Why AI Benefits New Employees Most

One of the most consistent and counterintuitive findings in the emerging literature on AI productivity is that the workers who benefit most are often not those with the greatest expertise — they are those with the least. Across multiple peer-reviewed studies conducted in different sectors and with different AI tools, less experienced workers show the largest proportional productivity gains from AI assistance. This pattern has direct and significant implications for healthcare organizations navigating workforce strategy in an era of high turnover and costly onboarding.

The Evidence Base

The experience-level effect appears across the most rigorously conducted studies of AI in knowledge work settings. In Brynjolfsson, Li, and Raymond's examination of AI assistance in a large customer support operation, workers in the bottom skill quartile saw productivity improvements nearly four times larger than those in the top skill quartile. The most experienced agents — those whose performance already exceeded organizational benchmarks — saw minimal benefit and in some cases modest performance declines.

A similar pattern emerged in Noy and Zhang's controlled study of AI writing assistance. Workers who rated their own writing skills below average before the study saw larger improvements in output quality and task speed than those who rated their skills above average. The quality gap between low- and high-skill workers narrowed substantially when both groups had access to AI.

The GitHub Copilot study of AI-assisted software development found that less experienced developers completed assigned tasks 55 percent faster with AI assistance, compared to approximately 20 percent faster for their most senior peers. The gap in task completion time between junior and senior developers — a persistent feature of software development team dynamics — narrowed substantially in AI-assisted conditions.

The Mechanism: AI as Encoded Expertise

Why does this pattern appear so consistently? The most compelling explanation is that AI tools trained on large corpora of human knowledge encode and make accessible the kind of tacit expertise that experienced workers have accumulated over years of practice.

An experienced customer service representative, for instance, has internalized thousands of past interactions — she knows which responses de-escalate difficult callers, which documentation is required for different claim types, and which edge cases require escalation. An AI system trained on similar data can surface relevant guidance in real time, effectively giving a new employee access to a compressed form of institutional knowledge she would otherwise take years to develop.

This mechanism explains both why new employees benefit most and why experienced workers benefit least: the AI provides the most value precisely where the worker lacks the knowledge the AI encodes. For workers who already possess that knowledge, the AI adds less marginal value — and may occasionally interfere with established personal workflows.

Implications for Healthcare Operations

Healthcare is an industry where the experience-level dynamics documented in productivity research are especially salient. Several features of healthcare operations make the new-employee AI benefit particularly relevant.

High Turnover and Onboarding Cost

Healthcare organizations — particularly in administrative functions, contact centers, and non-clinical support roles — experience significant workforce turnover. Turnover rates in healthcare administrative roles have ranged from 20 to 40 percent annually in recent survey data, with some contact center functions experiencing even higher rates. The cost of this turnover — recruitment, training, and the productivity trough during onboarding — is substantial.

If AI tools can meaningfully accelerate the rate at which new employees reach acceptable performance levels, the economic implications extend beyond direct productivity gains. Shorter onboarding periods reduce the revenue and quality costs associated with the performance gap between new and experienced staff. In organizations with high turnover, this effect compounds: a faster ramp-up curve applies to a larger share of the workforce at any given time.

Contact Center and Member Services Functions

Health plan contact centers and member services operations share structural characteristics with the customer support environments studied in the peer-reviewed AI productivity literature. Representatives handle high volumes of inquiries requiring accurate information retrieval, policy interpretation, and documentation — exactly the task types where AI assistance shows the largest performance benefits for less experienced workers.

Organizations operating these functions face a persistent challenge: experienced representatives are expensive to recruit and retain, and the performance gap between new and experienced staff creates service quality variability that is difficult to manage at scale. AI assistance that narrows this gap — by surfacing relevant policy information, suggesting response approaches, and guiding documentation — addresses the operational problem directly.

Clinical Administration and Revenue Cycle

Revenue cycle operations — coding, billing, denial management, prior authorization processing — are characterized by complex rule sets that take significant time to master. New coders, for example, may require twelve to twenty-four months to reach expert performance levels, during which their error rates and productivity significantly lag those of experienced peers.

AI assistance in revenue cycle functions could plausibly accelerate this development trajectory by surfacing relevant coding guidance, flagging likely errors, and suggesting documentation requirements — the same type of encoded expertise that produced performance gains for less experienced workers in adjacent industries. Healthcare-specific evidence on this effect is limited but the structural conditions are closely analogous.

Cautions and Limits

The experience-level effect observed in research studies warrants cautious extrapolation. Several limitations deserve explicit acknowledgment.

First, the studies establishing this effect were conducted in non-healthcare settings on tasks with clearer performance criteria than many healthcare roles. The degree to which findings transfer to clinically adjacent or clinically complex functions is not yet established.

Second, the possibility of skill atrophy deserves attention. If new employees develop competency by relying on AI assistance, they may reach acceptable performance levels without building the deeper knowledge that produces expert performance. Long-term effects on workforce capability development are largely unstudied.

Third, the performance gains documented in research settings may not fully replicate in operational deployments where AI tool quality, workflow integration, and staff adoption vary. Research conditions typically involve controlled tool deployment with active monitoring; real-world adoption is messier.

Strategic Implications

For healthcare organizations, the experience-level effect suggests a specific lens for evaluating AI tool investments: rather than asking whether AI makes experienced workers more productive — an effect that the evidence suggests may be modest — the more productive question is whether AI can accelerate the development of new workers and narrow the performance gap that drives operational variability and workforce cost.

This reframe has practical consequences for deployment strategy. Organizations optimizing for workforce economics may find the strongest return on investment in functions with high turnover, significant onboarding burdens, and large performance gaps between new and experienced staff — rather than functions where experienced workers dominate and the performance ceiling is already high.

The research does not suggest that experienced worker productivity is unimportant or that AI tools should be deployed only for new employees. It suggests that workforce strategy and AI deployment strategy are more tightly linked than many organizations currently recognize — and that the experience-level effect is a predictable feature of AI adoption that can be anticipated and planned for.