Why Giving Employees AI Tools Is Not the Same as Getting Value from AI

Enterprises are handing out AI access faster than they are building the capability needed to use it well — and the gap between the two is where most AI ROI disappears, according to an AI literacy strategist working across healthcare, biotech, manufacturing, and finance.

Alicia Kimiagarov, Senior Manager of Data Analytics at Cerrowire and advisor to the AI Huntsville Taskforce Workforce Development Committee, describes two recurring failure patterns: employees who never adopt the tools they are given, and employees who adopt them in low-value ways. In both cases, the problem was not access — it was the absence of training, context, and peer-to-peer reinforcement.

One example: a business user was given access to an enterprise-approved, low-code agentic AI platform and told to build a solution. With no guidance on what agentic AI could do or how to scope a problem, the project was abandoned. In another case, a biotech director with a full enterprise AI license ended up using it only for emails and presentations.

Key Takeaways

• Three types of literacy are required: Kimiagarov defines DDA literacy as digital (how systems support workflows), data (interpreting quality, context, and fitness for use), and AI (understanding capabilities, limitations, and when human review must take over). These are not technical specialties — they are foundational enterprise expectations.
• Research confirms the pattern: Gartner's 2026 research on AI-augmented citizen development argues that successful scaling requires rethinking roles, skills, and decision rights alongside the technology operating model. McKinsey's 2026 research identifies knowledge and training gaps as the leading barrier to accountable AI implementation.
• Peer-to-peer learning outperforms formal training alone: At an aerospace and defense company, one peer demonstrating a practical use case — cutting 20 hours of weekly manual work — converted a skeptical colleague faster than any corporate training program could. Manager reinforcement and shared success recognition cemented the behavior.
• Measure outcomes, not activity: License counts, active users, and prompt volume are poor proxies for value. The right question is whether AI adoption is producing measurable improvements in time, quality, cycle time, or decision accuracy.

Read the full article on CIOReview