The Surprise Bill Nobody Should Be Surprised By

A recent piece in the New York Times describes a pattern now emerging across corporate America: organizations that moved aggressively to integrate AI tools are confronting unexpected cost overruns, with the gap between projected and actual expenditure large enough to stall further investment (Herrman, 2025). The framing in coverage of this story treats it primarily as a budgeting failure, a matter of procurement teams underestimating API costs or licensing fees. That reading is not wrong, but it is incomplete. What corporate AI sticker shock actually represents is a structural mismatch between how organizations theorize competence and how algorithmically-mediated systems actually distribute it.

The Procurement Assumption and Its Theoretical Flaw

Classical coordination theory, whether in markets, hierarchies, or networks, assumes what I call ex-ante competence: the relevant actors arrive at the coordination problem already equipped to participate effectively (Kellogg, Valentine, & Christin, 2020). When a firm purchases enterprise software, it typically assumes that users will extract roughly proportional value from roughly equivalent access. The pricing models of most SaaS vendors reinforce this assumption by charging per seat. You buy a seat; you get the tool; the tool gets used.

Platform-mediated AI systems do not work this way. Access does not equal capability, and capability does not distribute uniformly across users with identical licenses. The Algorithmic Literacy Coordination framework I am developing in my dissertation research treats this as the core coordination problem: competencies in algorithmically-mediated environments develop endogenously through participation, not prior to it. This means two employees at the same firm, using the same AI subscription, on the same day, can produce outcomes that differ by an order of magnitude. The cost structure of enterprise AI does not reflect this variance. Organizations pay uniformly for something whose returns are profoundly non-uniform.

Why Awareness of Cost Did Not Prevent the Shock

One might argue that finance teams simply failed to read the pricing documentation carefully enough. But the sticker shock phenomenon is too consistent across firms to reduce to procurement negligence. A more theoretically interesting explanation involves what algorithmic literacy researchers call the awareness-capability gap (Gagrain, Naab, & Grub, 2024). Workers and managers can develop accurate awareness that AI tools exist, that they are costly, and that they require careful use, without this awareness translating into improved deployment outcomes. Knowing that costs will scale with usage does not automatically produce the behavioral schemas required to manage that scaling effectively.

This is the distinction Hatano and Inagaki (1986) draw between routine and adaptive expertise. Routine expertise applies known procedures to familiar problems. Adaptive expertise applies structural principles to novel configurations. Enterprise AI deployment is a novel configuration for nearly every firm encountering it. Organizations that treated AI integration as a procedural rollout, issuing guidelines, training modules, and use-case checklists, generated routine expertise at best. When the environment shifted (and it shifts constantly in algorithmically-mediated contexts), that routine expertise did not transfer. The costs accumulated precisely in the gaps between what the procedures anticipated and what the system actually produced.

The Organizational Theory Problem Underneath the Budget Problem

Rahman (2021) describes how algorithmic control systems create what he calls invisible cages: structural constraints that shape behavior without being legible to those being constrained. Corporate AI deployment presents a variation of this problem that runs in the opposite direction. The cage here is not invisible because it is hidden from workers; it is invisible because the organization itself lacks an accurate structural schema of what it has purchased. Executives approved budgets for AI based on folk theories, individual impressions of how the technology would behave, rather than accurate models of how algorithmically-mediated cost and value actually distribute across user populations (Gentner, 1983).

This is not primarily a technology problem. It is an organizational knowledge problem. Firms that will navigate AI deployment successfully are not necessarily those with the largest budgets or the most sophisticated tools. They are the ones that develop accurate structural schemas of the coordination mechanism they are participating in, schemas that capture topology (the shape of how AI systems amplify input variance into output variance) rather than just topography (which buttons to press in which order). The sticker shock story is, at its core, a story about what happens when organizations buy into a new coordination mechanism without first developing the structural literacy to participate in it effectively.

References

Gagrain, A., Naab, T. K., & Grub, J. (2024). Algorithmic media use and algorithm literacy. New Media & Society.

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. Research and Clinical Center for Child Development Annual Report, 11, 27-36.

Herrman, J. (2025). Corporate America is experiencing AI sticker shock. The New York Times.

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.