AI Skills Gap Assessment Starts At The Project

AI skills gap assessment reveals a dangerous blind spot: 93% of leaders cite skills barriers to AI progress, yet 68% believe their teams are keeping pace.

Slalom's 2026 research found that 68% of leaders believe their workforce keeps pace with AI. The same survey found that 93% of those leaders and employees identify underdeveloped skills as the primary barrier to AI progress. Both numbers come from the same population. That contradiction is not a rounding error — it is a description of how AI project failures happen.

The gap you cannot see from a workforce dashboard

Leaders are not lying when they say their teams are ready. They are estimating. MIT's 2022 survey of executives found that the same group responsible for closing skills gaps estimated 38% of workers need fundamental retraining or replacement within three years. Estimated. Not measured. Not mapped to specific delivery requirements. The number floats free of any project, role, or deadline.

General workforce readiness scores cannot catch this. A dashboard showing 70% AI literacy across your organisation tells you nothing about whether the four people running your customer churn prediction model know how to audit its outputs for bias. Those are different questions, and only one of them predicts whether the project ships.

Why the gap compounds instead of closing

Julia Coronado, president of MacroPolicy Perspectives and a Robert Half board director, identified the structural problem in Fortune's 2025 analysis: AI is eliminating entry-level roles, which are the positions where workers historically develop the judgment needed to oversee AI systems at the mid-level. Remove the entry-level pipeline and you remove the development path. The workers you need to supervise AI outputs are not being grown inside your organisation anymore. [Inference: the rate at which this pipeline erosion affects specific industries varies and is not quantified in the source.]

The WEF's 2025 data shows AI literacy acquisition grew 177% since 2023. That sounds like progress until you read the next figure: demand for AI skills grew sixfold in the same period. Acquisition is moving in the right direction and losing ground.

What a project-scoped assessment actually looks like

The organisations that catch gaps before delivery fails do not start with the workforce. They start with the project. Step one is mapping the specific AI capabilities the project requires — not "data literacy" as a category, but the ability to validate training data for a specific model type, or to interpret confidence intervals from a specific output format.

Step two is assessing the people assigned to that project against those requirements. Not a general skills survey. A structured gap analysis tied to the roles on the delivery team, using the project's technical requirements as the benchmark.

Step three is prioritising which gaps to close before kickoff versus which ones to manage through pairing, documentation, or scope adjustment. A Protiviti C-suite respondent told Fortune in 2025 that external hiring for niche AI skills is prohibitively expensive for most organisations, which means the prioritisation decision is also a resource allocation decision. You are not choosing between training and hiring. You are choosing which gaps are small enough to close internally before the project starts and which ones will sink it if left open.

The case for AI-driven skills inference

Johnson & Johnson's approach, documented in MIT Sloan's 2022 coverage, uses AI to analyse employee data and generate role-level skills insights without a manual assessment cycle. For a large enterprise with mature HR data infrastructure, this is a legitimate alternative. The system is faster, less dependent on self-reporting, and continuously updated.

The problem is what it cannot see. AI inference tools trained on historical employee data detect skills that have already been recorded — credentials earned, projects completed, roles held. Mid-level AI oversight capability, the specific gap Coronado identifies, has not been required at scale before. It does not appear in past project assignments because no one has needed to log it. A backward-looking system cannot surface the absence of a skill the organisation has never had reason to name. The J&J model works well for known skills in established roles. It fails precisely where the delivery risk is highest.

SHRM's 2026 report found 51% of workers identify enhanced training as their top priority for AI outcomes. That demand exists. The missing piece is not willingness — it is a targeting mechanism that connects the training to the project requirement before the deadline arrives.