On February 16, 1968, a call placed inside the Haleyville, Alabama city hall was answered on a red telephone at the police station a few blocks away, and that is the event generally recognized as the first 911 call in the United States. Almost six decades later a large share of the country still routes 911 over selective routers and databases whose architecture would be recognizable to the people who built that first system. This piece traces how the routing and location plumbing actually works, what Next Generation 911 changes, and why the transition stalls on money and authority rather than on engineering.

Haleyville, February 1968, and what came before it

The idea of a single emergency number was not American in origin. London put 999 into service in 1937, and other countries followed with their own short codes well before the United States did anything nationally. The push here came from the President’s Commission on Law Enforcement and Administration of Justice, whose 1967 report recommended that a single telephone number be established for emergency calls, because a citizen in most American cities at the time had to know a seven-digit number for police and a different one for fire, and often a third for an ambulance service that might be run by a funeral home.

AT&T announced in January 1968 that it would make the digits 911 available as a nationwide emergency number, chosen in part because the combination was short to dial on a rotary set and was not in use as an area code or a service code. The Alabama Telephone Company, an independent operating company outside the Bell System, put a working 911 system into Haleyville within weeks of that announcement. Rankin Fite, then Speaker of the Alabama House of Representatives, placed the inaugural call from the city hall, and Congressman Tom Bevill answered it at the police station. The accounts that circulate in the industry credit ATC president Bob Gallagher with the decision and inside plant manager Robert Fitzgerald with the engineering, and describe the motive as beating the Bell System to the milestone, which is a story told consistently enough to repeat but which I would treat as recounted history rather than documented record.

You will also see claims that Nome, Alaska had 911 service in 1966. That claim is not as well documented as the Haleyville event, and I have not seen primary material supporting it, so treat it as an unresolved footnote rather than a correction to the usual account. What the Haleyville system did not have is worth naming: no automatic number identification, no address display, and no ability to route a call anywhere except to the one answering point that the switch was wired to serve.

Why the national rollout took three decades

The White House Office of Telecommunications Policy issued a national policy bulletin in 1973 urging nationwide adoption of 911 and recommending a federal role in coordinating it, and that is roughly where federal involvement stopped for a long time. There was no statutory mandate, no appropriation, and no single agency in charge. Every 911 system in the country was therefore built by a county commission, a city council, a special district, or a state legislature that decided to create one, funded by a fee that the same body had to be persuaded to impose on telephone subscribers.

That structure explains the pace better than any technical constraint does. A county that wanted 911 had to consolidate answering points, negotiate with the telephone company for trunking and database service, build an address database, and in many rural places assign street addresses to properties that had never had one, because rural route and box numbers cannot be used to route a call or dispatch an engine. I have worked in Georgia counties where the 911 addressing project was the reason road names got posted at all, and where the fire department did the field verification because nobody else had the staff.

Congress did not designate 911 as the universal emergency number until the Wireless Communications and Public Safety Act of 1999. Later statutes built the federal support structure around it: the ENHANCE 911 Act of 2004 created a joint NHTSA and NTIA program office and a grant program, the NET 911 Improvement Act of 2008 addressed IP-based provider obligations, and the Next Generation 911 Advancement Act of 2012 authorized further grants. None of those statutes made 911 a federal service, and none of them changed the fact that the operating authority is local.

NENA publishes an estimate that 911 service now reaches roughly 96 percent of the geographic United States and that Americans place something on the order of 240 million 911 calls a year, with NENA also reporting that 80 percent or more of those calls originate on wireless devices. Both figures are association estimates rather than census counts, and the wireless share in a specific county depends heavily on how much wireline service survives there.

Selective routing, ANI and ALI: how legacy E911 works

Basic 911 delivers a call to one answering point per switch, which works only when the switch boundary and the jurisdiction boundary happen to agree. Enhanced 911 was built to solve that mismatch, and the mechanism has three parts that get conflated constantly, so it is worth separating them.

Selective routing is the routing decision. Trunks from the originating end office run to a selective router, historically a telephone company tandem switch, which looks up the calling number in a selective routing database. That database maps each telephone number to an Emergency Service Number, a code representing the Emergency Service Zone in which the subscriber’s address sits, and the ESN identifies the correct primary PSAP along with the law, fire and EMS agencies responsible for that zone. The zone boundaries come from the Master Street Address Guide, the MSAG, which is a table of every valid street name, address range and community in the served area with the responding agencies attached. When a new subdivision goes in and the addressing coordinator does not get the ranges into the MSAG, calls from those houses fall out as unlocatable or route to the wrong PSAP, and that is a records maintenance failure rather than a network failure.

