RETHINKING ACCOUNTS RECEIVABLE IN THE AGE OF AI

George Kurdin, CEO and Co-Founder of Monk, discusses how AI agents are changing accounts receivable, why ERP integrity matters and where human judgment remains essential.

BEYOND RULES AND ROUTINE FOLLOW UPS

Accounts receivable depends on customer relationships, fragmented information and exceptions. Context aware AI agents offer a different approach to work that fixed workflows have struggled to handle.

FLEXIBLE THINKING. CLEAR LIMITS.

Greater autonomy requires clear boundaries. While agents handle routine activity, people retain control over commercial terms, financial concessions and sensitive customer decisions.

In this ERP News Q&A, George Kurdin explores autonomous collections, cash application and ERP integration, explaining what it takes to give AI more responsibility without compromising financial control.

GEORGE KURDIN

CEO & CO-FOUNDER, Monk

George Kurdin is CEO and Co-Founder of Monk, which develops AI agents for accounts receivable. His approach centres on helping finance teams manage the exceptions, fragmented information and customer interactions that have made AR difficult to automate.

In this interview, he discusses how agents can support collections, cash application and invoice submission while operating within clear boundaries. He also explains why reliable ERP integration, traceable actions and human judgment remain essential as finance teams give AI more operational responsibility.

The thinking is flexible. The authority is not.

Why Traditional Accounts Receivable Automation Falls Short

Q: Accounts receivable has historically remained more dependent on spreadsheets, inboxes and manual processes than many other areas of finance. Why has AR proved so difficult to automate effectively, and what can AI address today that previous generations of automation could not?

GEORGE KURDIN: Start with the size of the problem. More than $10T is stuck in unpaid invoices around the world. DSO has climbed into the high 70s in most industries since the supply chain shocks. That is cash sitting in someone else’s bank account.

AP is a ledger problem. The inputs are yours. One process, start to finish, under your control. AR is a relationship problem. The inputs come from another company: their portal, their remittance format, their approvals, their inbox. Every customer is a different system. Rules need things to stay the same shape. In order-to-cash, nothing does. So teams stayed on spreadsheets and email, because those take anything you throw at them.

And the work is exceptions. Exceptions eat the day. Exceptions drive DSO. Each one needs its own approach and real thinking about that account, that short-pay, that deduction. You can’t write that down in advance.

Here is what changed. We now have agents that can work a problem over a long stretch and hold a lot of context. One agent can carry the whole history of an issue across email, portals, the ERP and call notes, then use tools to go do the work. That did not exist three years ago.

Solve exceptions properly and automation finally means something to the person doing the job. That is what we work on at Monk.

Q: Much of traditional AR automation has focused on predefined workflows and rules. What changes when AI can interpret the context of a customer interaction or payment situation and determine what should happen next, rather than simply follow a fixed sequence?

GEORGE KURDIN: A workflow engine holds what you expected to happen. AR is mostly exceptions. So a rules tree with 200 branches tells you the problem was never rule-shaped.

What changes is that the system acts on what is true today. They short-paid two invoices by a freight charge. They are mid-renewal. They have filed two deductions before. Instead of sending dunning step 4 on day 31.

What does not change is the set of things it is allowed to do. The agent picks from a fixed list of actions, each with hard guardrails. The thinking is flexible. The authority is not. We call it boring agents by design.

Balancing AI Autonomy with Human Judgment

Q: Intelligent Collections is one area where agentic AI is already taking on more responsibility. Where do you draw the boundary between work that an AI agent can handle autonomously and situations that should be escalated to an AR professional?

GEORGE KURDIN: It depends on the business and the segment. A staffing agency billing through VMS portals, a mid-market software company with 40 big accounts, and an SMB with 800 small invoices carry very different risk. We set this per customer.

Above that, you can generalize. Draw the line at reversibility and money.

The agent handles anything you can undo without much damage. Reminders, chasing remittance advice, matching payments, submitting invoices to portals, research, passing work to the right person inside the company.

