bumped up a tier, and then everything else that comes in, it would be like our third, the third priority.

Richard Piet 9:59

I would imagine that your colleagues have a weight lifted off their shoulders in a way, because I mean there's a lot of weight there anyway when you're taking 911 calls, but uh to not have to worry about some other ringing line in the middle of a significant call you're trying to handle, this must be a big relief.

Human Oversight With The Live Dashboard

Michael Armitage 10:20

Yeah, and and as Callie's been implemented and trained, and as time's gone gone on over the last month or so, you know, we can start to really see now that transition to you know Callie being able to handle a lot of these calls. I think another important piece is we do have a dashboard too in dispatch where everybody's logged in and they can see what Callie's working on. So there is always still an element of human oversight, but the stress of not having that phone ring as much is really important um in the center and being able to, you know, it's a lot easier to multitask and maybe clear something off a dashboard than it is to have to put somebody on hold so that you can talk on another phone line.

Richard Piet 11:06

Yeah, absolutely. So this does not mean, does it, that uh suddenly you don't need as much manpower and uh you got fewer positions. That's what somebody's thinking. We hear this all the time, right? Oh, AI is gonna lay us all off.

Michael Armitage 11:22

Yeah, no, that's not the case at all. Actually, we uh last year added a position and included it in this year's

Michael Armitage 11:28

budget as well. So we are actually up a position authorized from where we were um you know a year, year and a half ago, and we don't anticipate that change. What what the struggle is, you know, there's there's a challenge in staffing across 911, and we're always behind the eight ball, unfortunately, which is probably the reason why we added this additional position. But you know, if you Google 911 staffing shortages, you'll see it across the country. And you know, we're trying to, yeah, I think we're a lot better off than a lot of those agencies are, but what what I'm getting at is you know, somebody can leave a position right in 911 and give us two weeks' notice. It takes us about eight months to fill that by the time you do the hiring process, the background checks, and then six months of on-the-job training.

Richard Piet 12:20

Oh my.

Michael Armitage 12:20

It takes from the time that we know that there's going to be a vacancy until the time we can get somebody in that seat, we're always at a disadvantage because it's just it's not like the fire fire departments and police departments that have academies that their employees can go to ahead of time. We're bringing people in off the street that may have never most likely have never worked in any sort of emergency services before.

Richard Piet 12:44

Yeah. You know, my experience in in hearing folks like you all in action is I don't know, dateline or 2020 when they play back emergency calls like that. And I'm always struck by how calm they are and reasoned they are and focused. This is not easy work.

CPR Coaching And Why Focus Matters

Michael Armitage 13:10

It's not, and I know this probably isn't necessarily the on topic, but one thing I do want to highlight because I think it's so important with our telecommunicators, you know, we have been working so hard collectively from police, fire, EMS to improve outcomes when it comes to uh cardiac arrest in Calhoun County, and we just got the latest report that shows a tremendous improvement, and that's one of the key elements of that, among many, is the ability to get hands on CPR over the phone. And our telecommunicators do a remarkable job if they detect through their protocols that somebody needs CPR, getting CPR started and an AED, if it's available to that patient as soon as possible, it saves lives. And when we talk about AI, it's so that our telecommunicators can focus on those calls and being able to give those instructions uninterrupted that really makes a difference.

Richard Piet 14:13

And when you say detects cardiac arrest, and now your your telecommunicators are focused on protocol, they might be talking someone through CPR who doesn't know how to do it. Is that right?

Michael Armitage 14:27

Yeah, absolutely. And and now think of the stress. Try to put yourself in that seat of helping instruct somebody how to do CPR while there's a phone ringing in your other ear.

Richard Piet 14:38

All right, Callie.

The Fight To Classify Telecommunicators

Richard Piet 14:40

Pick up that other line, will you? Man, this is incredible. You know, are telecommunicators at 911 considered first responders? I consider them first responders. Um seems like they ought to be, right?

Michael Armitage 14:53

Yeah, and that's at a national level. We've been fighting really hard. Um, we had to pass the Senate. Um, we're hoping that it'll pass the House at the federal level to change their classification to protected class because right now, which is a first responder, right now the federal OMB at the federal government classifies 901 telecommunicators essentially as secretaries. And this has been a fight that we've been you know trying to champion for a long time, and I believe strongly that they should be uh recognized at the federal level, state level, at every level as part of that first responder network. Wow, yeah, that's a whole other story, isn't it? But um somebody somewhere in the federal classification said these are telephone answers, so they're administrative. Wow, yeah, yeah, completely different. Yeah, all right. So how will you monitor Callie

Metrics Reviews And Weekly Tune Ups

Richard Piet 15:55

and know she is doing the uh the right thing here and also absorbing the training by doing, and you're able to see, yep, this is successful.