ANI is the calling number delivered with the call. On the classic CAMA trunks used for 911 it arrived as multifrequency digits, in most deployments eight digits consisting of one information digit and a seven-digit number, which is why area code handling required extra engineering and why some centers ran 10-digit or 20-digit ANI variants. ALI is the location lookup. The PSAP’s ALI controller takes the delivered ANI, queries an ALI database over a dedicated circuit, and gets back a record containing the civic address, the subscriber name, a class of service code such as residential, business, coin or later wireless and VoIP, and the responding agency identifiers for that address.

The important property of the legacy design is that the telephone number is the key to everything. Routing, address and agency assignment are all derived from a number that was provisioned to a fixed location, which is exactly the assumption that mobile phones and interconnected VoIP destroyed.

Wireless Phase I, Phase II, and the routing workarounds

The FCC took up wireless 911 in the proceeding usually cited as CC Docket 94-102 and adopted rules in 1996 that created two phases. Phase I required a carrier to deliver the caller’s callback number and the location of the cell site or sector handling the call. Phase II required the delivery of latitude and longitude. After a 1999 revision the accuracy benchmarks were set differently depending on the technology used, with network-based solutions required to meet 100 meters for 67 percent of calls and 300 meters for 95 percent, and handset-based solutions required to meet 50 meters for 67 percent and 150 meters for 95 percent. Those benchmarks applied to averaged carrier performance, not to any individual call, which is a distinction that has caused more arguments in PSAP conference rooms than almost anything else in this subject.

Deployment was slow and uneven because the carrier obligation was triggered by a valid request from a PSAP that was capable of receiving and using the data, and in the early years a state cost recovery mechanism was also part of the picture. A center that had not upgraded its CPE and mapping could not make a valid request, and a center that could not fund the upgrade stayed on Phase 0 for years while the surrounding counties moved. Deadlines were extended repeatedly through the 2000s.

The routing mechanics deserve attention because they show how much strain the legacy architecture was under. A wireless call has no fixed address associated with its number, so the mobile network assigns a pseudo-ANI, either an Emergency Services Routing Digit or an Emergency Services Routing Key, which is a routable number tied to a cell sector. The selective router uses that pseudo-ANI to pick the PSAP, and the PSAP’s ALI query on the same pseudo-ANI is steered to a mobile positioning center or gateway mobile location center, which returns coordinates. The rebid button on a dispatcher’s ALI screen exists because the first response is often the sector centroid and a second query, seconds later, may return a computed position.

The FCC adopted indoor location rules in 2015 that phased in a 50 meter horizontal benchmark and added a vertical, or z-axis, metric of plus or minus 3 meters relative to the handset for a defined percentage of calls, phased by market size. Those compliance dates have been extended more than once, so verify the current requirements and deadlines directly in the FCC’s rules rather than relying on a summary, including this one.

What NG911 actually changes about routing and media

Next Generation 911 replaces the circuit-switched selective router and the number-keyed databases with an IP network and a session-based call model. The reference architecture is NENA’s i3 standard, which NENA publishes and revises, and the transport is an Emergency Services IP network, the ESInet, carrying Next Generation Core Services. Calls arrive as SIP sessions through a border control function, and a routing proxy applies policy to decide where they go.

The substantive change is that routing is driven by location rather than by telephone number. The originating network supplies location with the call, either by value or by reference, in a Presence Information Data Format Location Object, and the routing function queries an Emergency Call Routing Function, which is a geospatial lookup: given this point, which PSAP serves it, and which law, fire and EMS agencies serve it. A companion Location Validation Function checks civic addresses against the same authoritative GIS data before an emergency ever happens, which is the NG911 replacement for MSAG validation. Policy rules layered on top of the routing function let a 911 authority define, in advance, what happens when the primary PSAP is at capacity, is evacuated, or is out of service, and that alternate routing executes without a telephone company technician touching anything.

Media is the change that gets the most attention and delivers the least by itself. An IP call path can carry text, images, video and telematics data, and i3 defines how text sessions are handled through a text control center. Whether your center can accept a photograph from a caller depends on the call handling equipment, the CAD integration, the retention and public records treatment of that image under your state’s law, and whether a telecommunicator handling three calls has any capacity to look at it. The network capability arrives years before the operational answer does.

