Why is finance transformation stalling at source-to-pay exceptions?

Most finance teams have done the hard work. They've digitized the standard requisition-to-purchase order (PO) and invoice-to-payment paths. The routine transactions flow and the data moves. The systems do exactly what they were built to do.

But somewhere in that process, a gap opened up that automation didn't touch, and that's where the real costs hide. Exceptions are where this breaks down: these are the cases your systems were never designed to handle.

To close this gap, 84% of finance functions are now investing in AI, hoping to break through on cost and efficiency. Only 7% say those investments are delivering high or very high impact.

The reason so much of that investment stalls isn't that AI can't spot problems. It's that spotting a problem and understanding it aren't the same thing. Without a shared definition of what your data actually means, AI doesn't fail loudly—it fails confidently. It hands back a plausible-looking answer, and no one questions it until the audit.

The exception trap

An invoice arrives that doesn't match the purchase order (PO). A requisition comes through without the right approval structure. A supplier changes their payment details mid-contract. A requester gets frustrated with approvals and processes a payment outside the system just to move forward.

This surrounding work is where the real cost lives, and the scale of it is bigger than most finance and procurement leaders expect:

• Only 7% of organizations describe procure-to-pay as nearly or fully automated

• If you consider the full range of activities, processing a single purchase order can cost anywhere from about $14 to more than $54

• For every $1 million in revenue, accounts payable processing costs range from $380 for top-performing operations to $920 for average performers, representing more than $500,000 in potential annual savings for a company generating $1 billion in annual revenue

• Procurement organizations that reach "Digital World Class" performance report 19% lower functional cost, fewer full-time equivalents (FTEs) tied up in manual work, and 58% shorter requisition-to-PO cycles than their peers

Automating the standard transaction path frees teams from much of the routine work, giving them more capacity to resolve exceptions. But when transformation roadmaps overlook exception handling, they leave significant improvement potential unrealized. Exceptions stay manual and fragmented, and that's exactly where cost concentrates and delays pile up. To transform source-to-pay, exceptions are where you have to look.

Why AI alone hits a wall

AI is good at spotting patterns in clean data. It can flag a duplicate invoice and predict which invoices will be late with high accuracy. But when an exception surfaces (such as incomplete information, a policy violation, or a missing signature), AI can tell you there's a problem, but it can't actually resolve the work around it.

So, a finance team gets an alert from their AI system, but they still have to:

• Track down the right person in the approval chain

• Pull context from three different systems

• Manually coordinate the fix

• Make sure nothing falls through the cracks

AI adds one more data point to a pile of manual work someone still has to get through.

There's a second problem that is less visible, but more risky.

Even when AI is wired into your systems and can act on an exception, the action is only as reliable as the AI's understanding of what it's looking at. Take a simple example: which amount field reflects the reporting currency, or which receipts count toward a cycle time calculation. Get that wrong, and AI doesn't hesitate. It acts on the wrong answer with the same confidence it would show on the right one, and the override goes through unchallenged.

In a dashboard, that's an embarrassing correction. In a payment instruction or an approval threshold check, it's a compliance failure or a wrongly released payment—the kind of error that surfaces at audit, not before.

The systems that close this gap don't just orchestrate work. They anchor every action in a shared semantic layer that defines what each object means across your systems. It enforces your governance rules at the moment of decision, and produces a machine-readable evidence record every time an exception is resolved. That's what makes AI genuinely trustworthy in finance.

Designing for exceptions

Today, when an invoice exception surfaces it can take up to four days just to get resolved: 24 to 48 hours to route the exception to the job site, then another 24 to 48 hours to process it once it's approved. That's before counting the four or five people who typically touch the invoice earlier in the process, or the up to five-day lag from offshore capture.

A better workflow would collapse that timeline. It would automatically retrieve the relevant records: the PO and payment terms for an invoice, and the supplier contract for a request. It would check those records against your existing policies and preferred suppliers. Where it can resolve the exception immediately (flagged against clear policy), it does. Where it can't, it routes the decision to the right person with all the context attached (not an alert, but the actual comparison or contract mismatch they need to decide).

Anything requiring human judgment still gets routed to where people work—Microsoft Teams, Slack, email.

And because every resolution is logged, patterns become visible. Orchestration is what makes that action possible: it's the layer that connects your systems, governs the exceptions, and keeps work moving across teams.

Building a process that actually handles exceptions

You don't need to rip out your systems: the orchestration layer described above works alongside your enterprise resource planning (ERP) and procurement system, not instead of them.

Some finance teams involve IT to build this in-house. It’s possible, but often slow and resource intensive. It means building around challenges that continue to plague most enterprises: how to route work across approval chains without bottlenecks, and how to handle exceptions at scale without adding headcount.

If you're evaluating an approach, whether building it internally or buying a platform, the questions worth asking fall into four areas: how much of the exception it actually handles, how well it integrates with what you already run, whether it enforces governance once people get involved, and how fast you can realistically get it live.

Specifically:

• Does it handle the full exception lifecycle or just flag problems?

• Can it retrieve records from multiple systems in real time, without manual rekeying?

• Does it enforce your governance rules even when people get involved in decisions?

• Can it learn from past executions and self-improve to cover more exception types?

• Is it purpose-built for source-to-pay workflows, with deep expertise in procurement and finance processes? Or is it a generic workflow tool that you'd have to configure from scratch?

• How long will it actually take to implement (weeks or months)?

The answers to those questions matter more than whether you build or buy.

A solution purpose-built for finance exception management will understand the nuances of your workflows, the governance rules that matter most, and the specific places where exceptions tend to surface. It will also come pre-configured with best practices from best-in-class industry use cases.

For the real payoff, start in procurement and accounts payable

If you're going to close the gap, start in procurement (from request intake to PO) and accounts payable (invoice processing). Those are where exceptions happen most frequently and where manual work is highest. You'll also see ROI fastest in those areas: fewer missed discounts, faster cycle times, fewer duplicate payments, and less time your team spends chasing approvals.

Address the exception work itself and you’ll build a source-to-pay process that's truly end-to-end: one that handles not just the standard transaction, but the exceptions, approvals, and handoffs that stop your team from focusing on the work that matters most.

Explore the UiPath Solution for Source-to-Pay.

Sources:

• APQC, "How Efficient Is Your Procurement Process? Benchmarks Reveal a Wide Performance Gap," April 29, 2026.

• APQC, "How Organizations Can Reduce Accounts Payable Costs," March 16, 2026.

• APQC, "Procure-to-Pay Automation: Moving Finance From Transactional to Intelligent," March 26, 2026.

• Gartner, "Gartner Says CFOs Need Structured Finance AI Roadmaps," June 8, 2026.

• The Hackett Group, "Achieving Digital World Class Procurement Excellence Through Gen AI," November 7, 2025.