Agentic Payments Need Judgment, Not Just Automation

Treasury has never been about moving money

Most people outside finance think treasury is about paying bills.

It isn’t.

Treasury is about deciding when, how, and sometimes whether money should move at all.

That’s a completely different problem.

Imagine a company with:

• $10M in available cash.
• $8M in supplier invoices due this week.
• A treasury fund yielding 8% annually.
• And short-term financing is available at attractive rates.

A purely automated payment engine might conclude:

• Cash is available.
• Execute every payment immediately.
• Nothing technically wrong happened.
• Every supplier gets paid.

No penalties. No overdue invoices.

From an operational perspective, the system performed perfectly.

Yet an experienced treasurer might make a completely different decision.

Instead of paying every invoice today, they might deliberately postpone a portion of the payments by 48 hours. Why?

Because the additional yield generated by keeping that cash invested exceeds the total cost of the late fees.

The company ends up with more money, despite paying some penalties.

To a rule-based system, this looks irrational. To a treasurer, it’s simply capital allocation. That is the main reason why treasury is not an optimization problem with only one objective.

It is about dozens of objectives constantly competing with one another.

• Should you optimize for liquidity?
• Working capital?
• Financing costs?
• Investment returns?
• Foreign exchange exposure?
• Credit line utilization?

None of these variables exists in isolation. Sometimes paying early is the right decision, sometimes paying exactly on the due date is. Even more, sometimes paying latecreates the highest enterprise value.

The answer depends on context.

A machine optimizing for “avoid payment fees” would never make that decision.

Get Alejo Lovallo’s stories in your inbox

Remember me for faster sign in

That’s why experienced treasurers are so valuable. They’re not executing payments. They’re continuously balancing trade-offs.

This is where Agentic Payments become interesting. Not because they can execute transactions. Because they can reason about trade-offs.

The best decision isn’t always the fastest one

One thing we´ve learned while building Fonder AI Treasurer is that agents lovelocal optimization.

Give an AI a single KPI, and it will become incredibly good at maximizing it. Unfortunately, businesses rarely operate with a single KPI.

Imagine telling an Agentic Payment system: “Minimize payment penalties”.

Congratulations! It will pay everything immediately.

Now imagine telling it: “Maximize investment yield”. It might postpone every payment until the very last second.

Neither system is “wrong.” They’re simply optimizing the wrong objective.

Real treasury is about balancing both. The best financial decisions often emerge from tension, not certainty.

Judgment is the real product

From the questions asked above, you can notice something interesting:

None of these questions has a universally correct answer. They’re contextual. They’re probabilistic. They’re judgment calls.

Which means no single optimization function is enough. The job isn’t finding the mathematically perfect answer. The job is making the best decision under uncertainty.

That’s judgment.

Judgment extends beyond spreadsheets

Now imagine two suppliers. Supplier A and Supplier B.

On paper, they look identical. Same invoice amount, same due date, and contractual terms.

A simple AI Agent sees no difference. A treasurer does.

• One supplier has supported the company for fifteen years. They accepted delayed payments during difficult periods.
• The other will probably sue the company immediately and will never sell to the client again.

The treasurer will choose to negotiate a long-term contract with Supplier A.

Maintaining that relationship may be worth far more than the interest earned by delaying payment.

• That decision doesn’t exist in the ERP.
• It does not appear in the invoice.
• It lives in experience.

If Agentic Payments are going to become real financial teammates, they’ll eventually need to reason about these softer dimensions as well.

• Business context.
• Relationships.
• Company philosophy.

Building the treasurer, not the payment engine

AtFonder, we’re focused on capturing the reasoning of great treasurers — not just automating transactions and payments.

It is about encoding the judgment of exceptional treasurers.

Because companies don’t hire experienced treasury professionals for their ability to click **“Pay.”**They hire them because they know **when not to click it.**That’s the kind of judgment we’re trying to build.

Automation is becoming accessible to everyone. Judgment won’t.

And Fonder believes that’s where the next generation of Agentic Payments will compete. Not on who moves money the fastest.

But on who makes the best financial decisions.