Nobody in finance wants a chatbot

"Financial services are too slow, too risk-averse and too set in their ways for AI to make a dent." I hear some version of this every month. Its European cousin goes: "Europe is behind on AI. Too regulated, too cautious, too fragmented."

Neither is wrong about the facts. Both are wrong about what the facts mean.

The complexity is the opportunity

I've spent more than a decade building software for fund managers, assurance firms, real estate businesses and other financial institutions. The industry carries enormous operational complexity, some of the heaviest compliance burdens of any sector, and talented people who spend a disproportionate share of their day on work that doesn't need their expertise. That isn't a criticism. It's how these organisations have had to operate, because there was no better way.

Look at the conditions: high document volumes, repetitive rule-based processes, reporting cycles that consume entire teams, compliance monitoring that happens periodically when it should arguably be continuous. These aren't marginal inefficiencies. They're structural, and they have been for years.

That's exactly why I think financial services have more to gain from AI than almost any other industry. The bigger the pile of drudgery, the bigger the upside of clearing it.

What practitioners actually ask for

Over the past year I've been in a lot of rooms with asset management practitioners, talking about where AI fits. The conversations keep catching me off guard. Not because the ideas are wild, but because they're boring.

Nobody asks for a chatbot. Nobody wants a demo of yet another AI copilot. What comes up, again and again, falls into four buckets.

Repetition. The number one request is some version of "how do we stop doing the same thing forty times a day?" Trade reconciliation, NAV exception handling, investor reporting prep. Leaders want their ops team to stop copy-pasting between three systems all day.

Documents. Every fund is buried in PPMs, side letters and compliance filings. They've seen OCR tools and they don't want another one. They want something that reads for context and turns a document into clear next steps.

Decision support. I want to be precise here: not one person I've spoken to wants AI making investment calls. What everybody wants is speed. "Summarise that earnings call. Flag what's off in portfolio exposure. Tell me what I need to pay attention to, then disappear."

Compliance. This one is getting louder. Regulators are tightening, funds are putting more people on staying ahead of requirements, and everyone is looking for proactive monitoring. Nobody cares whether it runs on agents, a language model or a spreadsheet macro.

The through-line is that the sharpest people in this space aren't buying AI because it's fashionable. They're buying it because something specific costs them money or time, and they want that fixed. They couldn't care less which model runs under the hood.

New tools join the old ones

The same pattern shows up in the most quantitative corner of the industry. I spent some time comparing how portfolio behaviour gets predicted: the traditional methods used for decades, the machine learning techniques that came after them, and the generative AI initiatives everyone is pitching now.

When people say "AI" today, I hear a demand for magic. The magic that automates everything, the magic that earns all the world's money. But AI isn't new to portfolio analysis. Quant models built on machine learning, the unhyped part of AI, already do a decent job of predicting price swings. The real question is whether a language model does better.

The honest answer is that these aren't interchangeable tools. They're different approaches that work well together and poorly alone.

• Traditional methods aren't precise, but they keep regulatory supremacy. Regulators and investment committees understand them, and they will stay.
• Machine learning will keep generating alpha from the numbers, but more of its inputs will come from language models.
• Generative AI will supply more and more of that data, reading filings, transcripts and news at scale, rather than acting as a stand-alone oracle that tells you what happens next.

That's the shape of most good AI in finance: a new layer that feeds the existing machinery, not a replacement for it.

Regulation is an advantage, especially in Europe

Back to the European objection. European financial services carry some of the highest document and compliance burdens in the world. Fund managers and assurance firms navigate several regulatory frameworks at once, across jurisdictions. That complexity is precisely where well-implemented AI creates disproportionate value. The problem is bigger here, so the upside is bigger too.

The EU AI Act creates clarity as much as constraint. For a compliance-conscious organisation, a clear framework to work within is an advantage: it forces proper implementation, which is where durable value lives anyway.

There's also a maturity factor that gets underrated. European institutions adopt technology deliberately. The start is sometimes slower, but the implementations are built properly, embedded deeply and actually used. That's a fundamentally different outcome from a rushed deployment that gets quietly abandoned six months later. And the European talent pool for this work is world-class: deep technical capability, real domain expertise in regulated industries, and a culture that takes privacy and governance seriously.

Domain knowledge decides it

What my team and I have learnt from real implementations, rather than from conference slides, is that the technology rarely makes or breaks a project. What matters is whether you genuinely understand the workflow, the regulatory context and how decisions get made inside these organisations. That knowledge takes years to build, and no model ships with it.

AI won't just make financial services more efficient. For the firms that move with intention, it will change what's possible. The opportunity isn't despite the complexity. It's because of it.

For the numbers behind this, see AI adoption in finance 2025–2026: state and predictions on the BN Digital blog.