What percentage of your incoming invoices are currently processed into a posting proposal in your ERP without human intervention? If you want to automate accounts payable, we provide you with a single workflow: whatever comes in goes through that workflow and appears as a posting proposal in your ERP—and that’s where it stops. What’s correct moves on; what’s incorrect goes into a workbin with the reason noted.
So what exactly happens between the time we receive the request and the booking proposal?
What We Automate for You
For us , automating invoice processing covers the entire chain, from receipt to posting proposal, and that chain begins with a single data model, regardless of the channel through which an invoice is received. Invoices received via Peppol comply with the European standard EN 16931. Invoices received via email, PDF, or scan are converted to the same data model as Peppol, so that no separate rules are needed for each channel further down the line.
Based on that data model, the invoice moves through the chain. Each step in the chain makes a single decision and explains why, with a score and the reason; you can always check why an invoice was approved or rejected. At the end, the posting proposal enters the ERP system via the standard invoice import; the ERP remains the system where postings are made and payments are processed.
We make a proposal; you approve it. Your ERP system posts the transaction and makes the payment. By design, the agent does not post transactions, make payments, approve them, or modify supplier data.
And what about the team lead who keeps your AP team running? The workbench only shows him which lines are incorrect, along with the reason. Correcting them is faster than retyping them.
So why is there still manual work involved today, if invoices are already recognized?
Where Things Are Getting Stuck Today
The manual work that remains involves structure, not sloppiness. In accounts payable, a lot is already set in stone: the format of a Peppol invoice and the VAT rules. As of January 1, 2026, VAT-registered companies in Belgium must exchange their B2B invoices in a structured format, using Peppol as the standard. Still, an exception—such as a missing purchase order, a price that’s just off, or an approver on vacation—might end up on someone’s desk on your team.
According to Ardent Partners (AP Metrics that Matter in 2025), 53% of the AP and finance managers surveyed cite invoice exceptions as one of their biggest challenges. This is an indicative figure and not a survey specific to Belgium.
Detection is not the same as resolution. We don’t see the benefit in typing faster, but rather in the percentage of invoices that don’t need to be reviewed by a human, and in how quickly the rest are resolved. By that measure, a tool that detects anomalies and places them in a queue merely shifts the workload and doesn’t actually resolve anything.
What kind of verification is required for an invoice, with and without a purchase order?
With or without an order form: two routes, one supply chain
An invoice with a purchase order and an invoice without a purchase order require different verification procedures.
Invoices with a purchase order
For invoices accompanied by a purchase order, we perform a line-item comparison: we compare the invoice line by line with the purchase order and the receipt. This allows us to catch any discrepancies in price or quantity that might be overlooked in the total.
Invoices Without a Purchase Order
Without a purchase order, the validation process relies on coding and the approval workflow. We generate the coding based on previous bookings, and the invoice is sent to whoever the approval workflow designates.
Two routes, one supply chain: what benefits do they offer together?
What It Offers
We measure the results using four metrics, and each one is clearly defined here. For the cost per processed invoice and the touchless rate, we’ll agree on the target values with you during the baseline measurement.
Cost per processed invoice. In *AP Metrics that Matter in 2025*, Ardent Partners reports a cost of $2.78 per processed invoice for the top-performing group, compared to an average of $9.40; amounts are in dollars, as published, and fully loaded. Ardent Partners defines the top-performing group as the 20% of companies with the lowest average cost per invoice and the shortest average processing time. Keep these figures in context: the sample is predominantly North American (59% of respondents), and 53% work for a large company with revenue exceeding $1 billion. For us, the cost per processed invoice is a target value that we agree upon together during the baseline assessment, measured across all incoming invoices during the period, including licenses, infrastructure, and management.
Processing time. The same measurement shows 3.1 days per invoice for the top-performing group, compared to an average of 9.2 days, using the same sample size. We measure the processing time ourselves, from receipt to final approval. This is a measurement standard, not a target value.
Touchless rate. The touchless rate—broken down by invoices with and without purchase orders—is a target value that we agree upon together during the baseline measurement. “Touchless” means that an invoice enters the ERP system as a proposal without human intervention, where the ERP then posts it. This is how we define automatic invoice posting: the proposal enters the ERP system without human intervention, and the ERP posts it.
Exceptions resolved autonomously. This metric does not have a target value, but it does have an agreed-upon definition. We define the subset of exceptions that are resolved without human intervention during the baseline measurement: which cases count, what “resolved” means, and over what time period.
Anyone who says, “We’ve already tried this,” won’t get a promise from us, but rather a measurement metric. The baseline measurement follows the same definition as the follow-up measurement; otherwise, you’ll be measuring an improvement that doesn’t exist.
The question remains whether this applies to your situation.
Who This Isn’t For
There are three situations in which this approach is not appropriate—or at least not yet.
Do you process few invoices? If the number of invoices per month falls below a certain threshold, this won’t pay for itself—and we’ll be upfront about that.
If you don’t enter purchase orders, start with the master data rather than AI.
If you have an ERP migration planned, you don’t have to wait for it: the agents run on top of your ERP, and at go-live, you switch the source, not the agent. Waiting is only advisable if go-live is within three months, or if the new ERP resolves the issue on its own. Before going live, the workflow runs in the shadow of the current process, using actual invoices, without posting any entries—this is the shadow run.
If this applies to your situation, how do you get started?
How We Get Started
Our AP automation process begins with an AI Discovery Session lasting half a day to a full day; the outcome is a shortlist based on impact and effort, in that order. After the AI Discovery Session, we’ll build a working prototype using your own invoices within a few weeks. Then you decide: move forward, make adjustments, or stop. We scale up what works.
If you’d like to assess your numbers yourself first, you can do so with the self-assessment. A 20- to 30-minute online consultation is all it takes to see where AI can deliver value for you.
Where is this approach already being implemented today?
Where this chain is already in operation today
The same approach is currently in use in multiple client implementations across finance, customer service, and logistics. In a previous project by our team, the reconciliation of supplier statements went from requiring a team of five people to taking less than two hours of work per week. In another project by our team, order processing went from requiring eight full-time employees to requiring less than one. These are document flows other than invoices, but they use the same validation logic.
Do you know your AP scores, or are you just guessing?
A 20- to 30-minute online conversation is all it takes to see where AI can deliver value for you.