They’ll say “he grabbed his chest and went down.”
Nobody having the worst moment of their life reaches for the clinical term. They don’t say “active threat.” They say “there’s a guy with a knife.” They don’t say “structure fire.” They say “something’s burning and it’s spreading.” They don’t say “cardiac event.” They say “he just collapsed and he’s not breathing.”
We've spent decades around phone systems and emergency communications, and this is the detail that quietly breaks most “smart” 911 alerting on the market: it’s waiting for words people don’t actually use.
The problem hiding inside “keyword detection”
A lot of systems will tell you they do keyword detection. It sounds great in a demo. Then you look at how it actually works.
You build a list — “weapon,” “fire,” “heart attack,” “help,” “shooter” — and the system watches for those exact words to cross the wire. If the caller says the word, the alert fires. If they don’t, nothing happens.
So the entire safety value of the system comes down to one assumption: that a terrified person, mid-emergency, will happen to say the precise term that someone typed into a configuration screen months earlier. A receptionist watching a coworker collapse is not going to calmly announce “we have a cardiac event in the lobby.” She’s going to scream that he fell down and won’t wake up. And if “cardiac” was the trigger word, your system just sat there.
That’s not a safety feature. That’s a coin flip.
The worst part is that it fails silently. Nobody finds out the keyword list had a gap until the one call where it mattered most — and by then it’s a post-incident review, not a prevention.
What we built instead
This is the part of Echo we're proud of, and it isn’t the transcription itself. Real-time transcription is hard engineering, but plenty of people can transcribe a call. The thing that matters is the understanding underneath it.
Echo doesn’t just match strings against a list. It performs semantic analysis on the live call. It understands that “he grabbed his chest and collapsed” means the same thing as “cardiac” — and fires the same alert — even though the caller never used the word. It reads meaning, not vocabulary.
Consider how many ways a single emergency can be described:
- A cardiac emergency might come across as “he’s clutching his chest,” “she passed out and she’s gray,” or “my dad’s not breathing right.” None of those contain the word “heart.”
- A weapon might be “he’s got a gun,” “I think that’s a knife,” or “he’s threatening people in the parking lot.” A rigid list catches the first one and misses the rest.
- A fire might be “there’s smoke pouring out of the break room” or “the wall is hot and I smell something burning” — long before anyone says the word “fire.”
A keyword system treats each of those phrasings as a different problem, and it only solves the ones someone anticipated in advance. Echo treats them as what they are: the same emergency, described by a human being under stress.
So when your security team gets pinged, it isn’t because someone beat the odds and said the magic word. It’s because the system understood what was happening the way a trained human listening to the call would have understood it.
That’s the whole game. The caller should never have to be fluent in your keyword list while they’re trying to save someone’s life.
“Why is a 911 compliance company doing AI?”
It’s a fair question, and we get it a lot.
The honest answer comes down to where we sit. To make compliance work under Kari’s Law and RAY BAUM’s Act, 9Line has to be in the call path — we’re already part of the actual 911 conversation, not watching location data from the outside. That position is what makes any of this possible. A platform that only manages addresses and pushes a location to the PSAP can’t analyze a call it never hears.
So for us, the question was never “can we bolt transcription onto the product.” It was the opposite: we can already hear the call — why on earth are we throwing away everything in it except the location?
That’s the difference between sitting in the media and sitting beside it. Competitors who only manage dispatchable location can tell your team that a 911 call happened and where it came from. They physically cannot tell you what the call was about, because the content never passes through their system. We’re in a position to do something with it, so we did.
Notification is table stakes. Understanding is the difference.
Here’s where we've landed after decades in this work.
The notification telling your team a 911 call happened is the bare minimum. Everyone in this market does some version of it, and you should expect it as a baseline, not a feature.
The real difference is whether the system understands the call well enough to tell your people what they’re walking into — in the words real humans actually use, under the worst stress of their lives. A guard who knows “possible cardiac arrest, lobby, caller says he’s not breathing” responds very differently than a guard who just knows “a 911 call came from the building.” One of those is information. The other is an alarm with no context.
That context is what shortens the gap between the moment something goes wrong and the moment the right person is moving toward it with the right expectation. In an emergency, that gap is measured in seconds, and seconds are the entire ballgame.
Keyword matching waits for the magic word. Understanding doesn’t need one.
See it for yourself
If you want to see what semantic 911 alerting actually looks like in practice — not a slide, the real thing on a live call — we're happy to walk you through it.
Request an Echo demo and we’ll show you how Echo hears an emergency the way a person would.
Echo is part of 9Line’s cloud-based E911 platform — no software to install, built for organizations running Cisco Unified Call Manager, Microsoft Teams Direct Routing, and Cisco BroadWorks. Compliance with Kari’s Law and RAY BAUM’s Act, with intelligence built in.
