AI & Workflow Automation

Exception handling. The part of AP no one budgets time for.

Exception handling is where accounts payable really slows down. Missing POs, freight invoices with split tax, approval delays, supplier changes. None of this shows up in the sales demos, but it’s what bookkeepers deal with every day. This article breaks down the most common AP exceptions, how long they actually take to resolve, and why they’re so mentally draining. It looks at where traditional invoice automation tools struggle, why bookkeeping AI matters most when things don’t line up, and what better exception handling really looks like in practice. If you work with e-commerce, wholesale, construction, or any business with messy invoices, this one will feel familiar.

22 January 2026

If accounts payable were honest, exception handling would be the job. Everything else is just admin around it.

Because straight-through invoices? They’re easy. They flow. They behave. They barely need you.
But exceptions. That’s where time disappears. That’s where bookkeepers earn their keep. And that’s where most invoice automation tools quietly tap out.

You know the feeling. An invoice hits the inbox and something’s off. The total doesn’t match. The tax looks wrong. The supplier changed the layout again. Or worse, it’s freight. Multiple accounts. Multiple tax treatments. No PO in sight.

So you pause. You investigate. You message someone. You wait.

Multiply that by 50 invoices a week and suddenly your “automation” tool is just a document inbox with extra steps.

Let’s talk honestly about exception handling. The common ones. How long they actually take. And why they’re so hard to resolve without the right systems backing you up.

What we really mean by “exception handling”

Exception handling is everything that stops an invoice from being processed cleanly.

It’s not just errors. It’s uncertainty.

Any time a human has to stop and think, ask, check, or chase. That’s an exception.

Most bookkeeping teams don’t track this time. They just feel it. The drag. The backlog. The late nights.

And here’s the kicker. Exceptions don’t arrive evenly. They cluster. They spike at month-end. They show up when you’re already busy.

That’s why they hurt.

The most common AP exceptions (and why they keep showing up)

Let’s break down the ones that actually cause pain. Not the theoretical edge cases. The real stuff you see every week.

1. Missing or mismatched purchase orders

This one never dies.

The invoice says PO 4567. The PO says a different amount. Or there’s no PO at all. Or it exists but no one can find it.

Now you’re checking emails. ERP. Accounting system. Slack. Someone’s inbox.

Typical resolution time:
5 to 20 minutes per invoice. Longer if the approver is offline.

Why it’s hard:
PO data usually lives somewhere else. Most invoice automation tools extract the invoice but don’t actually reason about the mismatch.

So the bookkeeper becomes the detective.

2. Freight invoices with multiple accounts and tax treatments

Ask any bookkeeper what invoice type they dread. Freight always comes up.

Why? Because freight invoices almost never code cleanly.

One invoice.
Multiple shipments.
Some GST-free. Some taxable.
Some to COGS. Some to clearing. Some to landed costs.

OCR might grab the total. But it won’t understand the split.

Typical resolution time:
10 to 30 minutes per invoice. Sometimes longer.

Why it’s hard:
This requires context. What was shipped? Where? Under what incoterms? Is GST applied to international legs? Most tools weren’t built with physical goods logic in mind.

This is where bookkeeping AI either shines or completely falls over.

3. Incorrect or inconsistent tax coding

Tax errors don’t always look like errors.

The invoice “looks” right. But GST is applied where it shouldn’t be. Or missing where it should be. Or split across lines inconsistently.

If you miss it, the BAS pays the price later.

Typical resolution time:
5 to 15 minutes per invoice.

Why it’s hard:
Tax rules depend on supplier location, goods type, service type, and sometimes history. Simple automation can’t infer that. Humans can. But humans are slow.

4. Supplier invoice format changes

This one is sneaky.

A supplier you’ve processed hundreds of times suddenly tweaks their layout. New columns. New descriptions. New totals position.

Now the extraction breaks. Fields map incorrectly. Line items merge.

Typical resolution time:
3 to 10 minutes. Unless it cascades across dozens of invoices.

Why it’s hard:
OCR systems rely on patterns. Suppliers don’t care about your patterns. And most tools don’t adapt quickly without manual intervention.

