Sixty years of AP technology, and the hardest invoice in your queue looks a lot like the hardest one in 1965.

It doesn’t match the PO. Or the receipt. Someone has to figure out which of the three documents is wrong, and that someone is almost always on your team.

Every wave of AP automation attacked one of two problems. How the invoice travels: mail, EDI, scanners, portals, government platforms. Or how it gets read: keying, OCR, machine learning.

Both problems are now mostly solved. The exception isn’t.

Here’s the history, era by era. For each one, I’ll cover what it fixed and what it left behind. The leftovers are the useful part. They’re where your exception queue came from.

Era 1: The voucher package (pre-1970)

The invoice came in the mail. A clerk pulled the PO copy and the receiving report, stapled all three together, checked the math by hand, coded it in pencil, and walked it over for a signature. Checks were typed.

What it fixed: the control model. Three documents from three independent parties, matched by a fourth person. Segregation of duties and an audit trail a bank examiner could follow.

Every AP system since has been an attempt to reproduce that control, only faster.

What it left behind: cycle times measured in weeks. Early-pay discounts that nobody could realistically capture. And no cost-per-invoice number, because nobody could have calculated one.

Era 2: Batch on the mainframe (1970s)

AP became a sub-ledger. Invoices were keyed in batches, posted overnight, and printed back out on green-bar paper. The aging report was a stack on your desk Monday morning.

What it fixed: arithmetic, posting and the trial balance. The ledger became a file you could query and reconcile.

What it left behind: data entry became the bottleneck and the headcount. The invoice was still paper. The system held a transcription of it, so a keying error and a real dispute looked exactly the same from the inside.

Era 3: EDI, the invoice becomes data (1979 to 1980s)

ANSI chartered the X12 committee in 1979 and documents got numbers: 850 for the PO, 856 for the ship notice, 810 for the invoice, 820 for remittance. UN/EDIFACT followed in 1987 for international trade.

What it fixed: transport and transcription in one move. For an EDI supplier, the invoice arrived as structured data. No mail, no keying, and matching could run the moment it landed.

What it left behind: a mapping project for every trading partner. So only your biggest suppliers ever joined.

That created the two-speed AP department most companies still run. A small automated channel carries most of the spend. A long tail of suppliers on paper carries most of the documents, and most of the work.

Era 4: Imaging, OCR and capture (1985 to 1990s)

The mailroom became a scanning operation. Invoices were imaged on arrival, header fields were read automatically, and anything below the confidence threshold dropped to a validation screen.

What it fixed: paper stopped being the record. Anyone in any building could pull up the image, and nobody had to type supplier, invoice number, date and total anymore.

What it left behind: a template for every supplier layout, and a person to maintain them. Line-level extraction was poor, so line-level matching stayed manual.

And validation became a permanent job. Capture didn’t remove the keystrokes. It moved them all into one queue where you could finally count them.

Era 5: Digital workflow (late 1980s to 1990s)

Imaging got routing. The image went to a coder, then an approver, then a second approver over threshold, with every step stamped. For the first time, someone could tell you where an invoice was without walking to a desk.

What it fixed: approval routing, escalation and an electronic audit trail. Lost invoices mostly stopped being a thing.

What it left behind: workflow automated the movement of a decision, not the decision itself. Invoices now sat in an approver’s inbox instead of a tray, and AP inherited the chasing.

“Where’s my invoice?” became a supplier support function. Your team staffed it. The supplier still couldn’t see the answer for themselves.

Era 6: ERP and the three-way match (1990s to 2000s)

SAP R/3 landed in 1992, with Oracle and the other suites close behind. The match stopped being a clerk’s judgment and became a configured rule. PO, goods receipt and invoice reconciled against GR/IR. Out-of-tolerance invoices were blocked until someone released them.

What it fixed: control moved upstream and became measurable. PO compliance, tolerance policy and blocked-invoice reporting all became numbers you could hold someone to.

What it left behind: the finding every AP team eventually makes. Match failures are overwhelmingly data problems, not supplier misconduct.

Unit-of-measure mismatches. Receipts that were partial or never posted. Freight lines with no PO line. Price changes agreed on the phone and never amended. The supplier quoting the wrong PO.

The three-way match industrialized finding these. It automated none of the fixing. So first-pass match rates climbed fast, then flattened, and what’s left over is the expensive stuff.

Era 7: Networks, portals and the supplier side (1996 to 2000s)

B2B networks gave suppliers without EDI a place to flip a PO into an invoice, submit it without paper, and check status themselves. Later came early-pay discounts on a sliding scale. AP finally had a channel for the long tail.

What it fixed: self-service. Status visibility moved to the party who actually wanted it. That’s the only thing that has ever durably cut supplier inquiry volume.

What it left behind: adoption became the whole game, and adoption isn’t a technical problem.

A portal your supplier won’t log into is a paper process with extra steps. A network fee charged to the supplier turns your AP project into their cost center. Multiply that by every customer a mid-sized supplier serves and you get portal fatigue, the thing every AP team now negotiates against.

