The conversation in San Francisco was with marketing and compliance leaders navigating the CFPB, the SEC Marketing Rule, and FINRA. Important, complicated, and in many ways familiar territory for anyone who's spent time thinking about regulated financial services marketing.
Chicago was different.
Front Row Chicago was an off-the-record breakfast at Beatrix in the Loop. The two panellists were Jeremy Dela Cruz , Senior Manager of Global Marketing Compliance at Kraken , Payward and NinjaTrader , and Ed Ryan , CCO and COO at Ironbeam, Inc. . Both sit in CFTC-regulated businesses. Both run compliance functions where every piece of marketing content carries mandatory risk disclosures, every approval has a named owner, and the NFA or CME can audit the decision trail. The stakes are visible, the accountability is named, and the conversation reflected that.
One thing worth noting before we get into what was said: NinjaTrader is an Adclear client. Jeremy wasn't speaking to us as a prospect. He was speaking as a practitioner who uses the product and has been thinking hard about AI in compliance for longer than most. That context shaped the conversation.
AI is changing what's operationally possible, not just how fast things move
Jeremy described three active AI use cases at NinjaTrader: pre-publication review of marketing materials, post-publication monitoring of affiliate and influencer content, and regulatory summarisation.
The pre-publication piece is what most people expect, AI catching missing disclosures, misleading claims, and obvious problems before a human reviewer sees the content. Review time down from weeks in some cases. That's valuable, but it's not the most interesting part.
The monitoring piece is more interesting. Jeremy described the shift from sampling, where a firm reviews a percentage of affiliate posts and hopes the sample is representative, to being able to scan an entire YouTube channel in a day. Not a sample. The whole channel. That isn't an incremental improvement in the same process, it's a different process entirely, and the compliance question changes with it. Instead of "what can we afford to review?" it becomes "what actually needs human attention?"
The regulatory summarisation use case is the quietest of the three and possibly the most underrated. Summarising FINRA and NFA disciplinary events, tracking rule changes, and producing management reporting on regulatory developments are tasks that used to take days and now take minutes. The people freed up by that are the same people who should be focusing on the harder compliance calls.
Ed's use of AI at Ironbeam sits at an earlier stage, with the review process still primarily manual. But he raised a forward-looking use case that generated the most discussion in the room: real-time monitoring of live content. Firms running live streams face a compliance problem that pre-approval processes weren't designed to handle. Getting a script approved by the NFA can take two weeks, which makes spontaneous live content structurally difficult for regulated firms. Ed's thesis is that AI monitoring live audio in real time, listening for unbalanced risk/reward claims and flagging misleading statements as they're made, could unlock a content format that firms currently avoid. That's a different kind of compliance infrastructure than anything on the market today.
The CFTC has said so in two sentences
Ed raised this early and it set the tone for the rest of the conversation. The U.S. Commodity Futures Trading Commission 's position on AI is, in effect, two sentences: firms are fully responsible for what AI produces, and there is no regulatory carve-out for AI-generated or AI-reviewed content.
Jeremy made the same point from a slightly different angle. Regulators supervise firms and individuals, not software. If AI reviews content and a human approves it and it goes out wrong, the regulator will look to the named approver and the firm. The machine being in the workflow doesn't change who's accountable.
That framing had a practical consequence for how both panellists think about AI adoption. The question worth asking isn't "can we get the AI to approve this?" It's "can we use AI to make our human reviewers faster and their decisions better-documented?" Those are different product requirements, and they're the ones that actually matter for firms that will eventually face a regulatory examination.
Ed was notably cautious about generative AI specifically. Hallucination risk in compliance content is a different problem from hallucination risk in a customer service chatbot. If a compliance tool confidently invents a rule, produces a disclosure that doesn't match the actual requirement, or misrepresents the current regulatory position, the firm that relied on it owns that mistake. His concern wasn't about AI in general. It was about the gap between what generative AI is marketed as and what it can reliably do in a high-accountability context.
The hardest question: who owns an AI that makes financial decisions?
This was the most unresolved part of the conversation, and in some ways the most important.
The discussion of AI in compliance review is relatively settled, at least in its structure. AI reviews content, a human signs off, the firm is accountable. That model has friction and open questions, but everyone in the room understood its shape.
The model for AI making financial decisions is not settled at all.
The question was raised in the context of AI chatbots that give investment guidance, manage portfolios, or place trades autonomously. If a third-party AI system does those things, who owns it? Is the vendor effectively acting as a Commodity Trading Advisor and if so, should they be registered? If the regulated firm deploys the vendor's system and it makes decisions on behalf of clients, where does the regulatory responsibility sit?
Ed said this came up at a conference in Baltimore earlier this spring. There was no clear answer then. There still isn't one now.
Interactive Brokers and Robinhood are already moving toward AI-assisted portfolio management. The question isn't hypothetical, it's already arriving. The regulatory model for AI in compliance review exists, imperfect as it is. The regulatory model for AI making investment decisions doesn't yet.
What good looks like when evaluating AI vendors
Jeremy gave three criteria for how NinjaTrader thinks about AI vendors in the compliance space, and they're worth repeating because they're the right ones.
First, explainability. When a regulator asks how a compliance decision was made, "the AI said yes" is not an acceptable answer. Firms need to be able to explain the logic, the criteria applied, and the version of the rules in use. A vendor that can't provide that transparency leaves the firm carrying regulatory risk it hasn't accounted for.
Second, record-keeping. Can the system reproduce the results of a given review for an examiner months or years later? Audit trails aren't optional. If the AI reviewed content but can't show its work in a form an examiner can follow, the investment in AI hasn't solved the evidence problem, it's just moved it.
Third, partnership pace. AI regulation is moving fast and vendor product roadmaps need to move with it. A firm buying a compliance AI tool today is making a bet that the vendor will still be the right partner in eighteen months. The pace at which a vendor responds to regulatory developments, updates its models, and shares what it's learning matters as much as what the tool does today.
Ed added a consideration that applies specifically to firms that can build their own tooling. Ironbeam defaults to building internally, their CEO is also their CTO, but Ed acknowledged a clear advantage in doing so: when a regulator asks to open the box, it's easier if you built it yourself. Third-party vendors create a layer of opacity that firms have to manage. The trade-off is quality and speed. Both panellists acknowledged the opportunity this creates for AI-native compliance startups that can offer the transparency of a proprietary system with the quality of a purpose-built product.
The US opportunity is ten times larger than where we started
There are roughly 800 regulated trading platforms in the US. In the UK, the number is closer to 80.
Adclear started in London. The Financial Conduct Authority was the first regulator we built around and UK firms navigating financial promotions rules were our first customers. NinjaTrader was the first signal that the product translates, a US-regulated futures platform running their compliance workflow through a system built for FCA rules, because the underlying problem is identical.
The Chicago conversation confirmed that the adjacent market is large. Jeremy and Ed both pointed to verticals beyond trading platforms where the same compliance infrastructure problem exists and where AI tooling is still early: registered investment advisors, family offices, credit unions, and regional banks. Firms that face the same marketing compliance pressure but haven't had the same impetus to invest in solving it.
What's Next
New York is June 17 and 18. Front Row NYC CMO Dinner is co-hosted with Mint Studios , with marketing and compliance leaders from Adyen , Brex , and PensionBee US already confirmed.
If you're a marketing or compliance leader at a regulated financial services firm, this is the conversation we're building. Reach out to Doni Hoti or Joe Jordan.
Once you're booked in, we'll send you a free playbook on Financial promotions compliance for FinTechs.
