The Invoice Processing Problem Is Mostly About Not Processing Invoices

Companies receive invoices. They need to process them. This should not require three people and two weeks.

Vendors send invoices via email. Or regular mail. Sometimes fax, because apparently it’s still 1997 in accounts payable. You want the invoice recorded in your accounting system. Payment scheduled. Approvals obtained. Instead the invoice sits somewhere. In someone’s inbox. On someone’s desk. In a filing cabinet labeled “To Process.”

Probably when you first started in accounts payable, someone showed you the process. Print the invoice. Or forward it to processing. Enter data manually into NetSuite or QuickBooks or SAP. Route for approval based on amount and department. Wait for approval. Schedule payment. File the invoice. Mark it as processed. Simple enough.

After you’ve been doing this for a while, you realize the problems. Manual data entry has a 15-20% error rate. Wrong amounts. Misspelled vendor names. Incorrect GL codes. Each error requires correction. Which requires finding the original invoice. Comparing it to the entry. Making changes. More time wasted.

Another thing that surprisingly many companies do is just let invoices pile up? You can see the temptation. Instead of processing 50 invoices today, you process the urgent ones. Deal with the rest later. This creates an obvious problem. Eventually you have to process all the invoices. There is no actual “later” where processing becomes easier. So probably in two weeks you’ll have 200 invoices. And less time to process them. It compounds.

The Manual Processing Trap

One amusing irony? Invoice processing solutions talent is artificially scarce. Perhaps there are 10,000 people who are genuinely good at accounts payable in the United States. How many are processing invoices at any given moment? On some plausible assumptions the answer might be 6,000. The rest are in meetings. Or on vacation. Or quit last month.

Invoices get processed by whoever is available. According to whatever method makes sense to that person. At that moment. With no consistent data entry standards. Or approval workflows that make sense.

In a world of scarce invoice processing talent, processing times go up. The average company takes 15-30 days to process an invoice. That’s probably not everyone’s experience. But still. A meaningful chunk of invoices are sitting unprocessed in any given month.

If you’re running accounts payable, is this optimal? Like, “I will process maybe 60% of invoices on time, and when I do process them there’s a 15% chance the data is wrong”? I feel like when I put it like that it sounds not great?

The basic problem is volume meets manual labor. When you receive 1,000 invoices monthly, manual processing requires significant staff. Each invoice needs opening. Reading. Data extraction. Entry into your ERP system—Oracle,Microsoft Dynamics, or Workday. Validation against purchase orders. Routing for approval. Following up on approvals. Scheduling payments. Filing documentation. Maybe 10-15 minutes per invoice if everything goes smoothly. Which it doesn’t.

The Fake Automation Problem

At a 2023 procurement conference, someone from a major enterprise software company said something interesting. Many companies think they’ve automated invoice processing. They haven’t. They’ve just moved the manual work to different people. According to a source who attended.

If you’re trying to automate invoice processing, is buying invoice management software sufficient? The answer is no! Obviously not! Most “automated” invoice processing solutions still require humans to review every invoice. Confirm extracted data is correct. Fix errors. Make exceptions. Route for approval manually.

The software provides a nicer interface. Better reporting. Audit trails for Sarbanes-Oxley compliance. These are good things. But they’re not automation. They’re digitization of manual processes. Which is different.

Much like email 20 years ago, invoice processing solutions have reached a maturation point. Where basic digitization is standard. When you scan an invoice, the system captures an image. Stores it in a database. Lets you search for it later. This is better than paper filing cabinets. Obviously. But it’s not intelligent document processing.

I have sometimes mused about why companies accept this. Invoice processing solutions software, I sometimes write, is like hiring someone to organize your filing cabinet. They create nice labels. Arrange everything alphabetically. But when you need to find something, you still have to search through the cabinet yourself. Things like “automatically extract invoice data” or “match to purchase orders without human review” are best understood not as “ooh the new systems are so much smarter.” But rather “companies got used to manual processing and assumed it was inevitable.”

What Actual Automation Looks Like

Twenty years ago, companies cared about invoice processing speed. Much better to pay invoices quickly than slowly. But they also just hired more people to process them. Hoped for the best.

They thought about “the value of fast processing.” If you processed invoices in 5 days instead of 30, how much was attributable to having good staff? How much was attributable to simpler invoices?

