Latest / Fintech Conversations with Fexingo: Banking, Payments, and Financial Software Companies / How AI Agents Are Rewiring Corporate Treasury Management
Transcript
- Lucas: Most people think about artificial intelligence in finance as a dashboard feature — a fancy chart that tells you where your money went last month. But if you talk to the heads of corporate treasury at any major enterprise right now, they aren't looking at dashboards. They are watching their software make decisions before they even wake up. Luna: So we have moved past the 'analysis paralysis' phase? You mean actual execution? Lucas: Exactly. We are seeing a rapid deployment of autonomous agents that don't just predict cash shortfalls; they execute the trades, move the liquidity, and hedge the exposure automatically. It is no longer about giving the CFO better data. It is about handing the keys to the kingdom to an algorithm that operates on milliseconds, not business days. Luna: That sounds terrifyingly efficient. What happens when the model gets the signal wrong? Lucas: That is the million-dollar question, isn't it? Because in traditional treasury, there is a human in the loop who can say 'wait, this looks weird.' With these new systems, the latency between detection and action is so low that by the time a human notices something is off, the transaction is already settled. We need to talk about how this shifts the definition of risk. Luna: It shifts risk from operational error to model failure. Which is much harder to debug. Lucas: Precisely. And before we get too deep into the weeds of neural network architecture, I want to pause for a second. If these conversations are useful for what you're building or running, honestly, if today was worth a coffee to you, that's the link — buy me a coffee dot com slash fexingo. It keeps the lights on and the research sharp. Luna: Appreciate it. And yeah, keeping the ad-free format alive matters more than ever with this kind of niche content. Lucas: Right. So let's go back to the core tension here. In September of twenty twenty-six, interest rates have stabilized enough that the cost of capital isn't the only variable. Speed is. And these AI agents are optimizing for speed and yield simultaneously. Luna: But yield optimization used to be a manual process of shopping around banks. How does an agent do that without getting front-run? Lucas: They use what we call dynamic routing. Instead of sitting in one account, the cash is constantly fragmented across dozens of institutional partners based on real-time rate differentials and credit ratings. The agent moves dollars every hour, sometimes every minute, chasing half a percent efficiency gains that add up to millions over a year. Luna: So it is essentially high-frequency trading applied to idle corporate balance sheets. That feels like a regulatory grey area. Lucas: It absolutely is. Regulators are still catching up. Most frameworks were written for humans making deliberate choices. They weren't written for an algorithm that views liquidity as a fluid state rather than a static number. The SEC and the Federal Reserve are currently debating whether these agents should be classified as trading entities or just payment processors. Luna: And that classification changes everything regarding compliance and reporting requirements. Lucas: Correct. If they are traders, they need capital reserves. If they are processors, they don't. The industry lobby is pushing hard for the processor classification because it lowers the barrier to entry for smaller fintechs building these tools. Luna: So we might see a wave of aggressive adoption despite the uncertainty. Companies will just bet on the efficiency gains. Lucas: They are already doing it. Look at the mid-cap manufacturing sector. These are companies that never had a treasury department bigger than three people. Now they are using SaaS-based AI treasuries that compete with the systems used by Fortune fifty firms. Luna: Democratization of sophisticated liquidity management. That is a compelling narrative for investors in the B2B fintech space. Lucas: It is. But here is the catch that most pitch decks ignore. These models require massive amounts of historical transaction data to train. A company with messy, unstructured ledger entries is going to get garbage results. So the first winners aren't necessarily the ones with the smartest AI, but the ones with the cleanest data infrastructure. Luna: Which means the legacy ERP providers are actually the gatekeepers here. If they don't integrate well, the AI agents can't function. Lucas: Spot on. This is why we are seeing partnerships between cloud ERP giants and specialized AI treasury startups. It is not a replacement play. It is an augmentation layer. The AI doesn't replace the ERP; it sits on top of it, reading the API feeds and pushing commands back through the same channels. Luna: So the value prop is cleaner than the tech stack suggests. It is really about connectivity. Lucas: Connectivity and trust. Because you have to give the AI permission to move money. There is a psychological barrier for CFOs who are used to signing off on wire transfers personally. Even if the math is better, the liability feels heavier. Luna: Liability for what though? If the system follows the rules set by the CFO, isn't the outcome just optimized business? Lucas: Ideally, yes. But consider counterparty risk. If the AI decides to move two hundred million dollars to a new regional bank because the yield is slightly higher, and that bank fails overnight, who