How the US Postal Service reads terrible handwriting
At the Remote Encoding Center in Salt Lake City, keyers process 1.2 billion images of mail every year. It’s a more difficult job than I thought.
– Almost all the letters that get sent in the United States
are processed automatically.
You put a letter in a post box,
it’s taken to the closest processing center,
where cameras and computers take a picture of the front of the envelope,
read the address with optical character recognition, OCR,
and it’s then passed on to the right truck or plane.
That deals with pretty much everything, even the handwritten letters.
But what if your handwriting is really bad?
Like, really bad.
Or the envelope got a bit damp, and the ink blurred and ran?
The postal service sends the picture of the envelope here
to the Remote Encoding Center, the REC, in Salt Lake City, Utah,
and it’s the job of the folks here to turn scrawl and blurred ink
into actual addresses.
It’s a quiet day here today.
Mail volumes are low in the middle of summer.
So while it’s not busy inside,
I’m going to see if I’d be any good at this.
– We’re in the last Remote Encoding Center in the United States.
Back in 1997, when we had 55 RECs open,
all of those RECs combined keyed 19 billion images.
The OCR technology is so good that, in 2021,
as the last REC remaining, we only keyed 1.2 billion images.
Right now, we have about 810 employees here.
Shortly, we’re going to be teaching Tom how
to process some of this mail.
It’ll be a lot of fun.
– It’s important to say that I can’t show you any actual mail,
not the envelopes, not anything.
If any appears in the background on a screen,
it’ll be blurred out. That’s all considered strictly private.
And besides, I’m not going to flash random people’s addresses
on the screen of a YouTube video,
particularly when they’ll usually be written
in distinctive handwriting?
You can probably work out how that could go wrong.
But there is a stack of demonstration mail,
envelopes and addresses used for training,
and that’s what I’m going to be tested on.
– If you drop off a piece of mail,
it makes it to the local processing plant.
If the machine can’t read your handwriting,
one of the keyers here will type in the missing information,
and the information goes right back to the plant,
stays within automation, which is a whole lot cheaper
than having somebody physically sort that piece by hand.
– All right, what do I need to do?
– Put your last name and first initial into that.
– I looked at the keyboard to look for the numbers
and realized this is a-
– That is not a standard keyboard! – That is not a standard keyboard.
Other than the QWERTY letters and the numpad,
everything is different. Okay.
– To make everything faster for us,
they did some weird things with the keyboards
that they’ve provided us.
The home row, A, S, D, F, those also work as your numbers.
You can’t reach up to the top of the keyboard to hit 1, 2, 3.
– “How to key the outward portion of a mailpiece image.”
– We’re calling outward the city and the state.
– City and the state, okay.
Siemens came up with this in the ’90s.
So they decided “C coding” is going to mean zip code.
Outward is going to mean city and state.
Inward is going to mean the street address.
– And that’s now the language that you use,
because that’s what they decided.
– To make it faster to key these letters,
our keying is a small extract of what’s on the letter,
and then it compares it to the known-good addresses in the database.
Every known-good address in America is sitting
on our servers in the back.
If it’s good, the piece just goes away, and you’re done.
If it matches a couple of good addresses,
then you’ll get a list and you’ll choose out of that list,
which is the address you were trying for.
It makes it a whole lot faster than typing out the entire things.
– “Key in the first three characters of the first word in the city name.”
Okay, first character of the second word.
Two character state or territory of abbreviation.
– It’s the 3+1 rule. It’s the same for the street addresses.
– Okay. “Press the Go key when ready.”
So that is LE TMS.
– Oh. 3+1 rule.
If there’s not three characters, you got to do a space.
– Ah, okay.
– Now, you’ve got to clear it out with the Enter key.
– So that is LE, space, TMS.
– C/S key. – C/S, oh, “city/state”. Got it, okay.
– Some pieces are buffered and some pieces are live.
That live mail will be sitting on a conveyor belt,
going around the machine.
Those flats and those packages, if we don’t key them in 90 seconds
and get the information back to that machine,
it dumps it off into a reject bin
and somebody has to hand sort that.
So for the letters, they run them through the machine one time,
they store them off to the side.
A few hours later,
they put them in the machine the second time,
anticipating that we’ve processed everything that was missing.
– Oh. I’m guessing that’s, like… “Dixon Hill”? Or something like that.
I can’t tell if that’s DIC or DX…
– Unless it was… – Ox Hill! It’s Ox Hill.
– Remember, these are the easiest ones to read on the test.
– I realized that.
– Get ready.
Every year, we have fewer and fewer pieces.
We’ve got fewer people handwriting mail,
and we have OCR technology that’s really good.
I’ll bet we went from somewhere in the neighborhood
of less than half of the mail being read by a computer in the ’90s
to nearly 99% of letters being read today.
The OCRs that are reading the letter mail, I think it’s close to 99%.
So we get the 1% of the junk,
and we improve about half of that 1%. – Right.
– The job here may be getting harder because pieces are harder to read.
Because if the computer is so good and can’t read it,
it’s a lot more likely,
it’s not a good piece of mail to begin with.
The address was destroyed somehow,
or the customer had bad information for the address.
– So they’ve put “Ft” Lauderdale in the abbreviation.
So I would put F-T, no matter what? Not F-O-R?
I don’t autocorrect that. – Correct. Do it how the customer did it.
We’re not sleuths. We’re not picking out details.
You key by the rules,
you type in the address information from what you can clearly see.
Now, the street address works similarly,
where you’ll do all the numbers, and then 3+1.
– “Ignore the direction word and key the street name.
“If the direction word is the only word, key it.”
Okay. 3+1 rule.
So, Go. All right.
Ah, the… I looked down to find where the numbers are!
720WES, no, yeah, that’s West Boulevard.
So you include that and you hit Street.
400 Little Way Street.
– We have internet service from three different providers.
We have three fiber optic lines coming into the building at different points.
So if any one point gets chopped off, we can maintain service,
because we’re the only REC,
connected to every plant in the country, over 300 processing plants,
including Guam and Anchorage, Alaska; and Juneau, Alaska;
and San Juan, Puerto Rico.
It’s asking for numeric, and it’s in C coding,
so it wants the zip code.
– It wants the zip code, 35804, go.
– So if both the machine can’t read it
and a keyer isn’t able to decipher it here,
the machine at the plant sends it to the reject bin,
and then somebody will manually handle it.
And then, if they can’t figure it out,
it would either get returned to the sender.
And if there’s no return-to-sender information,
then it will go to the Mail Recovery Center.
– Okay, final test.
Here we go.
We don’t have a numeric.
So we hit None.
– Then, it’s outward, which is MOK, space,
HI, C/S, 2545HEAD… Street.
– Nice. – Okay.
I mean, what speed would I have to go here?
– 7,150 keystrokes per hour.
– It’s the speed with which you have to read that, parse it…
It’s much tougher than I thought it was going to be
to keep everything in your head.
And this is easy handwriting.
Nothing here has been blurred.
– And you haven’t even gotten to the ones
that are upside down or backwards, right,
or packages that have four sides to them.
– And it would be an average of four seconds
for each one to keep up the pace.
– On average, it’s about four seconds
that somebody’s got one of these on their screen.
– And you did an incredible job
for just learning these rules right now.
– All right, thank you. I’ll take that!
Thank you so much.
Share This- How the US Postal Service reads terrible handwriting