High-quality data will set the pace for AI in investment

By Kenneth Lamont, Principal, Manager Research at Morningstar

Before models and predictions, must come well-prepared data to fuel artificial intelligence.

Even the most advanced AI systems can be prone to hallucination—delivering flawed or misleading results. Structured, explainable, high-quality data isn't just the necessary fuel to AI in investment, it's the competitive edge needed to put investors ahead of the curve.

When Morningstar developed our global classification system, we needed to address that core challenge of making investment data truly AI-ready. Here's how we created a thematic fund dataset from the ground up using a blend of human expertise and machine learning.

Why not all data is analysis-ready out of the gate

The reality is many financial datasets are messy, unstructured, or inconsistent. They lack the explainability and traceability required for investment-grade analysis, and they’re often modeled retroactively to fit analysis tools that assume a high degree of standardization.

In thematic investing, unlike other sectors, thematic funds don’t always fit into standard categories, making them harder to analyse, benchmark, or compare. At Morningstar, we’ve worked to bring more clarity and structure to that space through our global classification system.

How we built a global thematic fund classification system from scratch

To tackle categorising over 50,000 thematic funds, we first identified funds based on intentionality, grouping together those that say they invest in similar themes. We reviewed investment objectives, legal names, and prospectuses to build a universe of funds explicitly focused on themes like AI, blockchain, or water.

From there we built a flexible three-tiered classification system. Funds could be grouped under broader and more specific themes at the same time:

Broad Theme

(e.g., Social)

Theme

(e.g., Demographics)

Sub Theme

(e.g., Ageing Populations)

Human analysts led the research, curating themes and validating categorisations. After establishing a rigorous framework, natural language processing (NLP) was layered in to help scale the process.

Machine recommendations and human validation

Our NLP system crawls all new fund launches by using a curated library of terms drawn from thousands of fund documents to flag potential thematic funds and suggest tags. However, the NLP-suggested categorisations aren’t just spat out and taken at face value.

Analysts review the recommendations and validate them based on their expertise. The analyst feedback also allows the model to learn and improve over time. Plus, this easily explainable process makes it so that we can show clients exactly how and why each tag was applied to a fund.

What makes data investment-grade

Inconsistent classifications can muddy comparisons and mislead investors. That’s why I place so much emphasis on structure, reliability, and transparency. Morningstar uses these thematic tags across products and teams, so as a power user, I need reliability and insist on that trustworthiness being the standard.

What makes data truly investment-grade can be broken into five core checkpoints:

Human context

Our classification system was created and is maintained by analysts. This human touch allows the framework to stay relevant to the needs of a range of clients globally.

Scalable automation

Leveraging Morningstar’s vast datasets, relevant inputs are processed at scale and used to highlight thematic intentionality and recommend appropriate tags.

Human oversight and validation

Every classification is reviewed by analysts, ensuring alignment with investor expectations.

Global consistency and local context

The classification system works globally, but accounts for regional naming nuances and disclosure differences. Local analysts validate where needed.

Traceability and explainability

Every tag and classification is fully auditable, with a record of sources and logic.

Speaking of explainability, here’s why it matters. Morningstar has taken steps to ensure our thematic tagging approach is fully transparent. Clients can see which terms were flagged, which inputs were used, and how the final decisions were made. In a world where regulators are increasingly scrutinising how data feeds AI models, this kind of explainability is a necessity.

Data prep should be a discipline, not an afterthought

To get data to investment-grade quality, asset managers can’t just aggregate. They must structure it, validate it, and document its journey from raw input to investment insight. Quality data planning should include:

Structure, validation, versioning, metadata

Benchmarks, performance tracking

Human oversight, quality control

New fund structures, emerging market trends

Error handling, backup data strategies

Regular data audits, data health checks

Real-time data corrections

The ability of data teams to swiftly respond to market shifts and emerging data needs is another skill essential to bringing data up to par.

Continuous optimisation is the way forward

At Morningstar, our goal is to help investors navigate complexity with greater clarity. That means building structured data that’s not just big, but brilliant – explainable, and built to go the distance.

Our next steps include enhancing our global classification system to ensure it remains the gold standard in a rapidly changing world, building our local language capabilities to ensure the deepest local market coverage possible, and expanding our coverage to new fund structures.

But, all of this can only be accomplished through the guardrails of maintaining explainability and other core standards. Thematic investing isn’t going anywhere, and neither is the demand for AI-powered insights.

Access more of Morningstar’s insights on investment data and artificial intelligence.