is responsible? The CFO who approved the parameters, or the vendor who built the model? Luna: That is a nightmare scenario for legal departments. They are going to push back hard on these deployments. Lucas: They are. And they are demanding what we call 'explainable AI' in treasury. Not just 'the bot moved the money,' but 'the bot moved the money because factor A, B, and C correlated to a predicted shortfall in D hours.' Luna: Explainability reduces the fear factor. It turns the black box into a white board. Lucas: Exactly. And the vendors are responding. The best platforms now provide a natural language audit trail. You can ask, 'why did you hedge this currency position?' and it will show you the volatility spike and the forward rate curve at that exact moment. Luna: That is a game changer for internal audit teams. They usually hate these new systems because they can't trace the logic. Lucas: It solves the audit problem. But it creates a new problem: alert fatigue. When the system is making thousands of micro-decisions a day, the human operator needs to know which ones matter. If every decision is flagged, nobody watches. If none are flagged, mistakes slip through. Luna: So the user interface becomes the critical control point. Designing the exception handling workflow is more important than the algorithm itself. Lucas: I would argue it is the single most important product feature right now. The UI determines whether the CFO trusts the system enough to leave it alone. If the interface is clunky or alarmist, they will revert to manual processes, and all that efficiency evaporates. Luna: Efficiency is useless if it comes with friction. So the competition is shifting from backend modeling to frontend experience. Lucas: Yes. And we are also seeing a bifurcation in the market. On one end, you have the hyper-automated systems for large multinationals with complex cross-border flows. On the other, you have simplified 'set-and-forget' tools for small businesses that just want to earn interest on their checking accounts. Luna: The retailization of treasury management. Everyone wants their spare cash working harder. Lucas: Right. And for small businesses, the risk profile is different. They aren't worried about basis points on a billion-dollar portfolio. They are worried about paying payroll on Friday. So the AI has to be extremely conservative with liquidity access, prioritizing safety over yield. Luna: Safety first, yield second. That is a very different constraint set than what you described for the Fortune five hundred. Lucas: Completely different. One is an optimization problem. The other is a solvency problem. And the algorithms reflect that. Small business tools often have hard stops — if cash drops below X, the AI freezes all investment activity immediately. Luna: Hard stops are good for peace of mind. They remove the ambiguity of 'how much risk am I taking?' Lucas: They do. But they also cap the upside. There is a constant tension between security and return. And as these systems get smarter, they will likely find ways to offer higher yields without increasing tail risk, but that requires incredible accuracy in forecasting. Luna: Forecasting cash flow is notoriously difficult. Especially with the variability in modern supply chains and consumer behavior. Lucas: That is why the integration with external data sources is key. These agents aren't just looking at your bank statement. They are pulling in weather data, shipping logistics updates, and even social sentiment to predict revenue spikes or delays. Luna: Holistic environmental scanning. Treating the business ecosystem as part of the liquidity equation. Lucas: Exactly. It is a systems-thinking approach applied to finance. You can't manage cash in isolation anymore. You have to manage it in the context of the entire operational reality of the company. Luna: This feels like the natural endpoint of digital transformation. Everything connected, everything optimizing. Lucas: It is. But it forces us to rethink what a treasurer does. Their job is no longer about moving money. It is about defining the rules for the machine. It is strategic governance rather than tactical execution. Luna: A shift from operator to architect. That is a significant career evolution for finance professionals. Lucas: It is. And those who adapt will be incredibly valuable. Those who cling to the old spreadsheet models will find themselves obsolete very quickly. Luna: Obsolete, or just redundant. Either way, the bar for competence just got raised. Lucas: Raised significantly. But the payoff is huge. We are talking about freeing up human capital to focus on strategy, growth, and innovation, while the machines handle the boring, repetitive, and complex arithmetic of survival. Luna: If the machines don't crash first. That remains the elephant in the room. Lucas: The elephant, the ghost in the machine, the unknown unknowns. But progress rarely waits for perfection. It waits for utility. And the utility here is undeniable. Luna: Undeniable. So the question is no longer if we will adopt this, but how fast we can build the guardrails to support it. Lucas: Building the guardrails. That is the work for the next decade. And for the CFOs listening, it starts with cleaning your data and trusting your team to learn the new tools. Luna: Clean data, clear rules, and a lot of confidence. Sounds like a recipe for success. Lucas: Or a recipe for disaster if you skip the basics. But assuming you get the foundation right, the future of treasury is already here.