The Specific Event

Two stories broke this week that, read separately, look like routine business news. Read together, they expose something worth taking seriously. First, Keelan Miskell wound down his crypto startup and discovered that its digital archives - user data, internal communications, behavioral logs - had real market value to AI labs. He netted roughly $2 million from data he initially assumed was worthless. Second, and relatedly, AI labs are now approaching operating companies directly, offering significant sums for internal communications: employee emails, Slack threads, internal memos. The market for organizational communication data is not emerging. It has already emerged, and it is pulling in assets that most organizations have never thought to classify as assets at all.

This is not primarily a privacy story, though the privacy concerns are real and serious. It is an organizational theory story. And the theoretical problem it exposes has not been adequately named.

What Organizations Do Not Know They Have

The Miskell case is instructive precisely because of his initial reaction. He did not think the data was worth much. Anonymous users, a young company, no obvious institutional value. He was wrong by approximately $2 million, which suggests that the valuation gap between what organizations believe their communication data is worth and what AI labs will pay for it is substantial and systematic.

This gap is not random. It follows a predictable pattern rooted in what Kellogg, Valentine, and Christin (2020) describe as the opacity of algorithmic systems to the workers and organizations operating within them. Organizations generate enormous volumes of communication data as a byproduct of coordination. They do not generate it as a strategic asset. The distinction matters because it means most organizations have no internal accounting framework for this class of resource. The asset exists; the cognitive schema to recognize it does not.

Rahman (2021) makes a related point in a different context, describing how platform-dependent workers often cannot see the structural features of the systems governing their labor. The parallel here is organizational rather than individual: firms participating in the broader digital economy are generating data that feeds AI training pipelines without any framework for recognizing, valuing, or negotiating over that contribution.

The Awareness-Capability Gap, Applied to Firms

Most of my research on the awareness-capability gap addresses individual platform workers. The gap describes a situation where workers know that algorithms shape their outcomes but cannot translate that awareness into effective behavior. The dead-startup data market suggests a firm-level version of this problem.

Firms are not unaware that AI labs want data. The Slack DM story has generated significant press coverage. But awareness of the market does not automatically produce the organizational capability to participate in it on favorable terms, to protect sensitive data from unwanted extraction, or to build internal governance structures that treat communication archives as managed assets rather than passive byproducts. Awareness and capability remain decoupled, and the decoupling is consequential.

Hatano and Inagaki (1986) distinguished between routine expertise, which handles known problem types through established procedures, and adaptive expertise, which applies structural understanding to novel situations. Organizational data governance frameworks were built for routine expertise: classify documents, restrict access, comply with known regulatory categories. The AI training data market is a novel situation. It does not map cleanly onto existing governance schemas, which is precisely why a startup founder can be surprised to learn his data is worth millions and why operating companies may be selling communication data without fully understanding the downstream implications.

The Organizational Schema Problem

Gentner's (1983) structure-mapping theory argues that transfer of knowledge across contexts depends on recognizing structural similarities rather than surface features. Applied here: firms that successfully navigate the AI data market will not be the ones with the most aggressive legal teams or the most detailed data policies written for a pre-AI context. They will be the ones whose leadership has developed an accurate structural schema for what AI training data markets are actually selecting for, and why.

That schema does not currently exist in most organizations. Internal communications were never conceptualized as training corpora. Slack was sold to firms as a productivity tool, not as a data asset with market value to third parties. The surface features of these systems - familiar interfaces, internal-facing design, routine business use - actively obscure their structural position in the broader AI data economy.

Why This Matters for Organizational Theory

Classical coordination theory, whether market, hierarchy, or network-based, assumes that parties entering an exchange have some working model of what is being exchanged and why it has value (Kellogg et al., 2020). The dead-startup data market violates this assumption at the organizational level. One party, the AI lab, has a precise structural understanding of what communication data enables and how to value it. The other party, the startup winding down or the operating firm fielding an unsolicited offer, frequently does not.

This is not a temporary information asymmetry that market mechanisms will eventually correct. It is a schema asymmetry rooted in different levels of structural understanding. Until organizations develop accurate models of why their communication data is valuable - not just that it is valuable - the asymmetry will persist, and the market will continue to price it accordingly.

References

Gentner, D. (1983). Structure-mapping: A theoretical framework for analogy. *Cognitive Science, 7*(2), 155-170.

Hatano, G., & Inagaki, K. (1986). Two courses of expertise. In H. Stevenson, H. Azuma, & K. Hakuta (Eds.), *Child development and education in Japan* (pp. 262-272). Freeman.

Kellogg, K. C., Valentine, M. A., & Christin, A. (2020). Algorithms at work: The new contested terrain of control. *Academy of Management Annals, 14*(1), 366-410.

Rahman, H. A. (2021). The invisible cage: Workers' reactivity to opaque algorithmic evaluations. *Administrative Science Quarterly, 66*(4), 945-988.