A person handles anything that changes a commercial term, gives up money, or spends goodwill. Write-offs, payment plans, discounts, credit holds, legal.

One thing worth saying: in mid-market and SMB, the agent can do more than people assume. Smaller invoices, more repeatable payer behavior, less politics per account. A lot of that book can run without anyone touching it.

The second line is confidence. Below the threshold, the agent stops and hands the work over with everything already gathered: history, how this payer behaves, a draft reply. Escalation is not a failure. It is routing. Most teams escalate badly today because the context shows up empty.

“Escalation is not a failure. It is routing.”

Q: If automation handles more routine collection activity, the cases reaching people are likely to be the more complex ones — disputes, unusual payment behavior, sensitive customer relationships or exceptions. Does greater automation ultimately make human judgment more valuable in AR rather than less important?

GEORGE KURDIN: Yes. We want to help people, not replace them. Take the best person on the AR team and make them better. Judgment dies in a queue of 300 follow-ups. It gets sharper with time and room to think. Give both back and people do the work that matters: run the book, work the real disputes, restructure a payment plan, decide which customer can take pressure this quarter.

Credit management is the clearest case. Good credit work is judgment plus data. Who pays, who does not, who is about to stop paying everyone. The policy needs a person. The edge comes from data most teams never see: how a customer pays across a whole network, not just your own aging report. Give a good credit manager both and they stop problems instead of chasing them.

Cash Application, ERP Integration, and Financial Data Integrity

Q: Cash application presents a different automation challenge. Partial payments, consolidated remittances, missing references and payments covering multiple invoices or entities can introduce significant ambiguity. What makes these scenarios difficult for conventional automation, and how is AI improving the ability to resolve them?

GEORGE KURDIN: Two problems in the old world.

First, the rules are weak. Match the string, match the amount, against open invoices. That breaks on a wire covering 14 invoices across two entities, short-paid for a credit memo, with the remittance advice in a scanned PDF in someone’s inbox.

Second, there is nothing behind the rules. The algorithm clears some share and hands the rest to the team to work from scratch. No partial credit. No head start. That is why the auto-match rate on the slide looks fine while the team is still drowning.

Our engine clears a much larger share straight through. We do not force a match. We only apply at full confidence, because cash applied wrong is worse than cash not applied. For what is left, the engine suggests ranked matches with the evidence attached. The team checks an answer instead of starting cold. That is how touchless cash application gets to a number that means something.

Q: Your contributed ERP News feature looks closely at what happens when AR automation has completed its work and the result has to land correctly back in the ERP. From a broader architectural perspective, how should companies think about the relationship between the ERP as the financial system of record and the increasingly intelligent automation layer operating around it?

GEORGE KURDIN: The ERP is the system of record. It stays the authority. The layer above it holds no competing version of the truth. It proposes. The ERP decides.

What that takes is an integration that never breaks, and this is the part nobody wants to talk about. You have to send accurate data. You cannot flood it. You need health checks at every stage of the sync. Every write has to be safe to repeat, traceable to who made it, and possible to undo. Reconcile as you go, not at month end.

The work varies a lot by system. NetSuite, SAP and Dynamics are each their own craft. Modern APIs, but different data models, different limits, different ways of failing. Then you hit something like Sun Systems, where there may be no API and no documentation, and you are solving a different problem. Any vendor who lists “integrates with your ERP” as one line has not done the work.

Building these integrations is an art. It takes real engineering time and it does not demo well, which is why most products are thin here. None of the reasoning matters if the sync is wrong. The books still close in the ERP.

Q: A large part of invoice-to-cash activity also takes place outside the ERP — through customer emails, remittance documents, AP portals, payment systems and conversations. Is connecting and interpreting this fragmented environment becoming as important as integrating with the ERP itself?

GEORGE KURDIN: Yes. Extremely important.