Shouldn’t AI help here? If invoice volumes have reached a critical point where manual processing doesn’t scale? Processing is so time-consuming that you miss early payment discounts? The market should create better automation tools. Part of the thesis is that data extraction can be fully automated. Or at least dramatically improved through machine learning.

Organizations implementing AI-powered invoice processing solutions report 70-90% time savings. In processing hours. Invoices that used to take 15 minutes now take 2 minutes. Data extraction that required manual keying now happens automatically. Error rates have improved from maybe 15% with manual entry. To under 5% with AI-powered OCR and natural language processing.

The technology works like this: AI reads the invoice using computer vision. Identifies document type—invoice versus purchase order versus receipt. Extracts vendor name, invoice number, date, line items, amounts, tax, payment terms. Validates extracted data against expected formats. Matches invoice to purchase order in your procurement system. Identifies discrepancies automatically. Routes for approval based on dollar amount and general ledger codes. Sends to appropriate manager via email or workflow system. Tracks approval status. Posts to accounting system when approved. Schedules payment according to terms. All without human data entry.

Another benefit is fraud detection. Modern invoice processing solutions can identify duplicate invoices automatically. Flag suspicious patterns like round-number amounts or unusual vendor activity. Detect invoices that don’t match purchase orders. Compare vendor banking details to historical records. This is particularly useful for companies concerned about accounts payable fraud. Where the cost of a single fraudulent payment can be substantial.

Looking Ahead

I’m generally skeptical of claims that AI will eliminate accounts payable departments. That companies cannot process invoices without sophisticated neural networks. That only Fortune 500 companies with massive IT budgets can implement intelligent invoice processing solutions.

Plenty of small and mid-sized businesses exist. Whose basic needs can be met with straightforward OCR and data extraction. Most organizations don’t need superintelligent invoice AI. They just need invoice processing solutions that can read an invoice. Put the data in QuickBooks or Xero.

Technology has gotten quite accessible. Cloud-based invoice processing solutions from companies like Bill.com,AvidXchange, and Tipalti offer rapid deployment. Implementation that used to require months now takes weeks. Pricing models have evolved from enterprise-only. To per-invoice fees that work for smaller organizations.

AI invoice processing solutions are prioritizing practical automation. Immediate cost reduction. Over long-term fundamental research into invoice theory. Companies are “under growing pressure to show ROI.” Their investment in accounts payable automation needs to pay off. Reduce processing costs. Tinkering with theoretical invoice processing systems for years won’t reduce costs today.

Perhaps a little odd that companies feel this pressure? Invoice processing has a long history of being manual drudgery. That nobody enjoys but everyone accepts. Still, when you get big enough, you care about efficiency. When you’re processing 10,000 invoices monthly at $10-15 per invoice in labor costs? It helps if you can reduce that.

The Bottom Line

If you gave me 1,000 invoices to process? I promised to enter them all accurately into your accounting system? In the meantime I just stacked them on a desk and hoped someone would get to them eventually? You’d be like “that’s not invoice processing!”

Using a manual process with spreadsheets and data entry? That was normal in 2005. An invoice management system with scanning and workflow? Better. Though there is still too much manual work.

An AI-powered invoice processing solution that automatically reads and processes everything? Significantly better. Artificial intelligence can extract data. Validate accuracy. Route for approval. High probability your invoices get processed correctly and quickly.

Maybe it’s expensive. These invoice processing solutions cost money to implement. Seems like a necessary investment though. Better than paying staff to manually enter data from 1,000 invoices monthly.

What if you didn’t automate? Just expected accounts payable staff to process faster? Well, same problem. People have limits. Processing times stay long. Errors stay high.

But you might have an answer. “Invoice processing solutions themselves aren’t very stable.” Complexity and integration challenges constantly eat away at their value. The only way to ensure accuracy is having humans check everything.

You might reasonably reply: “That doesn’t scale.” When invoice volumes double you need to double staff. Manual checking doesn’t eliminate errors.

This objection strikes me as correct. Though some CFOs will disagree.

Historically most organizations have manual invoice processing. With various levels of digitization. Hoping accuracy improves through training and effort.

The question is no longer whether AI will improve invoice processing. But how quickly your organization will implement it. The gap between companies with automated invoice processing solutions will widen. Those still using manual data entry will fall behind. Organizations embracing AI-powered invoice processing solutions today? Will be best positioned to actually process their invoices tomorrow.