Transition is handled by gateways rather than by a flag day. Legacy network gateways bring in traffic from originating networks that are still TDM, legacy PSAP gateways let a center that has not replaced its CPE receive calls from the ESInet, and legacy selective router gateways bridge traffic and transfers between the old tandem and the new core so that neighboring jurisdictions on different sides of the transition can still hand calls to each other. In 2024 the FCC adopted rules on the NG911 transition requiring originating service providers, upon a valid request from a 911 authority, to deliver 911 traffic in IP format to a designated point, with timelines that vary by provider type, and I would confirm the current compliance dates with the FCC before building a schedule around them.

GIS becomes the address database, and somebody has to own it

In legacy E911 the authoritative record is a tabular MSAG maintained by an addressing coordinator and reconciled periodically against the telephone company’s subscriber records. In NG911 the authoritative record is a set of GIS layers, principally road centerlines with address ranges, site and structure address points, PSAP boundary polygons, and emergency service boundary polygons for law, fire and EMS. Those layers feed the ECRF and the LVF, which means that a gap or an overlap in a boundary polygon is no longer a map display problem but a call routing problem.

The quality bar is higher than most local GIS shops are used to. Boundary layers have to be topologically clean, with no slivers between adjacent PSAP polygons and no areas claimed by two agencies, because the routing function has to return exactly one answer. Address points have to match the centerline ranges at a high rate, and NENA publishes data model and quality guidance that most state programs adopt as their provisioning threshold. The synchronization between the GIS data and the legacy MSAG has to be maintained throughout the transition period, because both are live at the same time and a discrepancy sends the same call to two different places depending on which path it takes.

This is ongoing labor, not a project with an end date. New roads get built, subdivisions get platted, cities annex, fire districts consolidate, and every one of those events is a data maintenance obligation with a routing consequence. The counties that struggle with NG911 are usually the ones that funded a one-time data cleanup with grant money and did not fund the position that keeps the data current afterward, which is a budget decision rather than a technical shortfall, and it is the reason I put GIS staffing in front of ESInet procurement when anyone asks me about sequencing.

Who pays for 911, and why the transition breaks the model

911 in the United States is funded principally by a fee assessed on telephone service, set and collected under state law, distributed by a state agency or retained locally, and supplemented by county general funds in most places. The fee was designed around a world of billed access lines, and it has been extended to wireless and VoIP with varying success. The FCC publishes an annual report to Congress on state collection and distribution of 911 fees, and each edition identifies a small number of states and territories that diverted fee revenue to purposes other than 911. The list changes from year to year, so read the current report rather than citing a count from memory. The Don’t Break Up the T-Band Act of 2020 also directed the FCC to address acceptable uses of 911 fees and set up an interagency effort on diversion, which tells you how persistent the problem has been.

The transition creates a specific and predictable funding crunch: for the length of the migration, a 911 authority pays for the legacy selective routing and ALI service and for the ESInet and core services at the same time, plus the gateways that connect them, plus the GIS work, plus new call handling equipment, plus training. The savings that eventually justify the move do not arrive until the legacy tariffed services are actually turned down, and they cannot be turned down until every originating carrier and every neighboring PSAP is off them.

The National 911 Program published a cost study in 2018 that estimated the national cost of implementing NG911 over ten years in a range of roughly $9.5 billion to $12.7 billion, depending on the assumptions used about architecture and existing infrastructure. That is a 2018 estimate with 2018 prices and it should be read that way. Bills introduced in recent Congresses have proposed one-time federal appropriations of a broadly similar order of magnitude for NG911 transition, and as of this writing on September 3, 2026 no appropriation at that scale has been enacted, so verify the current federal funding picture with the National 911 Program before you build it into a plan.

Three thousand authorities and the interoperability question

Counting PSAPs is harder than it sounds. The FCC maintains a PSAP registry and the National 911 Program maintains a national profile database, and the two produce numbers in the general neighborhood of 5,700 to 6,000 primary and secondary answering points without agreeing exactly, because they count secondary PSAPs, backup facilities and consolidations differently. Above them sit something on the order of three thousand separate 911 authorities with legal responsibility for service. Any national figure in this area should be quoted with its source and its date attached.

That fragmentation is the governance problem in one sentence. State 911 boards differ enormously in what they can actually do: some own and operate a statewide ESInet and set binding standards, some administer grants and publish guidance that counties may accept or ignore, and some exist mainly to allocate fee revenue. A state in the second or third category cannot compel a home rule county to migrate, cannot compel it to maintain GIS to a standard, and cannot compel it to accept transfers under common policy rules, which means the state’s NG911 completion date is set by its slowest county rather than by its contract.

