In this episode, Andrew Morgan, Chief Product Officer at MahiMarkets, joins host Atique Abdullah to discuss the FCA’s Mills Review, human oversight in automated trading, and what’s next for AI in financial services.
Atique: Andrew, I just wanted to understand what stood out for you when it comes to the Mills Review. We’re having this conversation because of the Mills Review, I believe — what stood out for you?
Andrew Morgan: I read the review and I was a bit mixed. I was fearful, a little, that some regulation would come in and kill a lot of the growth, a lot of the opportunity with AI. I was quite relieved in some respects that it leans more on existing regulation and frameworks rather than trying to control something that’s changing and innovating so fast. So on that side of things, I agree with it. On the other side, I thought targeting 2030 is quite a while — that’s a long time in AI terms. It’s exhausting keeping up with it as it is. What will the ecosystem look like next year, let alone in 2030? That was my first response. I also thought it might touch more on some of the regulation around things I’m seeing at the moment, and that felt a little absent.
Atique: Got it. And why do you think AI regulation has become such an important topic in financial services — why now, in 2026?
Andrew Morgan: I think everybody is adopting AI. We all realise the massive growth potential, the enabler that it is. Initially we all thought maybe it’s a time-saver, something to help us prepare for things. But as I’ve gotten more into it, it’s been the reverse — instead of being a time-saver, it’s reset my ambition. I’ve raised my ambition; it’s so powerful that you think, okay, we can do more with this. And with that power and responsibility, there’s no wonder there’s a huge amount of focus on regulation.
I think we’re quite nicely positioned because we’ve already been through an automation cycle. Trading already moved away from voice markets — it used to be RFQ, request for quotes, always a human in the loop to pin down the price on a ticket. Then it moved into automatic trading and ticket counts went huge. So we’ve already seen what happens when the machines go wrong. Our own founder ran trading and risk in FX and options at a major investment bank before founding MahiMarkets, so we’ve got a lot of experience going back to the electronification of the marketplace. AI is another layer, another opportunity on top of that. Everybody’s asking, “AI, can you make money for me?” So there’s absolutely a need, and that’s why regulators are interested.
Atique: People are increasingly talking to and taking financial advice from chatbots that don’t have regulation in place. What do you feel about that?
Andrew Morgan: Yeah, that’s a good one — this is the bit I thought might actually creep into the report and didn’t quite. It’s understandable — a lot of companies are just launching AI bolt-ons on their offerings. Without pinning it on one particular mover, you’re seeing brokers like Robinhood and eToro assign a balance to an AI that then trades autonomously on behalf of the retail user. I have to question whether users really understand the risk. Who’s responsible? It only takes one catastrophically wrong trade to unwind ten great ones. That’s an area where I’d hope for more guidance in future. That said, I think people are going to do this whether it’s allowed or not — the AI can sit in the browser and start trading on your behalf regardless. Some offerings are more thoughtful — I saw one from Interactive Brokers with a human in the loop: it tells you what it wants to trade and you approve it. I think it’d be good to be clear on who’s responsible for the loss.
Atique: Is that a consumer education problem, or a regulation problem?
Andrew Morgan: I’m hoping the brokers and trading platforms are going to be responsible about it. If you give someone AI capability with a polished narrative — charts showing how much money this would have made historically — and they go live and it doesn’t work out… adoption is already huge and increasing, so over time there’ll be more disgruntled examples of people saying “well, I didn’t know, the AI told me this was a good bet” — and it lost. Should they be asking OpenAI for a refund, or the broker who enabled the workflow?
Atique: What do you think are the biggest risks if innovation outpaces regulation?
Andrew Morgan: There is that risk, and it comes back to that 2030 number — in some ways you have to understand what you’re regulating before you can regulate it appropriately. The pace of innovation is exhausting at times; there’s something new every day, it’s an arms race to stay ahead. So there’s a risk the review is too far out — maybe more frequent, ongoing reviews would help, where firms can share evidential risk and reporting on an ongoing basis. All it takes is a new architecture or phrasing of current models — already powerful enough — for something to happen that needs regulation quickly.
Atique: Where do you think organizations should balance AI automation with human oversight?
