Beyond 3-Way Match: The Agentic AI Playbook for AP

This is moving faster than the controls around it

Someone on your team is using ChatGPT to draft vendor emails. Someone else is using Claude to explain a freight variance. Your shared services lead just sent you a deck about “agentic AP.” Three of your vendors just changed their homepage to say the same thing.

The question is not whether AI belongs in your AP function. It is already there. The question is whether any of it is defensible: in front of your auditor, in front of your audit committee, in front of the next duplicate payment your team finds in the wrong quarter.

That is what this session is about.

Why 3-way match isn't enough anymore

3-way match did its job. It took AP touchless rates from 0 to 60-70% across the enterprise. Then it stopped.

Most enterprises have sat at 50-70% touchless for three years despite continued investment in OCR, workflow tools, and ERP upgrades. The remaining 30-40% is not an invoice problem. It is the invoice from “Acme Logistics LLC” against a PO for “Acme Corp” because procurement created a duplicate vendor master in 2022. It is the 8% price variance because a contract escalation clause kicked in last month and nobody updated the PO. It is the invoice that arrived before the goods receipt because shipping logged the delivery in the wrong week.

None of these are invoice problems. They are data problems, and they live in the lineage around the invoice. Rules-based 3-way match has no way to reason about any of them.

What agentic AI does that rules cannot

Rules can detect a mismatch. They cannot reason about one. That is the gap agentic AI fills, and it is the gap that takes touchless rates from 60% to 90%-plus.

But there is a catch. Most of what is being marketed as “agentic AP” today is probabilistic AI giving you a confidence score, not a decision. “94% confident” is not an audit trail. It will not survive your next SOX review. What separates real agentic AP from rebranded probabilistic AI is governance: every decision in plain English, every step logged, every action defensible. That is the line this webinar will draw clearly.

Five minutes after you connect, your AP already looks different

Your team carries a knowledge graph in their heads. “Acme Logistics LLC” and “Acme Corp.” Same vendor. Different records. They just know. That tribal knowledge is why AP works, and why rules alone can’t automate it.

We rebuilt it in software. Connect Kognitos to your ERP and a context graph of your AP function builds itself in minutes. Then agents go hunting: duplicate vendors, duplicate payments, contract drift, anomalous approvals, fraud signals.

By the time your coffee’s cold, they’ve surfaced things your team has been carrying for quarters.

We did it live, on a real ERP, on stage. Watch the recording above to see all three workflows from start to finish.

Who you'll hear from

55 minutes, no fluff

Three things, in 55 minutes

1. The playbook for what comes after rules-based matching. Where ChatGPT, Claude, and the new wave of “agentic” AP tools genuinely help. Where they quietly fail. The gap between “94% confident” and “the auditor will sign off on it.” Stop reading vendor websites and start reading your own AP process.
2. Three workflows that show the difference, live. Non-PO invoice coding. 3-way match exceptions. Vendor statement reconciliation. The three workflows that consume the most AP labor in every enterprise we have seen, run end-to-end through Claude, governed by Kognitos. With the audit log on screen.
3. A framework your CFO will actually defend. A simple way to evaluate any agentic AI claim in your AP stack: what to ask in the RFP, what to demand in the demo, what to verify in the audit trail. Plus the leakage math for your own function, sized to your volume and exception rate.

Three things keeping AP leaders up at night

• The audit question you don't want to answer. “Walk me through how this invoice was coded, approved, and paid.” If the answer involves “the AI was 94% confident,” you have a problem. Probabilistic AI does not produce an audit trail. It produces a confidence score.
• The money already leaking out. Duplicate payments. Missed early-pay discounts. Non-PO invoices coded to the wrong cost center for three quarters before anyone noticed. Most AP teams are running on a 70% automated, 30% on-fire model, and the 30% is where the money actually goes.
• The team you cannot keep. Senior AP analysts are leaving for FP&A roles. The exception queue is what burns them out. Every cycle that ends in a Friday-night manual reconciliation is a retention risk you are not measuring.

If any of these are true, this hour will earn its slot on your calendar

• You sign off on financial statements and have started getting AI questions from your audit committee.
• Your AP team is already using ChatGPT or Claude informally and you have no governance around it.
• You have an OCR or AP automation tool that handles clean invoices and leaves the rest to your team.
• You are building an AP RFP and three vendors just told you they are “agentic.”
• You found a duplicate payment last quarter that should not have happened.
• Your senior AP people are leaving because of exception fatigue.

Job titles we built this for: CFOs, VPs of Finance, Controllers, AP Directors, Heads of Procure-to-Pay, Heads of Shared Services. Companies: 500–5,000 employees. Industries: manufacturing, logistics, healthcare, banking, retail, technology, CPG.

The four AP leakage dimensions covered in the recording

The Q&A section walks through how to size each of these for your own AP function. Use them as your post-watch checklist:

• Duplicate payment risk, sized to your invoice volume and vendor-master hygiene.
• Non-PO coverage gap, where probabilistic AI most often fails and where 30-40% of leakage hides.
• Exception aging, the labor cost hiding in your AP queue and the leading retention risk for senior analysts.
• Audit trail completeness, the question your auditor will ask next, and the line that separates real agentic AP from rebranded probabilistic AI.

Want a personalized version for your AP function? Book a working session and we’ll size all four against your invoice volume and exception rate.