Latest / Fintech Conversations with Fexingo: Banking, Payments, and Financial Software Companies / How AI Is Rewiring Financial Compliance
Transcript
- Lucas: The Securities and Exchange Commission just released its latest annual report, and the headline number that caught my eye was forty-two hundred million dollars in penalties levied last year alone. That is a record high, and it signals something structural about how regulators view compliance now. They aren't just fining you for breaking a rule; they are fining you for not having a system sophisticated enough to detect when someone might be breaking it. Luna: Right, because the old playbook of hiring more analysts to read emails doesn't scale when you have billions of transactions to review daily. Lucas: Exactly. And that is where the conversation usually gets stuck. People think compliance is a legal problem, but really it is a data processing problem. If you are still relying on static keyword triggers to flag suspicious activity, you are going to drown in noise. I spent time with a compliance officer at a mid-sized asset manager last week who told me her team spends eighty percent of their time clearing false positives generated by those old filters. Luna: So basically, the AI isn't replacing the human; it's trying to stop the human from burning out before lunch. Lucas: Precisely. The shift here is from reactive monitoring to predictive behavioral analysis. Instead of looking for a specific phrase like 'kickback' or 'off-book,' which criminals learn to avoid, newer systems map the network topology of transaction flows. They look for anomalies in timing, amount, or counterparty relationships that don't make sense for that specific client profile. It's less about what was said, and more about what the behavior looks like in aggregate. Luna: That sounds powerful, but also incredibly expensive to implement correctly. Who is actually buying this tech right now? Lucas: It used to be only the big banks, the ones with five billion dollar budgets for risk. But the market has fragmented. Now we see mid-market hedge funds and even some wealth management firms adopting lighter versions of these platforms. The barrier to entry has dropped because cloud-native regtech providers offer subscription models. You no longer need to build your own neural network from scratch. You can plug into a service that already understands the baseline patterns of legitimate trading. Luna: But does that mean we're trusting a black box to decide if a trade is suspicious? That feels risky from a liability standpoint. Lucas: That is the critical question. Regulators are asking for explainability. An AI model that says 'this trade is weird' is useless if it cannot point to the three specific variables that drove that conclusion. We are seeing a push toward hybrid models where the machine flags the anomaly, but the logic trail is visible to the human analyst. The goal is augmented intelligence, not autonomous policing. If the algorithm makes a mistake, the firm needs to know why so they can defend it during an audit. Luna: So the real bottleneck isn't the technology itself, it's the ability to interpret the output clearly for a skeptical auditor. Lucas: I would argue it is even deeper than that. It's about data quality. Garbage in, garbage out applies doubly here. If your historical transaction data is messy, unstructured, or siloed across different legacy systems, the AI will learn those inconsistencies as normal behavior. I saw a case where a bank's fraud detection system failed to flag a known money-laundering ring simply because the data migration from their old mainframe had stripped out certain metadata tags. Luna: Ouch. So before you buy the shiny new AI tool, you have to do the unglamorous work of cleaning up ten years of bad records. Lucas: Without a doubt. And that is why many CFOs are pushing back. They see the marketing pitch for ai driven compliance, but they haven't budgeted for the data engineering required to support it. It creates a gap between what the vendor promises and what the internal IT team can deliver. In October two thousand twenty-six, we are seeing a lot of stalled projects because of exactly this mismatch. Luna: Does that mean smaller firms are falling behind? Or is there a way around the data cleanup hurdle? Lucas: Smaller firms are actually moving faster in some ways because they don't have decades of legacy debt. A fintech startup built in the last three years has clean, api first data by default. They can deploy behavioral analytics on day one. The incumbent banks are hampered by their own history. They have to retrofit AI onto systems designed in the nineties. It is a fascinating competitive dynamic where the newcomers have the upper hand in operational agility. Luna: That reminds me of the supply chain finance episodes we did earlier. The startups were always ahead because they weren't carrying the weight of the old world. Lucas: Spot on. And it's not just about catching bad actors. There is a secondary benefit: efficiency. When you reduce false positives, you free up human capital. Those senior compliance officers who used to spend their days clicking through alerts can now focus on complex investigations, policy design, or regulatory liaison work. It changes the job description entirely. Luna: Which probably makes those jobs more attractive to younger talent who want to use data skills rather than just reading forms. Lucas: Absolutely. The career path for compliance professionals is evolving. It is becoming a blend of law, data science, and strategy. If you are a young professional entering this field now, learning how to interrogate AI outputs is just as important as knowing the statutes. The rules change every few years; the data patterns are relatively stable. Luna: So the skill set is shifting from memorization to interpretation. That seems like a healthier place to be. Lucas: It is. But let's talk about the downside for a second. Bias. If your training data reflects historical biases in lending or trading, the AI will amplify them. We have seen instances where automated underwriting systems denied credit to certain zip codes not because of income, but because of correlated spending patterns that the model misinterpreted as risk. Luna: That is a nightmare scenario. Not just legally, but reputationally. One bad algorithmic decision goes viral on social media before the legal team even drafts a response. Lucas: Correct. And unlike a human error, which can be framed as a mistake, an algorithmic error looks like systemic discrimination. Firms are now building 'fairness constraints' directly into their models. They force the AI to ignore protected characteristics and proxy variables. It adds complexity, but it is necessary for long-term viability. Luna: So we are moving from a phase of pure adoption to a phase of refinement and ethical guardrails. Lucas: Yes. The honeymoon period is over. Everyone knows AI works for compliance. Now everyone is arguing about how to control it responsibly. It is getting closer to the mature stage of any major technology cycle. Luna: Before we wrap this up, I want to touch on the investment side. Are investors pulling back on regtech valuations because of these implementation hurdles? Lucas: Not pulling back, but becoming more selective. Early stage funding was wild in twenty twenty-three and twenty twenty-four. Now, capital is flowing to companies that show clear ROI and low churn. Investors want to see that the platform actually reduces the cost per alert, not just that it has a cool dashboard. The metric that matters is operational savings, not just feature count. Luna: So the buzzwords are fading, and the balance sheets are speaking louder. Lucas: Exactly. If you can prove you save a compliance department fifty hours a week, you will get funded. If you just promise to disrupt the industry, you will struggle. Luna: That clarity is refreshing. It forces founders to get real about what their product actually does. Lucas: It does. And honestly, that is good for the entire ecosystem. We need solutions that work, not just solutions that sound impressive in a pitch deck. Luna: If these conversations help you navigate the complexities of business and technology, consider supporting the show. Your contributions keep our research deep and our ads away, allowing us to bring you these detailed breakdowns without interruption. You can find us at buy me a coffee dot com slash fexingo. Lucas: It truly helps us stay independent and focus on the substance. Thanks for listening. Luna: Let's circle back to that point about career paths. Do you think traditional law firms will start hiring data scientists directly into their compliance teams? Lucas: I already see it happening. Large firms are creating hybrid roles where a lawyer works alongside a data engineer. It bridges the gap between legal theory and technical reality. It is the future of how we handle regulatory risk. Luna: A hybrid role makes sense. It ensures the tech serves the law, not the other way around. Lucas: Right. And that synergy is what will define the next decade of financial services. It is not about choosing between tech and law; it is about merging them effectively. Luna: So the question for listeners might be: are you building the bridge between those two worlds yet? Lucas: Or are you still waiting for someone else to build it for you?