Andrew Morgan: It varies a lot by organization. At MahiMarkets, we’re a smaller, global firm, so we can adopt this kind of thing quickly. I see larger organizations that just aren’t able to move — internal mandates keep them on outdated versions of AI models that aren’t very effective. February this year was a real inflection point where adoption went through the roof industry-wide, and some firms still aren’t there. One thing I’d flag on risk: just because you can, doesn’t mean you have to. If you do a public launch of some LLM feature and clients get used to it, it’s hard to roll back — you end up dealing with AI-generated narratives from clients that they send to you asking “is this right or not?” — that’s an increased support and operational burden, and if you take it away, people notice.
Internally it’s a different game — we’re doing a lot of agentic-operating-model work internally, very controlled about what’s released publicly. We run beta groups exploring different models, longer-running orchestration, giving it larger tasks. It’s driving our ambition. So there are risks, but there are opportunities too.
Atique: How would you define good AI adoption, from your point of view?
Andrew Morgan: One thing we’ve been careful about at MahiMarkets is not just trusting the narrative. We measure outcomes against trusted, validated products and techniques, and we let the AI explain itself using those trusted products so we can index into something we can actually validate. We’re all about minimizing time-to-verification so we can act on it, with a human in the loop wherever the consequence really matters. Given our history in market-making and automation — running 24/7 — the cost of failure is huge, so we’ve built in firewalls and protections. Time-to-action is critical in trading, and so, consequently, is time-to-verification.
Atique: Any trends you’d highlight for financial services over the next few years?
Andrew Morgan: There’s a huge amount of change coming. I think we’ll see increased competition — people entering addressable markets they weren’t in before. You’ll see a lot more innovation and R&D because routine work gets automated, which buys back time for development and new markets. Things you’d never have done otherwise — you’d have said “the budget’s not there, we’re not getting this done this quarter” — now it’s “we’ll get it done next week.” We’re finding people across the organization with no coding background at all are now writing code. It’s democratizing, and that’s a good thing. How do you see it?
Atique: For me, a good AI implementation starts with getting everyone on the same page — that’s a genuinely hard job. Some people think AI just retrieves answers from the web; others realize a single person can get an entire body of work done with something like Claude Code. So it’s about educating people on what AI can actually do — not just in financial services, but in day-to-day life. I think we should double down on what’s already working and put more effort into educating people on what’s possible.
Andrew Morgan: Education is definitely a big part of it, because it’s changing so fast. Different people lean into it differently, so you get very different outcomes depending on who you are. AI isn’t a saviour — a lot of people say “AI wrote that” as if that settles it, but AI in the wrong hands can produce bad results too. It really is about responsible use and education.
Atique: Any advice for the current generation of AI-driven leaders and business leaders preparing for AI-driven futures?
Andrew Morgan: Going back to an earlier point — you don’t have to be first to market. A lot of people are rushing to bolt AI onto everything they can. It’s great that people are exploring, putting an MCP server on something — that’s trivial to do, but it doesn’t mean you should. It’s all about trust, and trust is easy to lose quickly. If someone in public is dealing with an AI that doesn’t really understand what they’re asking, they’ll get frustrated fast — sometimes having a human in the loop is just the sensible thing to do. AI can back that up — it’s tireless, it can be structured, it can support and scale, huge power. So my advice: this is coming, and if you’re not adopting it, you need to get moving — but be very careful about your positioning. I always ask AI for proof, not just numbers — I don’t want a polished narrative, I want something I can prove and something I can actually trade on. That’s still a blend of human and AI.
Atique: Last question — if an 18-year-old just out of college were listening to this, what two sentences would you want to leave them with to help build their career?
Andrew Morgan: I’ll ramble a bit and then give you the two sentences. I often think back to what I was like at 18. I went to Cambridge, and my older brother also went to Cambridge and went on to work with Sir Demis Hassabis, now head of DeepMind — so I’ve had a close view of how quickly this has turned into something massive. It’s an amazing enabler and a learning tool, but you have to get good at recognising when it’s wrong. Historically, you’d have to sit in rooms doing manual tasks for a long time just to get a foothold in a business — a lot of grafting to get anyone to spend time educating you. Now that’s being democratised — so much learning, so much capability. I think we’ll get to a one-person unicorn. I’d lean entrepreneurial — some of the traditional routes are closing down, but the amount of capital it takes to start and run a company now is a lot less. So: go big, start building, get experience, show people outcomes — that’s the massive opportunity.
Get in touch with our FintechEdition Editorial Team.