Interoperability between ESInets is the piece that most regions have not solved. Transferring a call with its location and its media from one ESInet to another requires interconnection between the networks, agreement on how routing policy and credentials work across the boundary, and a legal agreement covering liability and records. Two adjacent states can both be fully i3 compliant and still be unable to pass a call cleanly if nobody negotiated the interconnection, and mutual aid does not stop at a state line for a wildland fire or a mass casualty incident on an interstate highway.

The security dimension belongs in the governance conversation rather than the engineering one. Moving 911 onto IP networks changes the exposure profile, telephony denial of service and ransomware against call handling and CAD systems have both affected centers in the United States, and the responsible structure is a security program with an owner, a monitoring arrangement and an incident response plan agreed in advance with the state. I am not going to go past awareness level here, and any specific hardening work belongs with your state 911 authority and CISA rather than in a public article.

What to do at your agency

• Ask your 911 director in writing which routing path your calls take today, whether the routing decision is made by a legacy selective router on ESN or by an ECRF on GIS, and what the projected date is for turning down the legacy path, then file the answer with your communications plan.
• Have your training officer or a company officer place a test call to 911 from a landline or desk phone at each of your stations and confirm that the address and responding agency displayed at the PSAP match reality, and submit corrections through the county addressing authority for any that do not.
• Ask the county GIS coordinator for the current address point to road centerline match rate and the most recent boundary validation report for your department’s response area, and reconcile the fire, EMS and law emergency service boundary polygons against the response district maps your agency actually uses.
• Get the name and the submission process for MSAG and GIS address corrections into your department’s SOP or reference binder so that a crew that finds a wrong or missing address on a call knows exactly where the correction goes.
• Put NG911 transition status on the agenda of the county or regional public safety committee that already meets, with two specific questions: what the parallel legacy and ESInet operating cost is during migration, and which line item funds it.
• Ask your PSAP supervisor how the center currently handles a caller who offers a photograph or video, what the call handling equipment can accept, and what your state’s retention and open records law requires once it is received, and write the answer into the center’s existing procedure.
• Test one live transfer to each neighboring PSAP you routinely hand calls to and verify whether the caller’s location data travels with the transfer or has to be read aloud, then document the result for your mutual aid file.

Takeaways

• The first 911 call in the United States is generally recognized as the one placed in Haleyville, Alabama on February 16, 1968, built by the independent Alabama Telephone Company weeks after AT&T announced 911 as a national emergency number, following a 1967 federal commission recommendation for a single number.
• 911 spread county by county for three decades because no federal statute required it and no federal money paid for it, and Congress did not designate 911 as the universal national emergency number until the Wireless Communications and Public Safety Act of 1999.
• Legacy Enhanced 911 keys everything to the calling telephone number: selective routing maps the number to an Emergency Service Number through the MSAG, ANI delivers the number to the PSAP, and an ALI query returns the address, class of service and responding agencies.
• The FCC’s wireless E911 rules created Phase I, which delivers callback number and cell sector, and Phase II, which delivers coordinates against averaged accuracy benchmarks, with carrier obligations triggered by a valid PSAP request, and wireless calls are routed through a pseudo-ANI because there is no fixed address behind the number.
• NG911 replaces number-keyed routing with location-based routing over an ESInet using SIP, where a routing proxy queries an ECRF against authoritative GIS data and an LVF validates civic addresses in advance, and gateways bridge legacy and next generation traffic during the migration.
• An IP-based ESInet carrying legacy call flows into legacy CPE is a transition stage rather than location-based routing, and the test that distinguishes them is whether the ECRF is making the routing decision from GIS.
• The authoritative data in NG911 is a set of GIS layers requiring continuous maintenance to a defined quality standard, which is a permanently staffed function rather than a one-time cleanup funded by a grant.
• The National 911 Program’s 2018 cost study estimated national NG911 implementation at roughly $9.5 billion to $12.7 billion over ten years, no federal appropriation at that scale had been enacted as of September 3, 2026, and the migration requires paying for legacy and next generation service simultaneously until the last carrier and neighboring PSAP move.
• With roughly 5,700 to 6,000 answering points under about three thousand separate 911 authorities, and with state boards holding very different levels of authority, the completion date for a state’s transition is set by its least funded county and by whether anyone has negotiated interconnection between adjacent ESInets.