Would you let a robot chase your customers?

Train your customers to pay faster.

“Keep the change, ya filthy animal.”

Home Alone... not only is it the greatest movie of all time (fight me), but it also gave a young Secret CFO, watching at home, his first lesson in credit control.

Our hero, Kevin McCallister, is enjoying a well-earned giant bowl of ice cream while watching a gangster film. In it, a man turns up to collect money he’s owed, and his customer responds with a Tommy gun.

An extreme example of the difficulties that can arise after revenue recognition. But even when your customers aren’t armed (I hope), getting paid is still far slower and harder than it should be.

Today, I’ve partnered with Stuut Technologies to explore the future of cash collections, and to find out how much AI can (or cannot) help your business get paid faster.

More cash collected. With fewer collectors.

Stuut uses AI to chase unpaid invoices across email, SMS, and phone, handle customer pushback, and resolve payment issues. Less chasing for your team and more cash back in your business.

Stuut can be live in weeks, helping you attack overdue invoices, bring DSO down, and get cash in before year-end.

I once inherited a finance team with over $70m in overdue debt.

$20m of that would be stubborn and take time to shift, baked into an industry norm of paying a few days late. But that still left around $50m to get after. Yes, our order-to-cash cycle was about as watertight as a house after a visit from Harry and Marv...

My newly inherited Business Unit CFOs and AR teams were falling over themselves, sputtering excuses, rightly embarrassed by such a poor metric.

I needed to see what was happening for myself. So I spent several hours sitting with the front line of this problem: The credit controllers, listening to their calls (true story: My first summer job was working in credit control for a distressed company). 

They didn’t put a foot wrong. They followed a robust escalation procedure, knew when to push and when not to, and finished every call with clear next steps. But our customers’ AP teams were still running rings around them.

Our girls and guys were in a knife fight holding a banana...

Every account with major past-due issues (which was most of them) was a tangle of confusion: Price discrepancies, quality disputes, missing order numbers, etc. The worst account had over 200 new dispute notices landing every single day. No... not a typo.

These were all issues that started a long way upstream from collections.

Our customers had every excuse not to pay and had learnt how to use it. We needed to fix our side of the mess, then teach them their same behavior was unacceptable. 

It took six months to get overdue debt below $20m. But to get there, we had to make it business issue number one, not just in finance but in sales and operations too. All those calories spent getting paid for work we’d already done, rather than where it should be: Creating amazing products and serving customers.

Good O2C management is not something you can switch on and off. Doing it well is relentless, and mostly thankless.

This is why I think it’s an interesting area for AI... because agents do relentless very well (and don’t need thanking, although it would be polite).

I sat down with Ben Winter, co-founder of Stuut, an AI coworker for order to cash, to ask him why exactly getting paid is still so difficult... and whether AI can really make a dent.

“Companies are really good at selling stuff and terrible at collecting cash,” Ben told me. He co-founded Stuut after spending years in procurement. He was helping companies buy from suppliers but realized that on the other side of every transaction was a company trying to get paid.

It was an insight that... ahem, paid off.

Earlier this month, Stuut announced a $52.5m Series B led by Insight Partners, with participation from a16z. They’re also the proud custodians of the joint cutest mascot in finance.

Ben founded Stuut on the insight that overdue debt starts long before finance sees it.

“From the moment you order something to when the cash hits your bank account, information gets lost across systems,” Ben says. “And finance gets the short stick because they just have an invoice number, an amount, and a due date.”

“But if you can keep all of that information from disappearing, collect it, and make it usable for the finance team, you can fix the mistakes that keep cropping up, first in the AR ledger, and then at source.”

“Part of what we do is help centralize that and give you the ability to understand and analyze that data,” Ben says. “And then once you can analyze that data, your world becomes totally different.”

There’s a phrase in M&A: “Every deal dies a thousand times.”

Countless obstacles appear between agreeing the deal and closing. Unless someone takes ownership of removing them, the process becomes endless.

Getting a trade debt paid works much the same way.

You can’t control everything your customer does. But you can control whether your invoice is accurate, whether you can prove delivery, whether someone owns the dispute, and whether you follow up when the promised payment doesn’t arrive.

Your starting assumption should be that customers won’t pay until you’ve removed every reason they have not to.

I own 50% of an industrial chemical business (long story).

In our monthly board updates, our general manager, Matthew, will talk about overdue debt with a note of surprise. As though payment was inevitable until something unexpectedly got in the way. I keep reminding him: removing obstacles is the job.

