AI: Data Demand Revives Alternatives

The data collection frenzy by AI players is reviving an old vein on Wall Street: so-called alternative data. However, this potential comeback comes with the same pitfalls as before, as some players are already adjusting their offerings to meet the needs of models. A former Point72 employee, Kirk McKeown, has launched a startup to package these datasets for AI use.

Known Constraints: Market Caps and Data Quality

The previous cycle of alternative data was marked by growth difficulties, with providers struggling to expand their catalogs without compromising quality. The market also had a limitation: when too many players use the same datasets, the informational advantage disappears. Hedge funds, although they have significant resources, are few in number, which restricts the potential customer base. The current needs of AI companies could be just as demanding and difficult to satisfy. Moreover, some asset managers might believe that their own internal data remains, in some cases, the most relevant.

What AI Wants: Internal Archives, Recorded Tasks, and Heterogeneous Volumes

AI companies are primarily interested in documented work traces: internal memos, transaction records, contracts, and onboarding guides. There is a race to collect as much data as possible in a market described as rapidly evolving, with offerings changing frequently. Some holders of structured archives, such as law firms, are looking to monetize their document collections. In robotics, companies are creating their own datasets by funding the recording of daily tasks such as cooking or cleaning. Access to private content, including direct messages on Slack, is mentioned as potentially playing a key role, with the perspective that automation requires a deep understanding of actual tasks.

A Different Outlet from Finance, with Identified Pioneers

Alternative data, which emerged on Wall Street in the mid-2010s, encompasses unstructured sources such as satellite images or card payments, believed to provide an informational advantage, for example, by assessing foot traffic at Home Depot before financial results are published. The uses of AI are not the same as those of hedge funds, which primarily seek real-time flows related to markets, while AI favors larger and more detailed datasets. Some providers already estimated that the strongest demand would come from large enterprises; the rise of AI could fulfill this expectation. Players are positioning themselves: last year, Kirk McKeown, a former data expert at Point72, founded a startup to package datasets and make them accessible to AI models.