The Hidden Work in HR: Building on MIT’s Project Iceberg

I hope you're doing well. This week, we were deeply inspired by MIT’s newly released Project Iceberg, a study that revealed a profound blind spot in how nations measure AI impact. MIT showed that while only 2.2% of wage value reflects visible AI adoption, a much larger 11.7% sits beneath the surface—cognitive, administrative, and coordination tasks that AI can already technically perform. Their central argument is powerful: the indicators we track today represent only the tip of the iceberg.

Motivated by this insight, we examined how the same “iceberg problem” appears inside the enterprise. The findings are striking. Just as national labor metrics miss hidden AI exposure, enterprise metrics fail to reveal hidden work—the real source of productivity, cost, process friction, and AI opportunity.

The paper builds on MIT’s Project Iceberg, which shows that only 2.2% of visible U.S. labor activity reflects AI adoption, while 11.7% of wage value lies in hidden cognitive and coordination tasks. Draup applies this same insight to enterprises, arguing that most work inside companies is also invisible to traditional HR and organizational systems.

The paper notes that job descriptions capture only a small fraction of actual work, and that HR often relies on fragmented business inputs to understand jobs.

Process maps, workflows, KPIs, roadmaps, and skill taxonomies give some signals—but the deeper work involving judgment, coordination, exceptions, and cognitive load still needs to be studied and documented by HR.

Draup calls this the enterprise measurement gap, mirroring MIT’s national-level gap. This hidden work influences productivity, quality, delays, and the potential for AI automation far more than the surface-level tasks tracked by HR.

To address this, Draup introduces its Workload Iceberg Framework, which makes hidden work measurable through workload and task-level modeling. The methodology decomposes roles into workloads, tasks, skills, and interactions with the tech stack and AI models. Draup developed the Etter model, which includes a dedicated Input Repository that ingests diverse job-related documents, including JDs, process maps, SOPs, audit logs, call transcripts, case management data, portfolio plans, and peer benchmarks.

Using more than 500,000 tasks and skills, the platform models cognitive load, task complexity, task adjacency, skill transferability, and automation feasibility. Draup then applies its Dynamic Skills Architecture, which includes root, core, digital, and AI model skills, as well as sunrise and sunset skills and various skill-weighting techniques.

The framework is operationalized through ETTER, which measures automation feasibility across Etter, GPT-5, Claude, and Gemini, simulates human→AI→human loops.

From this analysis, several insights emerge.  (Here is a sample set of insights)

• First, the greatest AI automation potential exists in hidden tasks such as documentation, validation, exception handling, and coordination—none of which appear in job descriptions.
• Second, role convergence starts in invisible tasks, long before job architectures reflect it (e.g., cloud + data + security or HR + case management + AI review).
• Third, sunrise skills appear inside hidden workloads before they show up in job descriptions, echoing MIT’s finding that capability changes precede formal skill demand.
• Fourth, workforce planning must shift from job-based thinking to modeling workloads, task frequency, cognitive burden, and skill velocity.
• Finally, the paper notes that R&D work is especially ambiguous because it evolves through experimentation and rapid learning cycles.

The paper ends with an illustrative example from a call center. While the job description lists only answering calls, resolving issues, and documenting interactions, real work observed through double-jack listening includes emotional interpretation, multi-threaded system navigation, and back-channel coordination—none of which appear in formal artifacts but define the majority of cognitive effort.

The question we want to leave you with this weekend is simple: Can HR step up to lead the critical effort of documenting the real work happening across the enterprise?