Alternative data has spent much of the past decade moving from the margins of investment research towards the mainstream. Now, artificial intelligence is accelerating that process, making datasets easier to discover, evaluate, process and combine while allowing investment teams to tackle sources that would previously have required considerable engineering resources.
But that creates a new challenge. As the technology needed to work with alternative data becomes more widely available, where does the investment edge come from?
That question ran through the “Alternative data in production – from models to markets” leaders panel at last week’s A-Team/Eagle Alpha Alternative Data Conference in New York. The discussion suggested that AI is simultaneously lowering the barriers to using alternative data and raising the threshold for what constitutes genuinely differentiated information.
From finding data to extracting value
The alternative data market has changed considerably from the days when acquiring an unconventional dataset could itself provide differentiation. Datasets are increasingly structured, productised and available through APIs and cloud platforms, while tools for evaluating their relevance and predictive value have improved.
Simply knowing that a dataset exists therefore carries less value. Its usefulness increasingly depends on whether it can be integrated efficiently, applied across investment strategies and contribute enough value to justify its cost.
AI potentially broadens that opportunity. Data that once required dedicated quantitative researchers or engineering teams to ingest and manipulate can increasingly be explored by fundamental investment teams using AI-based tools. That could make highly specialised or niche datasets more accessible rather than less valuable.
At the same time, readily available public information is becoming harder to monetise as alternative data. If investment firms can use AI to discover, collect and process public sources themselves, paying a third party simply to aggregate that information becomes more difficult to justify. Proprietary and difficult-to-replicate data consequently assumes greater importance.
Compressing the route into production
One of AI’s most immediate effects is the compression of the journey between discovering a potential source and putting it to work.
The panel discussed applications ranging from dataset discovery and contract management to coding, process auditing, document extraction, quality control and combining multiple sources. AI agents can also help data teams search for information relevant to particular investment strategies, reducing the need for data managers to understand the detailed requirements of every portfolio manager or research team.
Development and evaluation times are also falling. One example discussed during the session involved dataset evaluation periods dropping from around four months to one. Another concerned information contained in handwritten oil-well completion documents, where different US states used different forms. A project once considered prohibitively difficult was revisited using AI and turned into something that could be delivered in around a month.
That ability to tackle previously impractical sources potentially expands the universe of alternative data itself. Information doesn’t necessarily have to arrive in a convenient, machine-readable format to become investable.
A renewed appetite for raw data
AI is also changing what sophisticated data consumers want from providers.
Rather than relying exclusively on vendor-generated analytics or finished signals, some investment firms increasingly want the underlying material – full transcripts, filings, tokenised text, metadata and other granular content – which they can feed into their own models and combine with proprietary information.
As firms build their own AI capabilities, more of the analysis can take place inside the investment organisation rather than being embedded in a standardised signal supplied to multiple customers. For data providers, that places greater emphasis on supplying high-quality underlying content that can be incorporated directly into clients’ own workflows.
Licensing models may have to adapt accordingly. The panel showed little enthusiasm for paying additional fees simply because existing data is being consumed through an AI agent or new delivery mechanism such as Model Context Protocol (MCP). There was greater recognition that the commercial equation changes when data is being used to train proprietary models or when providers make substantially more underlying content available for AI-based analysis.
That distinction is likely to become increasingly relevant as agent-based access develops alongside APIs, cloud marketplaces and traditional feeds as another means of distributing financial data.
Quality becomes more important, not less
Greater automation doesn’t remove some of alternative data’s longstanding problems. Processing more information at greater speed makes the quality of the underlying data more consequential.
The panel repeatedly returned to point-in-time accuracy, timestamps, consistent definitions, source integrity and human validation. AI can identify anomalies, categorise information and perform quality checks across enormous volumes of data, but blindly trusting its output introduces different risks.
One example involved historical market data containing gaps caused by market closures. An AI system attempting to create a continuous series could potentially fill or smooth those gaps, inadvertently removing information that an experienced market practitioner would recognise as significant.
Human domain expertise therefore remains part of the production process, even as AI takes on more of the labour involved in getting information there.
Where the scarcity moves
These changes are beginning to influence spending priorities. As AI tools themselves become increasingly commoditised, proprietary information that competitors can’t easily reproduce becomes more strategically important. At the same time, better usage analytics are making it easier for data managers to identify underused subscriptions, potentially encouraging consolidation around datasets and providers that can demonstrate sustained value.
AI is reducing the cost and complexity of discovering, processing and experimenting with information, potentially opening investment research to sources that would previously have been too difficult to exploit. Easier access, however, doesn’t necessarily translate into easier alpha.
As the mechanics of working with data become increasingly accessible, differentiation is likely to depend more heavily on proprietary sources, trustworthy point-in-time data, unusual combinations of information and the investment expertise used to interpret them.
The alternative data industry’s next phase may therefore see the competitive advantage shift further away from the tools themselves and towards the data and investment processes built around them.
Subscribe to our newsletter