ERP integration has an end, at least on modern systems. The outside world does not: thousands of AP portals, email threads, payer formats and bank files with no spec.

We are pushing hard to automate it. Invoice uploads into Coupa, Ariba, Tipalti. VMS portals for staffing agencies. Those cut DSO directly. Much of it is now possible with a smart mix of deterministic systems and AI. We are far from done, but I am proud of what we have shipped for customers so far.

The Future of Autonomous AR: Skills, Governance, and Trust

Q: As AI and automation take on more operational work across collections, cash application, invoice submission and routine follow-up, how do you expect the role of AR professionals to change? What skills become more important when people spend less time executing repetitive processes?

GEORGE KURDIN: From doing the work to overseeing it and negotiating.

What becomes more valuable: judgment on exceptions, resolving disputes and deductions, credit management, reading an audit trail well enough to catch the system being wrong, and enough comfort with data to question a number.

Collections starts to look like managing a portfolio rather than working a list. The analyst who can restructure a payment plan with a big account is worth several who can send dunning emails fast.

Q: Finance sets a particularly high bar for AI because an incorrect action can affect cash, accounting records and customer relationships. What should CFOs look for in terms of accuracy, auditability, controls and human oversight before trusting AI with more of the invoice-to-cash process?

GEORGE KURDIN: Security. This is customer financial data moving between systems. The controls should be plain and easy to check.

Accuracy and precision, reported by action type, with the confidence threshold published. Not one blended number. 95% across everything hides the 60% on the thing that matters. Same with auto-match rate — ask what happens to the rest.

An audit trail. Every action traces back to what it saw and the rule that allowed it.

A staged rollout. Shadow mode, then suggest and approve, then autonomous inside set bounds. Measured at each step.

And hard dollar ROI. It is time to see real money from AI. Saving time, running leaner, getting better reporting — not enough on its own. Revenue up or cost down. With $10T stuck in unpaid invoices and DSO in the high 70s, the return here shows up in days of working capital. That is part of why we like this space. Ask for the number and refuse the soft ones.

Q: Looking three to five years ahead, how autonomous do you think invoice-to-cash can realistically become? Which parts of AR do you expect to become largely automated or autonomous, and where do you believe human involvement will remain essential?

GEORGE KURDIN: Cash application goes first. Most structure, clearest feedback. Then invoice submission and portal work. Then routine dunning. Disputes and deductions go automatic on the research and stay human on the call.

Human work stays: credit strategy, negotiation, write-offs, and anything where the relationship is worth more than the invoice.

But people are bad at seeing power laws, so this will move faster than expected. A number of our mid-market customers already run collections on full autopilot. That did not happen by flipping a switch. It came from working with their team on operating procedures, getting clean data into the ERP, and building the harness around the agent. Once that is in place, the agent runs the book. It is in production, not on a slide.

Build AI that is reliable, secure and made for AR, and I think we will all be impressed by where this goes.

EDITORIAL TAKEAWAY

Accounts receivable remains difficult to automate because its work extends beyond the ledger. Customer communications, payment documents, external portals and commercial relationships all shape what happens next.

George Kurdin places exceptions at the centre of the AI opportunity. Agents can gather context and handle routine actions, giving professionals more time for disputes, credit decisions and negotiation.

Greater autonomy, however, requires clear limits. Financial concessions, commercial terms and sensitive relationships need defined controls and human judgment.

ERP integration is equally fundamental. Automation must return accurate, traceable results to the system of record. Strong reasoning cannot compensate for incorrect accounting entries.

Finance leaders should look beyond headline automation rates. Accuracy by action type, audit trails, staged deployment and measurable financial outcomes provide a stronger basis for trust.

The future of AR depends on combining contextual intelligence with dependable execution, while giving people the visibility and authority to intervene.

AUTONOMY EARNS TRUST THROUGH CLEAR LIMITS, RELIABLE ERP RECORDS AND HUMAN OVERSIGHT.