Lars

AI for Credit Management

[Lars @larsewallin]

Every AI application has the same weak point: the data it's built on. Get that wrong, and you get AI slop, confident answers sitting on shaky ground.

Yesterday I wrote about CreditPilot, an open source AI agents project for credit management.

Here's how it avoids that problem. Every agent writes what it finds into one shared log called Credit Events, the raw feed of everything happening across the customer portfolio.

But nothing gets into that log for free. Every Credit Event has to pass through a gateway function first. That gateway checks it against a strict, predefined shape: does it have a real event type from the fixed list, a severity, a summary, a source. Miss one, and it's rejected.

That gateway is also what lets more agents get added without anything breaking. A new agent doesn't invent its own way of writing to the log, it just plugs into the same gateway, follows the same rules, and its findings land in the same shared feed as everything else.

GitHub repo and a demo company are up at creditpilot[.]vercel[.]app

[Lars @larsewallin]

Hired a fourth analyst for CreditPilot: the Payment Behaviour Monitor.

https://t.co/HXzDWP2Wpu

The first three watch what's visible from the outside, headlines, filings, credit limits. This one watches something quieter: how a customer's payment habits actually shift over time, not just whether an invoice happens to be late today.

It compares two 30-day windows per customer, the current 30 days against the 30 before that. Average days late rises by 5 or more, that's deterioration. Falls by 5 or more, improvement.

And if payment dates start swinging unpredictably, some early, some very late, no clear direction either way, that gets flagged separately as volatility.

Same rule as every other analyst: it only writes to the shared Credit Events log. Doesn't touch anything else, doesn't talk to the other analysts, doesn't decide what happens next.

That's four analysts, more to come.

[Lars @larsewallin]

I recently started exploring an idea: building AI agents for credit management. The goal was, and still is, to see what could be built as internal tools rather than a SaaS product.

I'm calling the project CreditPilot.

I used Claude to collaborate on the structure and Claude Code to actually build everything.

It's open source, so anyone interested can check it out on GitHub. I also set up a company with dummy data to test how everything works. Both are available at: creditpilot[.]vercel[.]app

How it's built:
Think of the agents as a team of always-on analysts, each watching one specific thing. None of them talk to each other directly. They all just write what they find into one shared log called Credit Events, the raw feed of everything happening across the customer portfolio in real time.

When something in that feed is serious enough to warrant a decision, it becomes a proposed Action: a specific recommendation, with its reasoning attached, waiting for someone to approve or reject.

The point of this structure is that it's easy to "hire" more analysts, and each new one knows exactly what to do from day one.

I started with two fairly basic analysts: one monitoring the news, and one monitoring SEC filings. The third is an AR Aging analyst, watching credit limits, late payments, and the like.

Let me know what you think, or what the next analyst should do :)

[Lars @larsewallin]

@Code_Siam YT

[Lars @larsewallin]

@iam_chonchol AI

[Lars @larsewallin]

@AIToolReport SEND

[Lars @larsewallin]

@mordechaihoff @hnshah Too late! They have too many active users.

[Lars @larsewallin]

@SievaKozinsky So valuation based on cash flow is a crime??

[Lars @larsewallin]

@sulemanali @Bounce21NYC

[Lars @larsewallin]

@englishpaulm Never Say Never Again

[Lars @larsewallin]

@paulg @naval 😂😂

[NHL @NHL]

Game respects game. 🤝

The @MapleLeafs legend, Börje Salming, made his return to Toronto for the @HockeyHallFame induction ceremony.

Salming was the first Swedish-born player inducted into the Hall of Fame and remains the highest scoring defenceman in Maple Leafs history. 💙

[Lars @larsewallin]

@SecretCFO @hnshah Couldn’t you make a similar argument/thread about any stand alone key ratio? Revenue, Equity, EBIT, etc… “Revenue is a terrible way to measure a company’s size”
“Equity is a terrible way to measure a company’s financial health”

[Lars @larsewallin]

@5harath @beehiiv is worth checking out. Good luck!!

[Lars @larsewallin]

@tobyallen007 @Acquire_Me Awesome! Congrats on the launch 🚀

[Lars @larsewallin]

@DamianJanzi @matteomosca_ @bubble 🙌

[Lars @larsewallin]

@whit4th Congrats! Enjoy! It will be back home in the Bronx next year😉

[Lars @larsewallin]

@chazza @microacquire Congrats and good luck with whatever is next! 🍾🙌🚀

[Lars @larsewallin]

@shikuwangombe Keep crushing it @shikuwangombe 🚀🙌