Matthew, if you’re reading this, I love you. But get the team on the f*cking phones. Find out what’s stopping payment, fix our bit, and chase theirs (or I’ll set lil Stuut on you).

Ben at Stuut was clear where the problems start: “Most issues come upstream, in the order-management space, and multiply from there.”

Remember the account generating 200 new dispute notices a day I mentioned earlier?

When we got underneath it, they were all caused by just FOUR underlying product master data issues. One was our error. Three were the customer’s.

Four errors mucking up every delivery and distribution point. Our collections team was drowning in 200 new problems a day, all caused by the same four issues.

But... even if you get it all correct: Agree clear terms, keep the data clean, get the orders and invoices right, and resolve disputes quickly, you can still be left waiting for your money.

Someone at the customer still has to send you payment. But they’ve got other suppliers breathing down their necks at the same time. So, who gets paid?

“The squeaky wheel gets the oil,” Ben says.

I’ve been on both sides of that ledger enough times to know he’s right. Customers learn who chases, who escalates, and who lets it slide for another couple of weeks.

But letting them slide only teaches them your tolerance for unpaid bills.

There is real working capital alpha in just being boring, predictable, and relentless in your credit control routines.

Collecting a trade debt is like concrete... the longer you leave it, the harder it gets. Ben sees this as a hangover from the “bad habits during the ZIRP era,” when interest rates were near zero.

Cheap borrowing made a few extra days of overdue debt easier to shrug off. And then eventually to agree to longer and longer payment terms.

But shit gets expensive when you’re paying real money to fund your customers’ working capital.

A recent a16z report, drawing on Stuut’s data, put US non-financial businesses’ trade receivables at $7.2 trillion at the end of 2025.

Not all overdue, of course. Much of it is ordinary trade credit. But what struck me was the lack of any obvious correction as interest rates rose sharply in 2022–23. Extending credit to customers suddenly got a lot more expensive, and it looks like we CFOs... did nothing:

Let’s put some numbers on that.

You’re a business with $500m of annual sales and $100m tied up in net working capital because you are offering tragically generous credit to your customers.

If your cost of funding rises by four percentage points, carrying that same balance costs an extra $4m a year. At a 10% EBIT margin, you make $50m of operating profit, and that additional financing cost eats 8% of it. Ouch.

You’d expect that to create pressure to shorten credit terms. But there’s no obvious sign of it in the aggregate numbers.

And in my experience, long payment terms can create another problem: Complacency over invoice accuracy. There’s “plenty of time” to fix the pricing error. Find the missing PO. Resolve the dispute. Until suddenly there isn’t, and the person who could have fixed it has forgotten about it, moved jobs, or lost the paperwork.

As I say... debt is like concrete.

It’s also why I’ve never loved DSO as a standalone measure. It mixes together two different questions: How much credit have we agreed to give? And are customers paying when they agreed to?

I’d rather look at weighted average credit terms alongside conformance to those terms. What did we agree? What actually happened?

Then break that down by customer, channel, collections officer or sales rep. You can start to see whether you’re waiting because of the deal you made or because something is broken in the cycle.

So speed and analysis is important... which brings me to AI.

Now, the irony is that I put machine-learning-powered automation into my AR teams ten years ago.

The pitch wasn’t wildly different to the AI pitches you hear today. It learns, gets better, and takes manual work off your team.

And it did a pretty good job (within some very narrow guardrails).

So my question for Ben was: What can an AI-powered AR agent like Stuut do that those systems couldn’t?

“The old tools could flag a problem, but somebody still had to call the customer and sort it out,” Ben says. “Now an AI agent can have that conversation, send over a missing invoice, and offer a payment plan within the rules you set.”

And when the conversation exceeds its information or authority, it can bring in a person... live if you want.

Ben pointed out that every customer needs something different. And every customer interaction will be different. That multiplies the complexity across the entire process. Stuut is aiming to address that complexity further upstream, connecting contracts, price books, and orders across the full lifecycle.

“I think people just need to reimagine what’s possible in Order to Cash now. People are still fixed on pre-agentic assumptions,” Ben told me.

So let’s cut to the chase: This does mean letting a bot call your customer to chase payment. And how would you feel about that?

Let’s be honest, shitty automated customer service from your bank, airline, or cell phone provider has spent years destroying your confidence in a machine’s ability to understand basic instructions, let alone resolve discrepancies.

Press three... repeat your account number... explain the problem for the fourth time... wait for a human... human doesn’t pick up... back round the loop.

So “a robot is going to call your customers” might not immediately sound like progress.

