From Models to Markets:
AI's Growing Adoption in Investment Management
Is AI in Investment Management Finally Earning Its Place?
London, 23 March 2026 - Last week, at the Royal Automobile Club in London, we gathered professionals who face this question daily, leading to a practical and insightful conversation.
For the second year, Axyon AI hosted its annual roundtable, From Models to Markets: AI's Growing Adoption in Investment Management, bringing together senior practitioners from across the industry to move beyond the hype and discuss what is actually working, what is not, and where the journey is headed next.
Moderated by Anders Kirkeby, Head of Innovation at SimCorp, the event featured an exceptional panel of voices: Sam Zief from JP Morgan Private Bank , Sorin Ionescu from Deutsche Bank , Richard Taylor from BlueCrest Capital Management , Enrico Piccolini from Generali Asset Management, and our own CTO and co-founder, Jacopo Credi.
The Question Everyone Is Really Asking
At Axyon AI, we believe AI’s primary role is to augment human intelligence, not replace it. Whether through Predictive AI - our core expertise - or the new possibilities of Agentic AI, our focus is helping investment professionals gain better insights, make smarter decisions, and deliver improved client outcomes.
This belief shaped the morning’s tone. The goal was not to celebrate technology for its own sake but to reflect on its wise use - to spark ideas, challenge assumptions, and explore how AI can truly enhance alpha generation.
So, has the industry arrived? Not quite. But it is getting there.
From Experimentation to Habit
AI is no longer a novelty; it is starting to deliver tangible results. Yet its widespread adoption in investment decision-making is still in the early stages. The key question for the morning was clear: how do we turn AI from an experiment into a habit?
- A key distinction emerged between AI as a tool and AI as a transformative force—one that, like social media before it, may not just extend our capabilities but reshape the environment around it entirely.
- The direction of travel is clear: away from rules-based models and towards multi-model approaches combining short-term equity signals, macro regime classifiers, and asset-class-specific models.
- The human layer remains indispensable: translating single-stock AI signals into fund-level or ETF implementation still requires experienced judgement that no model fully replicates.
- One example is an agentic AI system that processes investment ideas submitted by email and returns fully quantitative, auditable KPIs within minutes, with humans involved throughout.
.jpeg?width=1200&length=1200&name=Axyon%20AI%20-001%20(1).jpeg)
Two Very Different Conversations
Not all AI adoption is equal. The panel drew a sharp and practical line between two categories of application that demand very different approaches.
-
Workflow automation is already happening at scale — internal LLMs drafting client communications, tools processing thousands of term sheets automatically, AI-generated meeting summaries landing in inboxes within minutes.
-
Investment decision-making is different: fiduciary duties, governance, and the need for long out-of-sample track records raise the bar and extend the timeline.
-
The industry is in the "saucer" of the ROI curve: investments have been made, but the full impact on productivity and headcount is yet to be realised.
-
Large institutions will likely be the last to adopt AI at scale in their investment processes — not from lack of ambition, but from appropriate caution with other people's money.
What Actually Works — and What Does Not
From practitioners with decades of experience, the panel offered candid lessons on where AI delivers and where it disappoints.
- Two use cases shared: an early warning system for credit rating downgrades, blending credit spread and balance sheet data; and a contagion effect model, tracking how stress in one issuer propagates across related names.
- Both succeeded by starting from a precise, back-testable business problem. The failures shared a common trait: questions that were too broad produced noise, not signal.
- The build-versus-buy debate remains live: building in-house offers control but carries significant opportunity cost; buying from specialist vendors offers speed and established infrastructure.
Explainability: The Black Box Is Opening
The "black box" perception of AI remains one of the greatest barriers to client adoption. Progress is being made, but the tension between model complexity and transparency is real.
- End-to-end explainability frameworks can now connect strategy performance, model behaviour, and raw data inputs into a single auditable chain, turning black boxes into defensible decision support systems.
- Instead of saying "the model weighted macro data more heavily," we can now explain that "the strategy rotated into defensive sectors due to rising FX volatility and slowing trade, as measured by declining shipping profits."
- A conceptual challenge remains: if a pattern can be explained simply, is it truly novel alpha? The tension between complexity and explainability is unresolved. Rigorous error analysis and problem decomposition help, but no easy answers exist.
Data, Bias, and the Jobs Question
Beneath every discussion of models and governance, three themes surfaced repeatedly — and are worth stating plainly.
- Data is everything. Wide, deep, representative, point-in-time data is the primary ingredient of any successful AI application. More data is not always better; curated, high-quality subsets increasingly outperform sheer volume.
- Bias is everywhere. For instance, any model trained on recent, US-centric data may carry subtle geographic preferences it was never designed to have. AI does not introduce bias into an objective process — it can codify and scale the biases that were already there. Rigorous, ongoing detection is not optional.
- Jobs are changing, not disappearing. AI is already displacing certain tasks, but the complexity of what remains — governance, explainability, data infrastructure, human-AI integration — continues to demand skilled professionals. As one participant closed the morning, “it is not a question of whether jobs disappear. It is a question of what the job becomes.”
Final Thoughts
The firms that will lead the AI adoption are not necessarily the largest or the best-funded, they are the ones willing to ask the right questions, start with the right data, and move with the right partners.
If that conversation resonates with you, we would love to continue it.
