Latest / Fintech Conversations with Fexingo: Banking, Payments, and Financial Software Companies / How Fintechs Use AI for Real-Time Invoice Discounting
Transcript
- Lucas: If these conversations are useful for what you're building or running, today's episode is a perfect example—we're drilling into a specific fintech tool that's quietly changing how mid-market companies manage cash flow. Luna: Real-time invoice discounting. I've been wanting to talk about this because it's one of those areas where the technology has finally caught up to what finance teams have wanted for years. Lucas: Right. So the old model: a company sends an invoice, waits 30, 60, even 90 days to get paid. If they need cash sooner, they can factor the receivable—sell it to a third party at a discount. But that process was slow, manual, and often opaque. Luna: And expensive. Traditional factoring could take days to approve, and the discount rate was often a flat percentage that didn't reflect the actual risk of that specific buyer or that specific invoice. Lucas: Exactly. What ai enabled platforms do now is assess each invoice in real time. They pull data on the buyer's payment history, the seller's track record, even macroeconomic signals. Within seconds, they offer a dynamic discount rate and can fund the invoice within hours. Luna: So it's essentially algorithmic underwriting applied to receivables. Let's talk about a concrete example—there's a manufacturer in Ohio, about 200 employees, that started using a platform called Taulia for this. Lucas: I know that case. They were selling parts to a large automotive supplier, payment terms were net 60, but they needed cash sooner to buy raw materials for the next order. Their traditional factoring arrangement required a 3 percent fee on every invoice, flat, regardless of who the buyer was. Luna: And with Taulia's AI, they got offered a discount of just 0.8 percent on invoices to that automotive buyer because the buyer's credit was strong and payment history was perfect. The system priced the risk accurately. Lucas: That's the key. The machine learning model ingests data from multiple sources—credit bureaus, bank transaction data, even public records. It learns which buyers consistently pay early or late, and it adjusts the discount rate dynamically. For a risky buyer, the rate might be 4 percent. For a strong one, under 1 percent. Luna: And this isn't just about speed. It's about access. Smaller suppliers who would have been turned down by traditional factors can now get funded because the AI can underwrite based on the buyer's credit, not just the seller's. Lucas: Right. It shifts the risk assessment from the seller to the buyer. That's a structural change. And we're seeing platforms like C2FO do something similar—they operate a marketplace where buyers can offer early payment in exchange for a discount, and the AI matches offers to sellers' cash needs in real time. Luna: C2FO claims that their platform processes over $100 billion in early payment requests annually. That's not small scale anymore. Lucas: No, it's not. And what makes it fintech-native is the integration with open banking. When a seller requests funding, the platform can pull their bank transaction data—with permission—to verify cash flow and flag any anomalies. That reduces fraud and speeds up approval. Luna: Let's talk about the numbers for that Ohio manufacturer. Their days sales outstanding went from 45 days to 12 days after adopting real-time discounting. That's a massive improvement in cash conversion. Lucas: And the cost? They paid an average discount of 1.2 percent across all invoices, compared to that flat 3 percent before. Net savings of about 1.8 percent on every invoice—on $5 million in annual receivables, that's $90,000 back to the bottom line. Luna: Plus the intangible benefit of never having to chase payments or worry about payroll gaps. The CFO told me it changed how they plan inventory. Lucas: Now, let's talk about the technology stack. These platforms typically use a combination of gradient boosting models—like XGBoost—for credit scoring, and natural language processing to extract invoice details like payment terms and line items from PDFs or e-invoicing feeds. Luna: So the AI has to understand the invoice itself, not just the parties involved. That's a hard problem because invoice formats vary wildly. Lucas: Exactly. Some are PDFs, some are XML, some are even paper scans. The NLP layer standardises that data before the pricing engine runs. And the whole pipeline has to operate in sub-second time because the seller is often waiting on the phone or in a dashboard. Luna: There's also a risk management angle. What happens if the buyer defaults after the invoice is discounted? The platform typically takes that risk—it's non-recourse to the seller. So the AI has to be extremely good at predicting buyer default. Lucas: And that's where the models are trained on historical payment data from thousands of buyers. They look at patterns like payment timing, frequency of late payments, industry sector, even macro indicators like regional GDP growth. The best platforms claim default prediction accuracy above 95 percent. Luna: But we should be clear—this isn't a silver bullet. For very small invoices, the fixed cost of the AI processing can eat into the benefit. Most platforms set a minimum invoice size, say $500 or $1,000. Lucas: Right. And there's the integration hurdle. The seller needs to connect their accounting system or ERP to the platform. That's getting easier with APIs, but for companies still on legacy systems, it can be a barrier. Luna: Still, the trajectory is clear. More and more mid-market firms are adopting this. A 2025 survey by PYMNTS found that 42 percent of companies with revenue between $10 million and $500 million now use some form of ai powered receivables financing, up from 18 percent in 2023. Lucas: And I think we'll see that number climb as open banking expands. In the UK and Europe, the second Payment Services Directive, PSD2, already mandates that banks share transaction data with authorised third parties. That makes it much easier for platforms to assess a seller's cash flow in real time. Luna: In the US, it's more fragmented, but initiatives like the Financial Data Exchange are pushing toward standardised APIs. So the friction is decreasing. Lucas: If you're a finance leader listening, the takeaway is: start with your largest, most reliable buyers. Those are the invoices where the AI discount will be lowest, and the savings most immediate. You don't have to onboard all your receivables at once. Luna: And if this episode helped clarify how real-time discounting works, that's exactly the kind of practical takeaway we hope you get from these conversations. Keeping this show ad-free and independent is important to us, and the way that happens is through listener support. If you found today useful, you can buy us a coffee at buy me a coffee dot com slash fexingo. Lucas: It genuinely makes a difference. Every contribution helps us keep drilling into the specifics like we did today. So thank you for listening, and for considering that. Luna: Let's close with one more angle. What about the future? I've seen some platforms experimenting with blockchain-based smart contracts that automate the entire discounting process—triggering payment automatically when the buyer confirms receipt of goods. Lucas: That's a natural extension. If the invoice, the proof of delivery, and the payment are all on a shared ledger, the AI can underwrite and fund instantly with zero manual steps. We're probably 18 to 24 months from seeing that at scale, but the prototypes are promising. Luna: So real-time discounting today, autonomous discounting tomorrow. The cash conversion cycle keeps shrinking. Lucas: And that's the core promise of fintech—not just faster, but smarter. Using data to allocate capital where it's needed, when it's needed, at a price that reflects actual risk. That's a win for everyone in the supply chain. Luna: Well said. That's all for this episode of Fintech Conversations. We'll be back next week with another deep dive. Lucas: Until then, keep building and keep running.