But the quality of AI-powered voice-to-voice conversations is staggering. I’ve played with this a lot while building my CFO role simulator, SimCFO. The technology at the cutting edge is light years ahead of the Alexa in your kitchen (which needs shouting at seven times just to play the new Strokes record).

In Stuut, you can choose from 27 voices, with at least 11 named accent styles, and the agent can hold conversations in 26 languages. If you have an export business, you’ll appreciate how useful that could be.

But a robot that sounds lovely while knowing nothing about your account is still completely useless.

The bigger change is context. Connected to your ERP, CRM and billing systems, Stuut can check the invoice, the payment history, and the correspondence while having the conversation. And then fire off the follow-up emails automatically.

Inside Stuut, you can set what the agent is and isn’t allowed to agree to. It might be authorized to accept a promise to pay by a particular date, or below a certain size. Or you can require it to defer that decision to a human. You can also configure escalation workflows: Start with emails, move to AI calls, then bring in a person when needed.

Ben says the key to success with agentic collections is giving the agent the information it needs to address the customer’s actual obstacle to payment.

“AP teams tell us they like it more than a human,” he says. “Do you want to know why? Because the agent answers the questions and gets out of the way. No more ‘press 0 to get jack shit.’”

The point is to give humans more time for the genuinely problematic accounts. The sensitive relationships. The decisions that require judgment about whether to agree to a payment plan, settle a dispute, or stop supplying a customer.

Ben has an interesting theory about how this could change ownership of cash collection.

“Eventually every seller should have an AI agent that’s managing their book of business, helping them collect on time and making it efficient,” Ben says.

In that model, finance handles the exceptions: Delinquent debts, sensitive negotiations, and decisions that involve a trade-off between P&L and cash flow.

That’s interesting to me. I’ve always pushed for sales commissions to have some kind of cash collection component. Salespeople need to understand that the job isn’t finished when the revenue is booked. We still need the money in the bank, and they share significant responsibility for getting it there.

So maybe sales owning collections, with an agent doing the running, is the natural end state of that thinking?

Finance would still handle the escalations and the judgment calls. But it puts ownership closer to the customer relationship. And to where so many of the obstacles to payment begin.

A lot of CFO tech comes with a substantial implementation lift.

But what excites me about agentic order-to-cash is that getting started is very easy, by comparison.

Ben says a typical Stuut implementation takes two to three weeks, with smaller customers sometimes going live in about a week.

Connect your ERP and CRM through APIs, set the agent's permissions and collection workflows, then put it to work. You can start observing results in days: Customer responses, disputes moving forward, and the thing we’re ultimately here for... cash received.

Getting connected isn’t the same as delivering value, of course. But because the feedback loop is short, you can quickly see whether it’s helping you get paid faster.

And even where an agent can’t resolve the problem, understanding the patterns in what’s stopping payment has value too.

If the same product keeps generating pricing disputes because a price file is wrong, the agent can flag the pattern... before you’ve spent weeks fielding 200 dispute notices a day.

Reading across messy correspondence, connecting recurring complaints, and identifying a likely common cause is the kind of work LLMs are well suited to. A human still needs to check the diagnosis and fix the price file. But they have a much earlier, better place to start.

The best way to decide how you feel about this technology is to try it yourself. Role-play as a customer and let Lil Stuut give you a call to chase you for payment.

Get a free call from Lil Stuut about your overdue debt

Seriously try it. It takes two minutes. And there is no better way to decide whether this is the future of Accounts Receivable (big thanks to the Stuut team for setting this up for Secret CFO readers).

Give the agent a hard time. Dispute the delivery. Say you never received the invoice.

See what happens. Then hit reply to this email and tell me what you think.

I’ve had a lot of fun testing it, but the phone calls are a smaller part of what interests me here.

The bigger opportunity is understanding why you’re not getting paid at a systemic level. Connecting hundreds of apparently separate reasons back to the handful of product or customer specific broken processes causing them.

Remember those 200 dispute notices a day? With just four easy to fix underlying issues. That’s the kind of insight I’d have wanted on day one.

Meanwhile, good luck getting your cash home in time for Christmas. Hopefully you won’t need John Candy and a polka band.

Want to know what it’s like to be chased for payment by Stuut? Get a call about a fictional overdue invoice and experience it for yourself, right now. It takes a couple of minutes

Disclaimer: I am not your accountant, tax advisor, lawyer, CFO, director, or friend. Well, maybe I’m your friend, but I am not any of those other things. Everything I publish represents my opinions only, not advice. Running the finances for a company is serious business, and you should take the proper advice you need.