5. Duplicate invoices and credit confusion

You’ve seen this one. Same invoice number. Slightly different total. Or a credit note that looks like a bill.

Is it a duplicate? A correction? A partial credit?

You don’t want to double-pay. But you also don’t want to block real bills.

Typical resolution time:
5 to 15 minutes.

Why it’s hard:
Duplicate detection is fuzzy. Especially when suppliers reuse invoice numbers or issue adjustments mid-month.

6. Approval bottlenecks

The invoice is fine. The coding is fine. The system is fine.

But the approver is busy. Or on leave. Or just slow.

So the invoice sits.

Typical resolution time:
Anywhere from a day to two weeks.

Why it’s hard:
This isn’t a data problem. It’s a workflow problem. And most tools stop at “sent for approval” and call it done.

The hidden cost no one calculates

Let’s do some uncomfortable maths.

Say a bookkeeper processes 300 invoices a month.
If even 30 percent are exceptions, that’s 90 invoices.

If the average exception takes 10 minutes, that’s 900 minutes.
15 hours. Every month.

That’s almost two full workdays spent not on bookkeeping, but on fixing broken flows.

And that’s conservative.

Now scale that across a practice. Or an e-commerce business with freight-heavy operations. Or a wholesaler juggling multiple entities.

Suddenly exception handling isn’t a side task. It’s the workload.

Why most invoice automation tools struggle here

Tools like Dext and Hubdoc do a solid job capturing documents. They were built for that era. Get the bill in. Extract the basics. Push it to Xero or MYOB.

But exception handling isn’t about capture. It’s about judgment.

It’s about understanding why something doesn’t line up and what to do next.

Most tools follow a simple path:

  • Extract

  • Push

  • Let the human fix the rest

Which means the messiest invoices still land squarely on the bookkeeper’s desk.

That’s not a failure of effort. It’s a limitation of design.

Why exception handling feels so mentally draining

Here’s the part people don’t talk about.

Exception handling isn’t just time-consuming. It’s cognitively expensive.

Each exception forces context switching. You go from routine processing to problem solving. From autopilot to focus mode.

You have to remember supplier history. Tax rules. Past decisions. Client preferences.

And because exceptions are unpredictable, you can’t batch them cleanly.

That’s why even a few exceptions can make a day feel heavy.

What better exception handling actually looks like

Good exception handling doesn’t mean eliminating exceptions. That’s unrealistic.

It means:

  • Catching them early

  • Classifying them correctly

  • Giving the bookkeeper context, not just data

  • Reducing the thinking required to resolve them

This is where accounting AI starts to matter.

Not just OCR. Not just extraction. But systems that understand confidence, patterns, and risk.

For example:

  • Flagging a freight invoice because it doesn’t match known freight patterns

  • Suggesting likely account splits based on prior invoices

  • Highlighting tax treatments that differ from historical behaviour

  • Escalating only the truly uncertain cases for review

When that happens, the bookkeeper isn’t fixing everything. They’re supervising.

That’s a very different job.

Why exception handling is the real buying decision

When bookkeepers say “this tool saved me time,” they’re rarely talking about straight-through invoices.

They’re talking about exceptions.

How fast they could resolve them. How little thinking was required. How few emails they had to send.

That’s why demos that only show clean invoices miss the point.

Show me the ugly ones. The freight bills. The messy supplier PDFs. The multi-entity invoices.

That’s where trust is built.

A quiet shift happening right now

More bookkeeping firms are starting to evaluate tools differently.

Not “does it capture invoices?”
But “how does it behave when things go wrong?”

Because things always go wrong.

And as invoice volumes increase, especially for physical-goods businesses, exception handling becomes the constraint.

Not headcount. Not clients. But cognitive load.

That’s why bookkeeping AI is moving beyond capture and into reasoning. Slowly. Imperfectly. But inevitably.

The takeaway

Exception handling is not a corner case. It’s the core of accounts payable.

Most invoices are easy. Most time is not spent on them.

The hard invoices are the ones that matter. They’re the ones that shape your day, your workload, and your stress levels.

If you’re evaluating invoice automation, don’t ask how many invoices it can process.

Ask how it handles the ones that don’t behave.

That answer tells you everything.