Michael Armitage 16:07

There's a couple elements we're looking at reporting and metrics that we'll start including in our annual report uh moving forward to be able to show how many calls Callie's been able to answer. But every you know, I mentioned the dashboard, there's a dashboard that the telecommunicators monitor so that they can see every call that that Callie handles and just kind of supervise that. But the other element is we meet on a regular basis, weekly basis right now with the vendor, and both the vendor and our administrative staff, we go through these calls too because we can see every call, what was said to Callie, and why Callie handled it the way that it did, and then we can go through and constantly make those tweaks and adjustments so that it it continues to be better in the future. So it is a constant review process that we've been going through since Callie's gone live and we'll continue to go through it as well.

Richard Piet 17:00

Well, we've said it repeatedly. I'll say it again, these are non-emergency calls that Callie's handling.

When Callers Refuse The Bot

Richard Piet 17:05

But if there's something that she can't handle, what happens?

Michael Armitage 17:09

If Callie can't handle it, or if the caller is not being willing to answer questions by Callie, there's what's called a resistance level, and Callie will transfer that call. So if a key if Callie can't handle the call, it'll get transferred to a telecommunicator.

Richard Piet 17:24

Uh-huh.

Michael Armitage 17:25

And I guess that brings me to another point real quick is yeah, I do ask, I know you can have preconceived notions about an AI bot or whatever, right? I do ask that consider the importance of taking these non-emergency calls off the shoulders of the dispatchers and just give her a shot. All right. Answer the questions. And I think I think most what we've seen is most of the time that a call gets transferred to dispatch is not because Callie couldn't have handled it, it's because the person that called wasn't necessarily giving Kelly the answers.

Richard Piet 18:01

Oh, interesting. So we have our own preconceived notions, we pesky humans, about talking to a machine sometimes that we don't want to do it.

Michael Armitage 18:12

Yeah, and again, I like I mentioned there's been so much fine-tuning and training. It's not just some bot that's searching the internet for things. This is specifically trained for Calhoun County.

Richard Piet 18:25

And in doing so, you probably had to teach her all the street names and all of these things.

Michael Armitage 18:31

Yeah, all of

Address Data Training And The True Cost

Michael Armitage 18:32

our address points, street data, all the GIS data that uh, which is our our spatial, our mapping data that we use, that all goes into to Callie to be able to verify the address to make sure that it is in Calhoun County. And so that's part of her process is when you give her an address, it's able to then verify that that is a valid address.

Richard Piet 18:56

All right. I think it's fascinating how this works. And so if you make a non-emergency call to the Calhoun County Consolidated Dispatch Authority, Callie will answer. And uh, maybe some of us won't even know the difference between Callie and a human, and she's being trained, and maybe you're helping with that training if you make that call to uh handle these efficiently. So is this pricey? I mean, you know, these this sounds complicated and uh expensive, Michael.

Michael Armitage 19:29

I can tell you it's it's less than a cost of filling a full-time position, especially since it's 24-7. So it's actually not a terribly expensive product. And I mean it's it's it's on par with a lot of software, but it's not what I would consider huge expense in comparison to what we spend on uh technology.

Richard Piet 19:49

Yeah, and and imagine if you had uh you would have more than this, but if you had somebody working 12 hours and a second person covering the other 12, it would be more expensive, I can imagine.

Michael Armitage 20:00

Yeah, it's about just about half the cost of a full-time position.

Richard Piet 20:04

Okay, there you go. Well, this is really interesting, and uh it'll be curious to see how this works, but we're acting uh I am acting all fascinated right now, but we're really headed in this direction, clearly. And and you said at least five uh consolidated dispatches in Michigan are already doing this, so this is uh the wave, the way we're going.

Michael Armitage 20:28

Yeah, absolutely. And uh I think you guys see more move in that direction as well.

Richard Piet 20:34

Yeah, all right. Well, thanks for uh bringing us up to speed on Callie. And uh hats off to uh you and your colleagues for what you do. I appreciate it. Thank you, Michael Armitage oversees the Calhoun County Consolidated Dispatch Authority. You are with Community Matters.