How does AI actually improve efficiency in accounts payable?

By handling the exceptions that break rule-based AP automation. The model reads non-standard documents (supplier statements, rebate contracts, off-template invoices), matches them against your ERP, and drafts entries for a person to approve. The efficiency gain is not "fewer clicks per invoice." It is capturing the work your team currently skips because there is not time.

Are AI agents in AP different from AP automation?

Mostly a naming choice. "AI agents" typically describes an AI-driven workflow that reads, decides, and drafts entries, while traditional ap automation software runs on templates and rules. In an honest build, both still route to a human for approval before writing to the ledger. Ignore the label and ask what the tool actually reads and what it actually writes.

Do I replace my accounts payable software or add AI to it?

Add to it, in almost every case. Your existing accounts payable software handles the 80% that fits its templates. An AI layer sits on top and covers the exceptions, reconciliations, and unstructured documents that fall outside those templates. Replacement is a much bigger project with far less certain payback.

Can the AI post entries directly without a person approving?

It can, and it should not. Any write to a money-touching record in your ERP goes through a person on your team. That is the guardrail we build into every system, including the one running inside Kelsan.

What does a build like this cost?

We do not quote before we understand the chore. Every engagement starts with a fixed-fee AI Capability Audit ($7,500), and the build itself is a separate fixed fee agreed in writing before any work begins. Cloud and model usage is metered like any other cloud service and quoted before the build. If AP is where your team is bleeding hours or your margin is walking out the door in unclaimed rebates and reconciliation gaps, the next step is a conversation. Book a call with Throughline, or start with a fixed-fee AI Capability Audit ($7,500). The call is free. Call 865-417-3554.

AI in Accounts Payable: What It Actually Does in an ERP

What does AI in accounts payable actually do inside a real business? It reads the invoices, statements, and supplier documents you already receive, matches them against records in the ERP you already run, and surfaces entries for a person on your team to approve before anything posts. That is the honest version. Everything else in this category is either marketing gloss on the same idea, or a claim of autonomy that no mid-market operator should sign off on.

Key takeaways

AI in AP is a system built into your existing ERP, not a separate product you swap in for the software you already run.
The mechanism is concrete: read the documents you receive, match them against ERP records, queue entries for human approval.
A person on your team approves every write to a money-touching record. That is the guardrail, not a limitation.
Throughline built a rebate-and-margin recovery system inside Kelsan, a multi-state distributor running Epicor P21, that has recovered over $100,000 in margin.
Every Throughline engagement starts with a fixed-fee AI Capability Audit at $7,500. The initial call is free.

What AI actually does inside an AP workflow

Strip out the vendor language and the work is narrow. An AP inbox receives PDFs, EDI files, portal exports, supplier statements, and rebate agreements. Someone reads them, keys line items into the ERP, matches invoices to POs and receipts, flags exceptions, and processes payments. The volume is high, the format is inconsistent, and the mistakes are expensive.

AI in this workflow does three concrete things. It reads unstructured documents and extracts structured fields (vendor, invoice number, line items, terms). It compares those fields against records already in the ERP, including POs, receipts, contracts, and rebate schedules. It writes proposed entries (a coded invoice, a credit memo, a rebate accrual) into a queue where a person on the AP team approves, edits, or rejects before anything hits the general ledger.

That is it. When you see "AI agents in AP" in a vendor deck, they usually mean the same three steps with a friendlier label. The interesting engineering is not the model. It is the plumbing: reading a supplier's messy PDF reliably, matching it against the right invoice in Epicor P21 or NetSuite or Sage, and getting the write into the ERP in a form the approver trusts. That plumbing is what intelligent document automation actually is when you build it for a real AP function.

AP automation vs. AI in accounts payable

The terms get muddled on purpose. Traditional accounts payable automation software runs on templates and rules: OCR a known invoice layout, apply a coding rule, route for approval. It works when your supplier base is stable and their documents are consistent. It breaks the moment a supplier changes format, sends a statement instead of an invoice, or bundles a rebate schedule into a PDF that does not match any template.

AI in AP handles the documents that break rule-based ap automation. It reads a document it has never seen before and pulls the right fields anyway. It reconciles a supplier statement against ten invoices already in your ERP and flags the two that do not match. It reads a rebate contract, watches for the volume threshold in your purchase history, and proposes the accrual entry. The AP automation software you already run stays. The AI layer sits on top of it, closing the gaps where templates fail.

If your current accounts payable software works for 80% of your volume, the question is not whether to replace it. The question is whether the remaining 20% (the exceptions, the rebates, the statement reconciliations, the non-standard suppliers) is where your team is bleeding hours and margin. That is where an AI layer earns its keep.

What we built inside Kelsan

Here is the first-hand version. Kelsan is a multi-state distributor in our group, running Epicor P21. Rebate programs are a real source of margin for distributors, and they are also a real source of leakage: contracts are long, tier thresholds shift, and the reconciliation between what suppliers owe and what actually gets credited is manual, tedious, and easy to skip when the AP team is buried.

We built a rebate-and-margin recovery system inside Kelsan that reads the supplier rebate documents Kelsan already receives, matches them against invoices and purchase history in P21, and surfaces the entries that should be claimed. A person on Kelsan's team reviews every entry before it posts. The system has recovered over $100,000 in margin that would otherwise have been left with suppliers.

Two things worth naming honestly. First, the dollar figure comes from a specific business with a specific supplier base; your numbers will not be ours. Second, the reason this works is not the model choice. It is that the tool is wired into P21 directly and the approver's workflow sits where the AP team already works. That is what ERP automation means in practice. If you want the longer version, we wrote up the rebate-and-margin tool we built inside Kelsan separately.

Where AI in AP is the wrong tool

Not every AP function needs this. If your invoice volume is low enough that one person handles it comfortably, the payback on any build is thin. If your supplier documents are already highly standardized (EDI-heavy, portal-driven), rule-based accounts payable automation software will get you most of the way there without an AI layer.

A build is also wrong if the org is not ready to keep a human in the loop. Anything that writes to a money-touching record in your ERP needs a person on your team approving it. If leadership is looking for full autonomy over payables, this is not the category to buy into, and it is not the build we do. The value of the human approval step is not that it slows things down. It is that it catches the 3% of cases where the model is wrong, and those cases are where a bad AP posting turns into a real problem.

What a real implementation looks like

The path is boring and specific, which is the point.

Audit the actual chore. Sit with the AP team for a week. Watch which documents cost the most time, which exceptions repeat, which reconciliations get skipped. This is where our fixed-fee AI Capability Audit ($7,500) lives.
Pick one narrow workflow. Rebate reconciliation, statement matching, coding non-PO invoices, three-way match exceptions. One. Not "AP" as a category.
Build the reader. Wire an AI systems integrator into the document intake so the tool parses what actually comes in, not a clean sample.
Wire it to the ERP. Reads pull from live records. Writes go into a queue, not directly to the ledger.
Put the approval queue where the team already works. If they have to log into a new tool to approve entries, adoption dies.
Measure against the pre-build baseline. Hours saved, exceptions caught, dollars recovered. If it does not move a number, kill it.

That is the shape of back office automation when it is built to last. It is also why we do not sell an AP product. Every ERP, every supplier base, every AP team is different enough that a shrink-wrapped tool leaves the highest-value cases on the floor.

About the author

Throughline is a small team of builders inside Keller Group. We build AI systems into our own operating companies first, then into yours.