Industries / Government, education and community / Government / Postal and courier
Addresses and tracking
This is the work of giving every delivery point a code a machine can read, and of telling senders and receivers where an item is on its way.
7
Milestones achieved, from 1857 to 2018.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1857
London split into districts
London had the same street names in different places, so Sir Rowland Hill divided it into ten postal districts: East Central and West Central, ringed by eight compass-point districts such as N and NE. The plan was carried out in 1857 and 1858, and the public was asked to add the district to the end of each address. Two districts, E and S, were dropped about ten years later. Numbers were added to the districts during the First World War to help newly recruited sorters who did not know London's streets; Royal Mail's British Philatelic Bulletin gives 1917 for this and Post & Parcel gives 1916.
Credit
Sir Rowland Hill
1959
Postcodes tried in Norwich
Eight sorting machines were put into use in Norwich in January 1959. To let them sort mail to each postman's round, the Post Office gave Norwich addresses codes starting NOR, introduced on 8 October 1959. Royal Mail's British Philatelic Bulletin says the experiment "did not prove a great success, mainly because many did not include the postcode in the address." Croydon, coded in 1966, set the pattern of an outward part for the place and an inward part for a few addresses, and every UK address had a postcode by 1974. Post & Parcel calls Norwich the world's first trial of address codes designed for machine sorting.
LessonA code helps only when senders write it, so the public has to be won over as well as the machines installed.
Credit
the Post Office; Postmaster General Ernest Marples
1983
Nine digits for the carrier
The US Postal Service added four digits to the ZIP Code in 1983. The extra digits name a block side, a floor of an office building or a group of post office boxes, so machines could sort letters down to the carrier. From 1 October 1983 mailers paid less postage for using ZIP+4. Under further discounts, the share of mail with barcodes printed by the sender rose from 7% in 1990 to 59% in 2000, by the Postal Service's count.
LessonDiscounts got large mailers to print the codes themselves, which moved much of the coding work out of the post office.
Credit
US Postal Service
1994
Tracking on the web
Federal Express renamed itself FedEx in 1994 and launched fedex.com, which FedEx calls the first transportation website with online package tracking. Customers could look up a parcel themselves instead of phoning. Online tracking is now routine for every large carrier.
Credit
FedEx (the "first" is FedEx's own claim)
2016
Three words as an address
Mongol Post, Mongolia's national post, adopted what3words as its national addressing system in May 2016. what3words names every 3 m by 3 m square on Earth with three words. Mongolia's three million people had no consistent addresses, and many collected mail from post office boxes dozens of kilometers from home, by what3words' account. A 2023 study by Rudy Arthur of the University of Exeter found that, counting only typing errors and simple sound-alike words, about a quarter of three-word addresses have more than three others they could be confused with, some close by. He noted that the word list is not public and the license terms make real-world tests hard.
LessonA closed address code cannot be checked by outsiders, so its error rate rests on the owner's word.
Credit
Mongol Post; what3words
2017
Photos of the day's mail
The US Postal Service rolled out Informed Delivery nationwide in 2017, after tests from 2014. It emails each subscriber pictures of the outside of the letters due that day, and package tracking was added to the emails the same year.
Credit
US Postal Service
2018
Address changes open to fraud
The US Postal Service processed 36.8 million change-of-address requests in fiscal year 2017. Online requests were checked against a credit card's billing address, and letters went to both the old and the new address. Its Inspector General reported in 2018 that the Postal Inspection Service had received nearly 25,000 complaints about such requests since January 2016, and recommended a government ID check for paper requests. A 2022 Inspector General audit found fraudulent online changes of address rose from 8,857 in 2020 to 23,606 in 2021; Postal Service managers disagreed, saying stricter checks would burden millions of customers to cut risk for a few thousand.
LessonThe form that redirects someone's mail is an identity check, and a weak one hands a thief the victim's post.
Credit
USPS Office of Inspector General
Watch for
India's grid code for addresses
India's Department of Posts launched its "Know Your DIGIPIN" portal on 27 May 2025. DIGIPIN gives each cell of about 4 m by 4 m in India a 10-character code worked out from latitude and longitude. It was built with IIT Hyderabad and the National Remote Sensing Centre of ISRO, its code is open source, and it is offered to ministries, state governments and institutions to build into their work.