Era 8: The tax authority joins the transaction (2003 to 2019)

Chile launched e-invoicing in 2003, the first country to do it. Brazil and Mexico followed in the mid-2000s. The model is clearance: the tax authority validates the invoice before or as it reaches the buyer. If it hasn’t cleared, it isn’t an invoice. Italy brought mandatory B2B e-invoicing to Europe in January 2019.

What it fixed: format and validity, by force. One structured format, one route, and no argument about whether the invoice was received.

What it left behind: compliance became country by country, and AP picked up a tax obligation it never asked for and wasn’t staffed for. Also, a cleared invoice is a legally valid invoice. It isn’t necessarily a correct one. Clearance says nothing about whether it matches your PO.

Era 9: Mandates become the default (2019 to 2030)

Europe is converting. Germany required every domestic business to be able to receive structured e-invoices starting January 1, 2025, with issuing phased in through 2028. Poland’s KSeF went mandatory in February 2026 for the largest taxpayers and April 2026 for everyone else. France went live on September 1, 2026, with smaller companies following in 2027.

Above all of it sits ViDA, the EU’s VAT in the Digital Age package, adopted in March 2025. It brings digital reporting for intra-EU B2B trade starting July 1, 2030.

What it fixes: the supplier adoption problem portals never cracked. Structured invoicing stops being something you talk a supplier into. It becomes something the law requires.

What it leaves behind: your AP process is now a regulated process in a growing list of countries, and legislatures are setting part of your roadmap.

And the big confusion of this era: format compliance is not process automation. A compliant e-invoice that fails the three-way match is still an exception. It just showed up faster, in XML.

Era 10: Machine learning and touchless (2010s)

Back up a few years, because while mandates were building, capture changed too. Extraction stopped depending on templates. Models learned supplier layouts from correction history, line items became readable, and non-PO invoices got coded by prediction.

The headline metric shifted from capture accuracy to touchless rate: the share of invoices that post without a human opening them.

What it fixed: template maintenance, and the cold start on new suppliers. Reading an invoice stopped being the hard part of AP.

What it left behind: accuracy without judgment. The model reads the invoice perfectly and still can’t tell you why it doesn’t match the PO, whether the price variance was approved, or which of three people can sign off.

Touchless rates rose on the clean invoices. The clean invoices were never the cost. The exception queue barely moved, and it’s the same queue the three-way match created in era 6.

Era 11: Agents aimed at the exception (2024 to now)

Language models can read a document without being trained on its layout. More important, they can reason across the invoice, the PO, the receipt, the contract and the email thread that explains the variance.

So the pitch has changed. It isn’t capture anymore. It’s working the queue: figuring out why a match failed, chasing the unposted receipt, drafting the supplier reply, proposing the coding, and building the case for a human to approve.

What’s actually new: this is the first wave aimed at judgment instead of transport or transcription. Every earlier era moved the invoice faster or read it better, then handed the exception to a person unchanged.

Read this era skeptically. Claims are running ahead of deployments. And the open question isn’t capability. It’s control.

Who approves? What’s auditable? What is an agent allowed to do unsupervised? How do you catch a wrong resolution and reverse it?

Era 1 built a control model that every era since had to reproduce. This one hasn’t reproduced it yet. Any vendor who can’t answer those four questions in plain language isn’t ready for your AP team.

Which era is your process actually in?

Your software’s era and your process’s era are two different things. The signal isn’t what your platform is called. It’s where the work still sits.

• Suppliers email invoices to a shared inbox and someone downloads attachments. You’re in era 4. The channel is email instead of mail, but capture is still the front door.

• Someone’s real job is chasing approvers. That’s era 5 residue. Workflow moved the decision into a queue and left the chasing with AP.

• Your first-pass match rate jumped, then stalled for years. Era 6 residue. What’s left are data and process problems upstream of AP. No matching engine fixes those.

• Your top supplier question is still “has my invoice been approved?” Era 7 never finished. Status visibility exists, but the suppliers making the calls aren’t in the channel that shows it.

• Touchless rate is strong but headcount hasn’t moved. Era 10 residue. You automated the clean invoices. The cost was always in the messy ones.

• A country deadline is driving your roadmap instead of a business case. You’re in era 9. Keep the compliance project and the automation project separate. They have different success measures and shouldn’t share a budget line.

The through-line

Read the eras in order and the pattern is hard to miss. Sixty years of investment went into how the invoice arrives and how it gets read. Both are now largely solved. Arrival is structured, and in more countries every year, mandatory. Reading is a solved machine learning problem.

What’s been carried forward untouched since era 1 is the exception. The invoice that doesn’t agree with the PO or the receipt, and the human judgment it takes to figure out which document is wrong.

Every era found exceptions faster. None of them resolved more.

So here’s the test I’d apply to anything pitched to your AP team from here on. Does it move the invoice, read the invoice, or close the exception?

If your team is still living in the exception queue, email me at dshore@accountspayable.institute.