Our Take
Self-attestation is not assessment; organizations measuring first are already building resilience while others staff projects with confident imposters.
Why it matters
As AI automation accelerates, HR leaders betting on course completion and self-report data are deploying workforces that cannot actually build or troubleshoot agentic systems. The gap between claimed and verified skills is now a business risk, not just a training problem.
Do this week
CHRO: Run a skills assessment on your top 100 employees this quarter before staffing your next AI initiative, so you can identify actual bottlenecks instead of discovering them mid-project.
88,753 assessments expose a massive skills-reporting gap
Workera's 2026 AI Skills Enterprise Benchmark Report, based on verified assessments of nearly 89,000 employees (per Workera), found that what people claim to know about AI rarely matches what they can actually do.
The platform uses a 300-point scale where scores above 200 indicate ability to design and build AI solutions, not just recognize concepts. Deep Learning Fundamentals averaged 142 across enterprise employees. Agentic AI Fluency and Engineering averaged 179—both in the "developing" range, meaning most employees can discuss these skills but cannot apply them effectively.
Entry-level skills with low technical barriers performed strongest. Data Storytelling Essentials, AI and Data Communication, and Responsible AI Essentials led the benchmarks. The collapse happens where technical depth is required.
ServiceNow's approach, detailed at the Wall Street Journal Leadership Institute's CPO Council Summit, shows what measurement-first looks like in practice. Chief People and AI Enablement Officer Jacqui Canney described assessing all 30,000 employees by job and level, setting percentile targets for each capability, then giving employees transparent access to their scores and personalized development paths. "We didn't make it a stick," Canney said. "It was more like an incentive."
The real risk: bottlenecking at scale
As enterprises accelerate automation and AI-assisted workflows, the gap between claimed and actual skill creates two overlapping costs. First, projects stall when organizations discover their deep technical staff cannot handle the work they thought they could delegate. Second, and more subtle, organizations with only a small cluster of advanced practitioners risk those people becoming single points of failure across the whole pipeline.
The report shows that targeted training works when focused. Employees who upskilled in Data Visualization and Storytelling improved by 77% on average. Generative AI Essentials improved by 51%. Responsible AI jumped from 25% accomplished to 94% (per Workera). But improvement rates vary significantly by capability. Some skills respond quickly to short courses; others like Machine Learning Fundamentals require sustained effort.
The real issue is not that employees are lying. It is that course completion and self-perception are not proxies for competence. HR leaders who treat them as such are staffing their AI initiatives with untested assumptions.
Three moves for HR and ops leaders
Measure before you staff. Before assigning employees to AI projects or building center-of-excellence teams, run a skills assessment on your candidate pool. Workera's framework is one option; others exist. The point is to know what you actually have.
Tier your training by skill type. Data communication and responsible AI train fast and stick. Deep technical skills do not. Budget accordingly and set realistic timelines.
Make scores transparent to employees and tie them to development, not punishment. ServiceNow's incentive model proved more effective than using assessment data as a compliance stick. Employees who see their gap relative to their role are more likely to close it.