If you dial 911 today, you are probably using a network designed for rotary telephones. That is not an exaggeration. It is the engineering reality that Next Generation 911 exists to correct.
The system we have
The 911 system most of the country still runs on was built in the late 1960s and extended, patch by patch, for half a century. It assumes a call comes from a telephone, that the telephone is attached to a wire, and that the wire ends at a fixed address.
Two pieces of equipment carry that assumption. A selective router decides which call center gets the call, based on where the wire comes from. A database called ALI — Automatic Location Identification — looks up the address on file for that line and shows it to whoever answers.
The assumption held for decades. It does not hold now. The great majority of 911 calls come from mobile phones, which have no wire and no fixed address. Callers text. They send photographs. Cars, watches, and medical monitors place calls on their owners' behalf. None of that fits down a circuit built to carry one voice from a known address to a known building. The workarounds have been ingenious and are reaching their limit.
What NG911 actually is
Next Generation 911 replaces the circuit with an internet protocol network. That sentence contains the whole change; everything else follows from it.
The architecture is defined by a standard called i3, published by the National Emergency Number Association. Its central idea is the ESInet — an Emergency Services IP network, shared by every public safety agency that might be involved in an emergency. Running on it is a set of functions called NG911 Core Services, which do the work the selective router and the ALI database used to do, but determine location at the moment of the call rather than looking it up in a table.
The structural difference. Legacy routing asks where the wire comes from. NG911 asks where the caller is.
Three capabilities follow, and each one changes the job of the person answering. Location is computed rather than looked up, so the system reports where a caller actually is instead of the address a carrier once recorded. Calls can move, carrying their location and data to any other center on the network — under the old system a misrouted call had to be re-taken by voice from the beginning. And the channel is no longer only voice: text, images, video, and telematics arrive alongside the call.
Why it is happening now
For most of the past decade NG911 was a goal without a deadline. States moved at their own pace, funded from their own 911 fees, which produced a map of wide disparity — some largely converted, others almost entirely on legacy equipment.
That changed in July 2024, when the Federal Communications Commission adopted its first nationwide transition rules, setting responsibilities and deadlines for carriers to build NG911 capability into their networks while preserving each jurisdiction's right to set its own configuration, timing, and cost arrangements. The rules took effect that November, with a two-phase structure triggered by requests from individual 911 authorities rather than by one national date. The scale is easier to grasp in dollars: forty-two states, the District of Columbia, Guam, and Puerto Rico reported NG911 spending in calendar year 2024, totalling more than $535 million in that year alone.
A decade-long aspiration became, within about eighteen months, a regulated programme with dates attached.
Where it gets hard
Here is the part that gets less attention than it deserves. The infrastructure work is expensive and complex but fundamentally solvable — it is engineering, and engineering yields to money and time. The harder problem is that the transition changes what a job requires, for tens of thousands of people, in a profession already under strain.
The federal National 911 Program has been blunt about this in its own operational analysis: the volume of incoming information rises sharply, and the people answering will have to be trained in a fundamentally different way — not only in the new tools, but in how to sort competing streams of data under pressure. The failure modes of abundance are real. Faced with too many inputs, a person can stall on the decision, or lock onto the input that matters least.
That is not a software training problem. It is a judgement problem, and judgement under time pressure is trained differently from feature knowledge. Someone who has spent fifteen years forming an accurate picture from voice alone now has video, and must learn when the video helps and when it costs the seconds that matter.
Two structural conditions make this harder than it needs to be. First, there is no single training authority: 911 centers are run by a wide range of local and state agencies, and that varied governance means each may set its own standards. National bodies publish recommended minimums and certifications exist, but no one entity can require a curriculum nationwide. A technology transition governed by federal rules meets a training landscape governed by nobody in particular. Second, the workforce is already stretched. Staffing shortages and burnout are the sector's dominant concerns. Training that assumes generous time away from the console is training that will not happen.
One transition, four audiences, four instructional problems. Treating them as one curriculum is the common failure.
Why this is a design problem
It is tempting to treat training as the last item on the deployment plan, scheduled after the equipment is racked and the network tested. That ordering guarantees the worst version of it: compressed, generic, delivered by whoever is free, measured by attendance rather than capability.
The alternative is not more training. It is training designed alongside the system, by someone who understands both the technology and how adults actually learn under pressure. Scope the audiences separately before writing a word. Decide what truly requires a person in a room and what can be four minutes completed between calls. Build assessment that measures decisions rather than recall. Treat the documentation as part of the instructional system, because in this field the reference material is consulted mid-incident by someone with no time to read.
None of that is exotic. It is ordinary instructional design applied early enough to matter. What is distinctive about NG911 is only the stakes: the failure mode of bad training here is not a poor completion rate. It is a delay in a response that had seconds to spare.
The short version
The country is replacing the network emergency calls travel over, from a fifty-year-old circuit-switched design to a shared IP architecture that knows where callers are and can carry more than their voices. Federal rules now set the pace where state initiative used to. The engineering is well understood and adequately funded.
What remains uncertain is whether the people who will operate it are ready on the day it turns on. That is a training question, and it is the one being answered last.