#533 1956 · Ken Eldredge / Stanford Research Institute (for Bank of America) · Banking / document automation
Machines couldn't read checks after a stamp or signature covered the number, so SRI printed the number in a channel stamps and signatures can't touch
the problem
Checks got stamped, endorsed and smudged before a machine could ever read the account number
background
In the early 1950s, Bank of America's checking accounts were growing by 23,000 a month, and an experienced bookkeeper could post only about 245 accounts an hour by hand — banks were closing their doors at 2 p.m. just to finish the day's manual proofing and posting. Bank of America hired Stanford Research Institute to design an automated system, and one requirement was non-negotiable: the account number on every check had to be readable by a machine fast enough to keep pace with the volume, without adding a separate step like punching the number onto a card.
The obvious path, and the one Bank of America itself first investigated, was optical character recognition — print the account number as a bar code or font and read it with a photoelectric scanner, the same approach an outside vendor had already built a prototype for. It failed on contact with how a check is actually used: by the time it reached a reader, a check had usually been stamped with cancellation and endorsement marks, folded, and handled by hand, and any of that layered optical noise could defeat a scanner reading the same spot. A pure bar code compounded the problem, since a human teller could not read one at all if a machine ever rejected it.
what everyone would do
The approach already on the table, and the one Bank of America itself first investigated buying from an outside vendor, was optical character recognition: print the account number as a bar code or font and read it with a photoelectric scanner. That was also exactly what the requirement broke on — optical readers of the era were defeated by anything printed or stamped over the number afterward, which every real check accumulates on its way through processing, and a pure optical bar code could not be read by a human teller at all if a machine ever rejected it.
what they saw
Eldredge saw that the problem wasn't reading speed, it was that an optical reader cannot distinguish the ink that identifies the account from every other mark later added to the same piece of paper — a cancellation stamp and an account number look identical to a photoelectric scanner. Switching the physical channel from light to magnetism let a reader respond only to the specifically magnetized ink, ignoring everything else printed, stamped or smudged onto the same check, while designing the character shapes to still be legible by eye solved the second problem — a machine-only code no human could check — in the same stroke.
the move
SRI engineer Ken Eldredge, heading the institute's Control Systems Laboratory, designed a numeric font printed in magnetic ink — later standardized as E-13B — shaped so it could be read both by a magnetic sensing head and by a human eye. Because the reader responded only to the specific magnetized ink pattern rather than to light reflected off the whole check, cancellation stamps, endorsement marks and creases added after printing simply didn't register, leaving the identifying number readable regardless of what else had since been stamped or written on the same paper.
why it works
Because a magnetic read head only registers the check's magnetized ink and not the light reflected off cancellation stamps, endorsement marks or creases layered on afterward, a check could pass through the full damage of ordinary handling and still yield a clean, unambiguous account number — something no optical scanner of the era could guarantee. And because the same E-13B characters were legible to a human eye, unlike a bar code, the bank traded away nothing in exchange for that immunity: a teller could still read a rejected check directly. That combination is what let the American Bankers Association standardize on it in 1958 and let per-employee processing rates jump roughly 130-fold over manual bookkeeping.
the payoff
SRI was issued U.S. Patent Number 3,000,000 for the invention (assigned to General Electric, which manufactured the check-processing hardware), and the American Bankers Association trialed the E-13B font in July 1956 and adopted it as the U.S. standard for negotiable documents in 1958. Where a manual bookkeeper processed roughly 245 accounts an hour, magnetic-ink automation lifted check processing to an initial 33,000 checks an hour per employee, eventually reaching as high as 100,000.
where it breaks
The approach only pays off when the marking medium can be produced cheaply and reliably enough to print at the scale a whole national document supply requires, and when a genuinely fragmented industry — thousands of independent banks in this case — can be coordinated onto one single standard font and ink formulation, which is why this needed the American Bankers Association to mandate it rather than Bank of America simply deploying it alone. It also assumes the document keeps circulating in physical form with its ink intact; once processing moves to purely digital images, as check imaging later did, the entire advantage of a separate physical channel disappears.
what came after
The American National Standards Institute recognized the ABA's MICR standard as the American standard for check printing in 1963, and the same E-13B magnetic-ink font remains the worldwide standard for check processing today, printed at the bottom of virtually every check issued anywhere in the world. The Federal Reserve required magnetic ink on all checks it processed starting in 1967, cementing it as the backbone of the U.S. check-clearing system for the rest of the 20th century.
references
- [1]Banking Automation: ERMASRI International, 2024sri.com
- [2]Check PaymentsFederal Reserve History (Federal Reserve System), 2023federalreservehistory.org