Achieved whenIndia Post or an outside party publishes how many deliveries carry a DIGIPIN and a measured change in failed deliveries where it is used.
Claimed by
India's Department of Posts, May 2025
Open
01
Location codes read aloud
A test of how often location codes go wrong when spoken, as in a phone call to a courier or an emergency line. Generate 1,000 codes for public places in each open system, Google's Plus Codes and India's DIGIPIN, and what3words where its license allows. Have volunteers and synthetic voices read them over phone-quality audio, transcribe them with speech-recognition models and with people, decode each transcript, and measure how far the decoded point lands from the true one.
The briefCredit · unclaimed02
Tracking scans checked by logger
A public log that compares each tracking event of test parcels with an independent tracker packed inside. Volunteers mail parcels to each other with several carriers, within one country and across borders. A script lines up each event's time and stated place with the tracker's trail and flags events recorded late, early or far from where the parcel was, and stops the carrier never reported.
The briefCredit · unclaimed
Open 01 · the brief
Location codes read aloud
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A test of how often location codes go wrong when spoken, as in a phone call to a courier or an emergency line. Generate 1,000 codes for public places in each open system, Google's Plus Codes and India's DIGIPIN, and what3words where its license allows. Have volunteers and synthetic voices read them over phone-quality audio, transcribe them with speech-recognition models and with people, decode each transcript, and measure how far the decoded point lands from the true one.
Why it is still open
Arthur's 2023 study of what3words modeled only typing errors and sound-alike words, and said its license terms make real tests hard. A 2026 test of 15 speech-recognition models from OpenAI, Deepgram, Google and Microsoft found an average error rate of 44% on spoken US street names, but did not test location codes. This research found no controlled spoken test of any location code.
Built with what exists
Google's Open Location Code library (github.com/google/open-location-code, Apache 2.0); India Post's open DIGIPIN code (github.com/INDIAPOST-gov/digipin); text-to-speech voices; speech recognizers such as OpenAI Whisper and ElevenLabs Scribe v3 (see the Note "Fewer word errors"); ffmpeg to cut audio down to telephone quality.
First step
Have five volunteers read 100 Plus Codes and 100 DIGIPINs for public landmarks, run Whisper on the recordings and score the decodes.
Done looks like
1,000 codes per system, each read by at least 20 volunteers and three synthetic voices and transcribed by at least two recognizers, with the share decoded to the wrong place, the share decoded to a wrong place within 1 km (the dangerous case, since it looks plausible), and the median error distance, published with the recordings.
Open 02 · the brief
Tracking scans checked by logger
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public log that compares each tracking event of test parcels with an independent tracker packed inside. Volunteers mail parcels to each other with several carriers, within one country and across borders. A script lines up each event's time and stated place with the tracker's trail and flags events recorded late, early or far from where the parcel was, and stops the carrier never reported.
Why it is still open
The largest check found is the USPS Inspector General's audit of about 2 billion package scans from July to December 2016, which used the carriers' own handheld GPS: of 25.5 million scans made late in the evening, about 1.9 million were "stop-the-clock" delivery scans made at the delivery office instead of at the address, and 8.3 million had no location data. The European method for measuring parcel transit times, CEN/TS 15472:2016, also works from the carriers' own tracking records. Hobbyists have mailed Apple AirTags to watch a parcel's journey, such as the writer Kirk McElhearn across England in 2021, without scoring the trail against the carrier's scans. This research found no independent test across carriers.
Built with what exists
Bluetooth trackers such as Apple AirTag or Samsung Galaxy SmartTag2, which report through passing phones, and GPS loggers with a cellular modem for exact times, with batteries that meet each carrier's lithium-battery rules; the carriers' public tracking pages for the volunteers' own parcels, saved by a browser agent (see the Note "Deleting 80% to start over"); Python with GeoPandas to match events to trails.
First step
Read each carrier's rules on batteries in parcels, then mail ten parcels with two kinds of tracker between two volunteers in one country and score their events by hand.
Done looks like
200 parcels across at least four carriers and five country pairs, with every tracking event scored as matching, late by a stated number of hours, or off by a stated distance, published with the code and the anonymized trails.