AI in Asset Management
A practitioner's view on how AI and code are reshaping the investment process — what works, what doesn't, and where the real edge lies.
How AI can be used as a portfolio manager
The conversation about AI in asset management tends to oscillate between two extremes: breathless disruption narratives where algorithms replace portfolio managers, and dismissive scepticism where nothing changes. The reality, as usual, is more interesting and more nuanced.
Over time I have built over dozens of tools that augment investment processes — from Monte Carlo simulation engines and factor correlation platforms to automated research pipelines and probabilistic financial modellers. I believe so strongly that there will be a seismic impact on the industry that I have invested in an existing software company and helped them set up the Finance arm of that business.
However, I’m still of the view that AI in asset management has to be handled carefully to ensure the benefits which AI brings doesn’t damage the human aspect which delivers value for clients.
What AI is good at in investment management
Data processing at scale
Screening 5,000 companies across many countries and giving a first pass used to take weeks. Now it takes minutes. AI and automation excel at ingesting, cleaning, and structuring financial data — pulling from APIs, parsing filings, normalising across currencies and accounting standards. The value isn’t in the analysis itself but in freeing human attention for the parts that actually matter.
Pattern recognition across large datasets
Factor analysis, correlation matrices, and base rate computation are natural territory for code. My factor correlation tool analyses how stocks and portfolios relate to macro factors across multiple time periods. My decision journal tracks 2,400+ stock-level decisions with measured outcomes, computing success rates, forecast accuracy, and bias detection — the kind of systematic self-assessment that would be impossible manually.
Simulation and scenario modelling
Monte Carlo simulation, fan charts, and probabilistic scenario analysis are transformative for portfolio construction. Instead of a single forecast, you can explore the full distribution of possible outcomes and ask: how does this portfolio behave across 1,000 different futures? This shifts the conversation from prediction to preparedness.
Research synthesis and acceleration
AI-powered tools can summarise earnings calls, extract key data points from annual reports, and surface relevant precedents from historical filings. This doesn’t replace reading — but it dramatically compresses the time between question and relevant information. This can be extended to help review portfolio holdings in an efficient way.
Visualisation and communication
D3.js and modern web frameworks make it possible to build interactive, explorable data visualisations that communicate complex ideas more effectively than static charts ever could. This site is itself an example — the interactive widgets in the philosophy section let readers explore the evidence rather than just reading about it.
AI (specifically LLMs) are incredibly powerful at writing deterministic code to build tools to enhance investment processes and for me that is where a large part of the power lies currently. Using LLMs to parse information is also a possible area of benefit, but needs to be handled carefully with an understanding that LLMs provide probabilistic results and can be prone to pitfalls such as hallucinations. Any agents need to be designed carefully and with appropriate structure and evaluation methodologies.
AI capability × investment task
A qualitative fit matrix. Each cell rates how well one AI capability (scale, repetition, pattern matching, content generation) maps onto a given investment task. Darker cells indicate stronger fit. Hover for context.
Top rows: AI-strong tasks. Bottom rows: human-core tasks.
What AI cannot replace
For all its power, AI fails at the things that make good investing genuinely difficult. These are the areas where human judgement remains irreplaceable — and where, paradoxically, technology makes it even more important.
Judgement under uncertainty
Models work with quantifiable inputs. Real investment decisions involve ontological uncertainty — the things you don’t know you don’t know. No model captures a regime change it has never seen. Markets are also complex, adaptive systems. Identifying a past relationship and acting upon it changes the system itself. The human judgement which sits behind strong track records becomes even more important.
Conviction and contrarianism
Buying when everyone is selling requires emotional fortitude, not computation. AI can tell you a stock is cheap. It cannot tell you whether you have the temperament to hold it through a 40% drawdown. To this extent, it may be a better replacement for quantitative methods of investing rather than methods which require human judgement.
Management assessment
Capital allocation skill, incentive alignment, and corporate governance are qualitative judgements that resist quantification. A management team’s integrity doesn’t show up in a spreadsheet until it’s too late. Before getting too carried away, meeting management and analysing their intentions and motivations will remain a very difficult thing to do.
Behavioural discipline
The value premium exists because exploiting it is psychologically painful. No algorithm can give you the discipline to ignore short-term noise, resist narrative seduction, or accept looking wrong for extended periods. There is a chance that AI use focuses on ‘solving’ near term problems, offering a greater opportunity for those willing to exploit the time arbitrage which exists in markets, using the power of patience.
Relationship and trust
Client communication, team leadership, and the ability to explain complex ideas simply are fundamentally human skills. Trust is built through dialogue, not data.
Asking the right questions
AI is extraordinarily good at answering well-specified questions. It is much less good at knowing which questions to ask. The sweet spot will be the use of AI by someone with rich domain knowledge. There will be real pitfalls in inexperienced workers using AI to provide credible yet wrong answers. Enhancing critical thinking rather than replacing critical thinking should be the aim.
My belief is that there is a lot of room for humans in the loop in a world where AI is used more. This is particularly true in areas where AI may struggle such as decision making in uncertain environments, critical thinking and understanding when AI use is appropriate (such as linear systems vs complex systems).
A walk through AI in the investment process
Having set out where AI may be useful and where it may be less so (or even a hindrance), let’s walk through an investment process and think about the areas of opportunity for adopting AI fully.
The investment process, step by step
17 stages of the active investment process. Click any step to expand. Colour indicates my subjective assessment of where AI provides the most leverage today.
Research & Idea Generation
Finding, filtering, and understanding companies
Decision
Weighing the case and challenging biases
Portfolio Management
Risk, construction, and ongoing monitoring
Meta & Feedback
Learning, simulation, and explanation
01Finding ideas
Screening features at the start of most investment processes. While it doesn’t have to change hugely, there is greater scope to screen on more factors (including non-quantitative factors) and visualise the opportunity set in different ways. A successful implementation here would be ensuring that the appropriate opportunity set is identified and the portfolio manager doesn’t feel like opportunities are being missed.
The code-as-edge thesis
One big area of potential AI benefit for me is the ability to build your own tools is itself a competitive advantage.
When you build a valuation heatmap, you understand the data at a deeper level than someone who receives a vendor’s pre-packaged output. When you code a Monte Carlo simulation, you internalise the mechanics of uncertainty in a way that reading about it cannot replicate. When you build a decision journal that tracks your own forecast accuracy, you develop a feedback loop that most investors never create.
The tools I’ve built are not about replacing investment judgement. They are about augmenting it — creating an environment where better decisions become easier to make and harder to avoid.
Proprietary tools built for the investment process:
Each of these tools encodes a specific way of thinking about investment problems. Collectively, they represent a process infrastructure that would be difficult to replicate — not because the code is complex, but because the domain knowledge embedded in the design is hard-won.
Where this is going
- •Investors who can code may have a structural advantage over those who cannot — not because code replaces judgement, but because it amplifies good process.
- •The value of domain expertise will increase, not decrease. AI makes generic analysis a commodity; it makes deep, specialised knowledge more valuable.
- •The competitive landscape will bifurcate: firms that integrate AI into their investment process and firms that treat it as a bolt-on reporting tool.
- •The best use of AI in investing will remain invisible to the end client. It will show up not in flashy demos but in better decisions, more disciplined processes, and more honest self-assessment.
The investors who will thrive are those who use technology to strengthen the fundamentals that have always mattered: rigorous analysis, disciplined process, honest self-assessment, and the courage to act on conviction under uncertainty. AI does not change what good investing looks like. It raises the bar for doing it consistently.
Key Takeaway