Accounts Payable Automation in 2026: What Touchless Actually Means

Every week, we have the same conversation with a CFO or finance controller. They say: "We already have AP automation. But our team is still handling most of the invoices manually. What are we doing wrong?"

The answer, almost every time, is the same. They have automation. They do not have touchless AP. Those are two very different things.

The definition problem

Touchless. That word appears everywhere among AP automation vendors. Basware uses it. Tipalti throws it around. Stampli includes it in every pitch. HighRadius too. Ask each one what the word actually means? The answers shift depending on who you ask. For some, it means skipping manual data entry. Others point to automatic routing, with most invoices moving through without a person. Then there are those who claim full end-to-end processing, no human needing to confirm anything at any stage.

Each of those works differently than the next. One costs more per invoice. Another gives back hours to your finance team. Some barely change anything at all.

Here is the question we start with instead. When scoping a new deployment, the first thing we look at is not the touchless rate. It is what falls through the cracks: the invoices that do not move cleanly. Then comes the real point: who picks those up, and how.

What the benchmarks don't tell you

Ardent Partners' 2025 State of ePayables report is the most cited reference across AP circles. Top performers process an invoice in just over three days, spending less than three euros each time. Meanwhile, typical companies drag that out past two weeks, paying nearly thirteen euros per invoice. That gap, sharp and quiet, tells its own story.

We bring up those numbers in almost every client conversation. But we always add the context most vendors leave out.

Achieving the 3.1-day, €2.78 outcome depends on four things being in place: ERP systems fully linked so entries post directly, supplier records kept accurate, purchase orders strictly followed, and exceptions that get resolved automatically rather than rerouted back to staff.

Most finance teams start without any of those. The distance between them and the benchmark is not a technology gap. It is a process architecture gap. That is exactly where most AP automation projects quietly underperform.

Three generations of AP automation and why they produce different results

From working with organisations that process between 20,000 and 200,000 invoices annually, we find their AP tools almost always fall into one of three categories. The pattern repeats across industries. Where systems differ most is not in design but in how far they actually go.

The first generation: OCR-based capture. Machines read invoice files and pull out key details. No more manual typing. But someone still checks every result. Matching records, signing off, fixing mismatches, all of that stays hands-on. The person just appears later in the flow. Cost per invoice drops a little. Speed barely moves.

The second generation: workflow automation. Rules route invoices based on set conditions: spending thresholds, department codes, supplier categories. Approvals move faster. Digital logs appear automatically. But at 20% exception rates, the industry average according to Ardent Partners, people are still manually resolving thousands of cases every year. At 50,000 invoices annually, that is 10,000 times someone steps in by hand. Most organisations sit here today. Peel back the marketing and most so-called AI tools land in this category too.

The third generation: AI agent-based automation. The system reads each invoice. It checks against purchase orders and delivery records. Then it posts directly into the ERP under a designated login, with no person needing to press confirm. When something unusual shows up, it does not disappear into a queue. Every exception gets tagged clearly: why it stopped, what the likely fix is. Only the genuinely complex cases move forward to a person. That week, the finance team reviewed 47 items. Not 2,000.

What separates second from third generation is visible in one number: how many invoices per month still require a human to touch them.

What exceptions actually reveal

A standard invoice does not reveal how good an AP system is. Every system handles those fine. A PDF from a familiar supplier, matching purchase order, correct amounts, tidy reference details, that moves smoothly across almost any modern AP platform without anyone stepping in.

Exceptions are what separate real touchless AP from the marketed version.

In every deployment we run, the ones that matter most are predictable. Invoices where the line-item price has shifted outside the agreed tolerance. Bills from suppliers not yet loaded into the master data. Multi-page invoices where the system cannot confidently assign GL codes. References to project numbers that do not match any open purchase order.

In a workflow automation system, every one of those creates a job for a person. In a well-configured AI agent system, the agent checks the variance against your policy, applies the resolution where the rules are clear, and escalates only the genuinely ambiguous cases.

That is the difference between a system that reduces hours and one that changes what the job actually is.

What we saw at CitizenM

CitizenM runs hotels across multiple countries and processes invoices across multiple entities. Each month, more bills arrive from different corners of the globe. Their finance team was spending significant time pulling numbers by hand, matching them to purchase orders, then posting everything into the system. As volumes grew, the hours did not shrink.

After deploying an AI agent integrated directly with their ERP: 2.5 FTE freed per 50,000 invoices processed annually. 99.7% workflow accuracy, not extraction accuracy. Full workflow accuracy. That covers extraction, three-way matching, GL coding, and direct ERP posting.

That distinction matters when evaluating vendor claims. Extraction accuracy is typically high and easy to claim. Workflow accuracy, which accounts for matching errors, coding errors, and posting errors, is where implementations usually fall short of what was promised.

CitizenM's finance team no longer handles standard invoices. They manage the exceptions the system flags. That is what touchless AP looks like when it actually works.

The ERP integration question nobody asks

One factor consistently separates genuine touchless AP from near-touchless AP. It is ERP integration depth. And it is the question we find most buyers forget to ask.

Reading an invoice and extracting data is one step. Posting a validated, matched, coded invoice directly into SAP or AFAS as a confirmed ledger entry is a completely different step.

Most AP platforms do the first. They hand structured data to your ERP and wait for a human to confirm the post. Some call that integration. In practice, it means someone on your team still initiates the final step for every single invoice.

The systems that hit best-in-class benchmarks operate as named users inside your ERP. They do not hand off. They book directly. When the system processes the invoice, the invoice is processed. No confirmation needed.

For organisations handling 50,000 or more invoices annually, that final manual step represents a significant portion of remaining AP labour, even after "automation" is live. Removing it is the difference between 90% touchless and 99% touchless. Across large volumes, the cost difference between those two rates is real.

Three questions to ask every vendor

After working through hundreds of vendor evaluations with our clients, we have settled on three questions that reveal more than any product demonstration.

First: what is your straight-through rate specifically on invoices with exceptions? Not standard invoices. Invoices with price variances, missing PO references, and new vendors. Every system claims a strong overall touchless number. The exception rate is where the actual difference lives.

Second: how does your system post to the ERP? Direct post as a named system user, or a data export that needs human confirmation? That single answer tells you whether the final manual step in your AP process disappears or just shrinks.

Third: what does the AP team actually do on Monday morning after deployment? Ask for a documented breakdown of time allocation before and after, not a touchless percentage. The answer shows whether the nature of the work changes or just the volume goes down.

The CFO who asks all three will get meaningfully different answers from different vendors. In our experience, the organisations achieving 3.1-day processing at €2.78 per invoice are the ones who asked exactly these questions before signing anything.

If you are evaluating AP automation for a high-volume operation, see how Freeday approaches AP deployments, including ERP integration model, exception handling, and outcome-based pricing.

‍