Your AP Team Isn’t as Automated as You Think.

By Shaun Leisegang MAICD Chief Automation, Data and AI Officer at Tecala | Follow me on LinkedIn

Accounts Payable is one of those processes almost every organisation has already tried to fix. 

Most have bought something. A scanning tool. An OCR engine. A digital mailroom. 

And most of them are still doing the work by hand. 

I've written before about the Old Way of AP: invoices arriving by email, someone downloading and renaming files, data keyed in manually, approvals crawling between inboxes, month-end turning into a scramble. That version of the problem is well understood. 

What I want to talk about here is the second version. The one that shows up after you've automated. Because a partly automated AP process has its own failure mode, and in my experience, it is the more frustrating of the two, precisely because the organisation believes the problem has been solved. 

The paper disappears. The work doesn't. 

Here's what most finance teams are still doing after they've automated: 

🔶 Validating supplier details against master data by hand 

🔶 Verifying ABN and GST registration 

🔶 Chasing approvers who have let an invoice sit for days 

🔶 Investigating duplicates, missing POs and changed bank details one at a time 

🔶 Re-keying the extracted data into Xero, MYOB, NetSuite or Dynamics 

🔶 Guessing at where invoices are getting stuck, because reporting is retrospective 

The scanner solved the filing. It did not solve the process. 

Extraction was always the easy part.

This is the part I think most AP automation projects get backwards. 

Reading an invoice was never the hard bit. Pulling a supplier name, an amount and a date off a PDF is a solved problem, and it is roughly ten per cent of the job. It is also, almost always, the only part the tool actually does. 

The other ninety per cent is where the cost and the risk live. Validating against your business rules. Matching to a purchase order. Catching the duplicate, the unfamiliar supplier, the bank detail that changed last week. Routing to the right approver under the right threshold. Posting to the right GL account. That work does not disappear when you buy a capture tool. It gets handed straight back to your team the moment the data has been read. 

There is a number that makes this visible, and I'd encourage every finance leader to go and find theirs. 

What straight through processing actually measures.

Straight Through Processing is the percentage of invoices that go from receipt to posting without a person touching them. Manual AP runs at about five per cent. Basic automation, the scan-and-extract tooling most mid-market organisations already own, plateaus somewhere between twenty and forty. Which means that after the project finished and the invoice was declared automated, six to eight out of every ten still need a human to finish the job. 

That isn't automation. It is a faster inbox. 

The Real Cost of Manual vs Automated AP,

The cost sits underneath it. Manual processing runs $12 to $30 an invoice against under $3 when it is genuinely automated. A manual cycle takes 10 to 15 days where an automated one takes 2 to 4, which is the difference between capturing early-payment discounts and missing them. Manual data entry carries a 1 to 4 per cent error rate. Around 60 per cent of finance team time goes to transactional AP work. None of that appears as a line item. It appears as a finance team too busy processing invoices to analyse anything

What the Frontier Way looks like in AP. 

The fix is not a better scanner. It is redesigning the process across the three types of work. 

Robots handle the routine. Collecting invoices from inboxes, portals and scanners. Routing and saving documents. Updating systems. Sending the reminders nobody enjoys sending. 

AI Agents handle the intelligent work. Reading and understanding the invoice. Validating and enriching against your rules. Predicting the GL code. Spotting the anomaly. Routing the approval by policy and escalating when it stalls. Posting into your finance system with the coding and tax treatment correct. Not one generic engine, but specialist agents each doing one job and handing off to the next. 

People handle judgement. Approving what exceeds threshold. Resolving the exceptions that genuinely need a human. Making the call on a flagged fraud risk. Managing suppliers. Setting the policies everything else runs inside. 

Done properly, that is the difference between a twenty to forty per cent ceiling and eighty to ninety-five per cent straight through processing. 

This is what we built Transformr to do. It is not a plugin. It is not a bolt-on to your ERP. It is the full invoice lifecycle, receipt through posting, run by eight specialist AI Agents inside a governed workspace, configured to your approval hierarchies and controls rather than custom built from scratch. Built and hosted in Australia. Live now, processing invoices for Australian organisations. 

And the team's work becomes more meaningful, not less. They stop processing invoices and start managing the function. 

Start with the question. 

If you take one thing from this, make it this question, and ask it of your finance team this week: 

What percentage of our invoices are processed end to end without anyone touching them? 

If the answer is under fifty per cent, the gap between what you bought and what you actually automated is measurable. So is the return on closing it. 

👉 Contact us to book a Transformr Assessment.