As AI continues to dominate boardroom agendas, it's easy to focus on the latest model, capability or breakthrough.
This week I sat on a panel at Indicium AI’s Applied AI Symposium in London, talking about where AI is actually delivering in banking and capital markets — beyond experimentation and into real deployment.
A few things crystallised for me coming out of that discussion. Five themes in particular stood out.
1. AI success starts before the model: There is no shortage of excitement around AI's potential. However, the businesses seeing the most value are consistently focused not just on the AI itself, but on the foundations that enable it.
In financial services, this means thinking about the data, governance and codified domain expertise that ensure AI systems are explainable, trustworthy, and consistent.
That's reflected in some recent research from Celent, which found that over half of lenders surveyed ranked data integrity, lineage, privacy and control among their top three priorities. Trust and traceability are not nice-to-haves, they're prerequisites for scaling AI responsibly. Get that right and speed follows. Skip it and you’re just scaling your exposure.
And as regulation continues to evolve, organisations investing in these foundations now, rather than adding them later, will be able to move faster and adapt more easily to future change.
2. Data matters. Context is the differentiator: For years, the conversation has revolved around data quality. That remains essential, but data alone is not enough.
To deliver trusted outcomes, agents also need machine-readable context: standardised metric definitions, and an understanding of customers, business priorities, risk appetite, decision-making frameworks and industry expertise. This is particularly important in highly regulated sectors because it isn't simply providing AI with access to data; it's ensuring that data can be interpreted and applied appropriately.
3. Measure outcomes, not activity: AI is very good at generating the illusion of progress — prompts run, tokens burned and dashboards that look busy. The real test is whether decisions get better.
For example, we are currently working with a large UK high-street bank using an Insights Agent in our Analytical Sandbox, and it’s cut our client’s analytical execution time from months to hours. But what matters more is what that unlocks: banks intervening proactively in cases of financial vulnerability, instead of waiting for the crisis moment. That’s the difference between impressive activity and real-world impact.
4.Scaling AI takes partnerships and people: Successfully scaling AI is not just about the technology. It requires the right combination of partnerships, people, and expertise. AI solutions increasingly rely on businesses, platforms and data providers working together, with ecosystem collaboration and shared standards critical to success.
At the same time, businesses need the right foundations internally. Success depends on clear ownership, the right skills, and accountability to help AI make appropriate decisions.
But the biggest barriers to AI success are often less technical and more operational. Models will continue to improve. The harder challenge, however, is creating the trust, context, and organisational capability needed to apply the models effectively. This means as AI becomes more accessible, this is where the competitive advantage will increasingly be won or lost.
5. What’s next: Multiplayer AI and agent identity: Two things I’m watching over the next 18 months.
First, AI will continue to move from “single-player” productivity tools to genuinely multi-player systems — agents talking to agents, across organisations, not just within them.
Second, adoption is outpacing trust and standards, and nowhere is that more visible than in agentic commerce. When you imagine AI agents making purchases on a person’s behalf, the ability to tie an agent action back to a real, accountable human — verifying who’s actually behind the agent — stops being a nice-to-have and becomes a key component enabling agentic finance scale safely.
The takeaway
AI is moving past the pilot stage in financial services into real deployment. The businesses winning this next phase aren’t just the ones with the latest models — they’re the ones that built data, governance and infrastructure foundations early enough to move fast when it matters.