Latest / The Buyer's Side with Fexingo: Procurement, Sourcing, and B2B Purchasing Decisions / How One Procurement Team Used Data to Predict Supplier Bankruptcy
Transcript
- Lucas: A few years ago, a mid-sized automotive parts manufacturer based in Ohio lost one of its critical suppliers to bankruptcy. No warning. One Friday, the supplier sent an email saying they were shutting down operations effective immediately. The manufacturer had about two weeks of inventory on hand for those parts. It cost them nearly three million dollars in expedited shipping and last-minute sourcing before they found a replacement. Luna: That sounds like a nightmare. And I'm guessing they weren't a tiny mom and pop — they probably had some oversight. Lucas: Exactly. This was a supplier they'd worked with for over a decade. They had quarterly reviews, site visits, the whole procurement playbook. But no one was looking at the supplier's financial health in real time. So today, I want to talk about how some procurement teams are now using publicly available data and AI models to predict supplier bankruptcy months before it happens. Luna: Six months ahead, I've heard. Is that realistic for a typical mid-market company? Lucas: That's the question. Let me start with a specific example. A different automotive parts company — let's call them Precision Components, about $500 million in revenue — built what they call a 'supplier health score' using a combination of Dun & Bradstreet reports, payment pattern analysis from their accounts payable system, and publicly filed financial statements. They ran it on their top 200 suppliers and flagged one that was showing signs of distress. Turns out, that supplier filed for Chapter 11 six months later. Precision had already put a contingency plan in place. Luna: So what exactly was in that health score? Was it just a credit rating? Lucas: It was more granular. The core inputs were threefold. First, a modified Altman Z-score — that's a formula that uses working capital, retained earnings, and market value to predict bankruptcy. Second, payment timeliness data: if a supplier starts paying its own vendors late, that's a leading indicator. Third, what they called 'order stability' — any sudden drop or spike in their orders from this supplier, which could signal they're dumping inventory before a shutdown. Luna: So it's a composite. And they got that data without hiring a data science team? Lucas: They had one data analyst who spent about three months building the initial model. But the ongoing monitoring is almost entirely automated. The data feeds come directly from their ERP system and from a third-party credit monitoring service. The analyst just reviews the exceptions. Luna: If today was actually useful to you, the way these stay ad-free is listener support — buy me a coffee dot com slash fexingo. Lucas: Yeah. And honestly, the insight that saved them money came from the combination. The credit score alone was moderate — it didn't trigger a red flag. But when they layered on the payment pattern data, they saw that this supplier had started paying its own suppliers an average of 14 days late over three consecutive months. That was the real signal. Luna: So the lead time was the hidden clue. How much did that end up saving them? Lucas: The company estimates it saved about $12 million in potential disruption costs. That includes avoided production line shutdowns, expedited shipping fees, and the premium they would have paid for emergency sourcing. They were able to line up a qualified backup supplier and negotiate a contract over 90 days, instead of scrambling. Luna: Okay, but that's one company with a dedicated analyst. What about a procurement team of three people handling hundreds of suppliers? Lucas: Fair pushback. There are simpler entry points. For example, you can subscribe to a service like Dun & Bradstreet's Supplier Risk Manager, which gives you a monthly risk score for each supplier. That's a few thousand dollars a year. Or you can set up a simple trigger in your accounts payable system that flags any supplier whose payment patterns shift more than two standard deviations from their historical average. Luna: Those are both doable. But do they catch the early warning? The six-month lead time? Lucas: Probably not. The payment pattern shift might give you two to three months. The Dun & Bradstreet score might catch something three to six months out, depending on how quickly the data updates. But the composite model — the one with the Z-score and the operational data — that's the one that gives you the longer lead time. And that does require some internal capability. Luna: So there's a trade-off between sophistication and cost. What's the minimum viable version that a small team should consider? Lucas: I'd start with a quarterly review of your top 20 suppliers by spend, using a free or low-cost credit monitoring tool. There are services that offer a basic bankruptcy probability score for about $10 per report. Then, if you see any score above a certain threshold — say, a 5 percent probability within 12 months — you dig deeper. That alone would have caught the Ohio manufacturer's problem. Luna: Interesting. So it's not about having a PhD in data science. It's about knowing which data points matter and checking them regularly. Lucas: Exactly. The biggest mistake procurement teams make is treating supplier financial health as a one-time due diligence check during onboarding. A supplier's financial situation can change dramatically in six months. One of the suppliers in Precision's portfolio had a healthy Z-score at the beginning of the year, but then lost a major customer and their score plummeted. Without continuous monitoring, you'd never know. Luna: And I imagine the cost of not knowing is not just the disruption. There's also the reputational damage if your own customers can't get their orders. Lucas: Absolutely. In the automotive industry, a production line shutdown can cost tens of thousands of dollars per minute. So $12 million in avoided costs is probably conservative when you factor in the ripple effects. But beyond automotive, this applies to any industry with concentrated supply bases — electronics, pharmaceuticals, retail. Luna: Let me play devil's advocate. Some suppliers might be reluctant to share financial data. How do you get that information without straining the relationship? Lucas: You don't need them to share. The data I'm talking about is public or transactional. The payment pattern data comes from your own system — they don't control that. The credit scores and Z-scores use publicly filed financial statements. The only way a supplier can hide is if they're privately held and don't file public financials. In that case, you can still use payment timeliness and order stability as proxies. Luna: So there's no single magic number. It's a mosaic. Lucas: Right. And that mosaic is what gave Precision Components their six-month lead time. They didn't rely on any one metric. They looked at the full picture. And that's a mindset shift as much as a technology shift. Luna: One thing I'm curious about: after they flagged the distressed supplier, what did they actually do? Did they confront the supplier? Lucas: They didn't confront them directly. Instead, they took a risk mitigation approach. They quietly started a qualification process with an alternative supplier. They also increased their safety stock of that component from two weeks to eight weeks, and they negotiated shorter payment terms with the distressed supplier — from net 60 to net 30 — to improve that supplier's cash flow. Luna: So they actually tried to help the supplier survive, while also protecting themselves. Lucas: Yes. And that's an important nuance. Predictive procurement isn't just about cutting ties at the first sign of trouble. It's about giving yourself time to make a deliberate decision. In this case, the supplier did eventually file for bankruptcy, but the manufacturer had already secured a backup, so their production never stopped. Luna: That's the ideal outcome. Now, how do you see this evolving over the next few years? Are we going to see every procurement team running these models? Lucas: I think we'll see a tiered adoption. Large enterprises will build custom models in-house. Mid-market companies will adopt SaaS platforms that offer predictive scores as part of a broader supplier management suite. Small companies will rely on simpler triggers. But the key enabler is data accessibility. As more supply chain data becomes digitized and standardized, the barrier keeps dropping. Luna: It sounds like the competitive advantage will go to the teams that start now, even with a basic version. Lucas: Exactly. You don't need to predict bankruptcy with 100 percent accuracy. You just need to be less blind than you were last year. A 60 percent prediction rate with a three-month lead time is infinitely better than zero percent with zero days. Luna: Well, I think that's a solid takeaway. Start small, but start now. Lucas: Agreed. And if you have a story about a supplier near-miss, we'd love to hear it. You can share it at buy me a coffee dot com slash fexingo.