How We Helped a 5,000-Person Company Reskill for AI in 6 Months

When a large insurance company's technology leadership team decided they needed to integrate AI tooling across their operations, they had three options. Hire from outside — expensive, slow, and increasingly competitive. License AI tools and hope employees figured them out — almost universally unsuccessful. Or systematically reskill the workforce they already had.

They chose the third option. What followed was a six-month program that moved 5,000 employees from baseline digital literacy to active AI tool usage, with measurable productivity outcomes and retention that outperformed their industry benchmark. This is the story of how we built it together.

The Starting Point

The company had two serious structural problems before any training could happen. First, they didn't know what AI readiness actually meant for their workforce. They had a vague mandate from their CEO to "prepare the organization for an AI-enabled future," but no definition of what specific behaviors that required from a claims adjuster, a customer service rep, or a data analyst.

Second, they had no reliable baseline. Their HR system had skills data that was three years out of date and based entirely on job descriptions, not actual assessed competency.

We started with a three-week competency definition phase. Working with functional leaders across the five largest business units, we defined AI readiness at three levels: foundational (every employee), applied (team leads and project-adjacent roles), and advanced (analysts, developers, and process owners). Each level had specific, observable behavioral descriptors — not "understands AI" but "can identify and document a manual task in their role that could be augmented by AI and articulate the expected output format."

The Assessment Phase

Week four through six was baseline assessment. We used a mix of scenario-based assessments, work sample reviews, and manager evaluations to place all 5,000 employees on the competency map. This sounds slow. In practice, the scenario-based assessments took about 35 minutes per employee and were completed during existing work time over a two-week window.

What came back was illuminating. The distribution looked nothing like what the company expected. Their most confident self-assessors — typically senior employees in traditional roles — scored lowest on scenario-based tasks. Younger employees in customer-facing roles, who the company had assumed needed the most development, scored significantly higher on foundational competencies. The actual skill gap was the opposite of where they'd planned to focus.

This alone saved the company from running a training program aimed at the wrong population.

Building the Program

With accurate baseline data, we built three parallel learning tracks. Each track was eight weeks long with two hours of required engagement per week plus an optional office-hours session. Content was a mix of internal-use-case videos developed with the company's operations team, curated external modules from two specialist providers, and weekly applied assignments tied to each learner's actual job responsibilities.

The applied assignments were the critical differentiator. A claims adjuster's week-three assignment wasn't "complete this module about AI." It was "document three repetitive tasks in your current claims process and use the AI drafting tool to attempt one of them. Submit your output and a one-paragraph reflection on what worked and what didn't." Their manager received the submission and was prompted to discuss it in their next 1:1.

That feedback loop — learner tries it, manager sees it, they discuss it — is what converts information into habit.

What Six Months Produced

At the end of the program, we ran post-assessments against the same competency framework. The results:

87% of the foundational learner population reached or exceeded the defined competency threshold, up from 31% at baseline. 64% of the applied track completed all eight weeks, and post-assessment showed 79% reaching the target competency level. The advanced track had 71% completion with 82% reaching target — the highest performer in absolute terms, partly because this group had more intrinsic motivation and more direct application opportunities.

On business outcomes: the company measured three things over the following quarter. Claims processing time for adjusters who had completed the program was down 18% compared to the control group that hadn't yet gone through. Customer service resolution accuracy was up 11%. Manager-reported confidence in AI tool usage — a softer metric but one the company tracked — went from 2.4 to 4.1 on a five-point scale.

Voluntary turnover in the cohort during and after the program was 6.2%, compared to an industry average of 14.8% for comparable roles. People stay when they're growing.

What Made This Different

The company had tried two previous "AI readiness" initiatives over the prior 18 months. Both failed to produce behavior change. When we reviewed them, the difference was stark. The previous programs delivered content. This program delivered competency development — and the distinction isn't semantic.

Content delivery says: here is information. Competency development says: here is information, now apply it to your work, have your manager observe you applying it, and be assessed on whether you can do it without assistance. The second is more demanding. It's also the only one that works at scale.

The other factor was manager activation. We ran a four-hour manager enablement session for all 280 people managers in scope before the program launched. We told them explicitly what their role was: not to teach the content, but to create application opportunities and hold reflection conversations. Their involvement converted individual learning into team behavior change.

Applicability

This program was large and well-resourced, but the architecture works at smaller scale. The core principles — accurate baseline assessment, role-specific competency definitions, applied assignments tied to real work, manager accountability loops — are as effective for a 200-person department as for a 5,000-person company. Scale changes the logistics, not the model.

The organizations that succeed at large-scale reskilling are not the ones that run the most training hours. They're the ones that build the clearest definition of what they're trying to change and then hold the program accountable for changing it.