Overview
At a senior-healthcare company, I led the company-wide AI adoption (AX) project. I designed 5 tracks (sales dashboard, new business discovery, LLM Wiki vault, onboarding, and skills training) and scaled a system I had validated in my own vault to the whole team.
Problem
Company data scattered across email, Slack, and documents never built up into knowledge, and the “what and how” of AI adoption was unclear. Everyone shared 1 common AI account, which made personalized use of AI impossible in the first place.
Approach
I defined step 1 not as “let’s use AI” but as “first build a path that leaves a record.” I learned the LLM Wiki methodology (started the course in May → completed it in August) and validated an ingest, tag, lint, and hook system in my own vault first, then secured approval through an executive report (5 build objectives, a 4-stage data-connection roadmap, role-based use cases, and 5 safety principles). I ran 1:1 setups with team leads for 5 consecutive business days and turned the sticking points into checklists, STEP A (9 manual steps with the instructor, 0 AI usage) and STEP B (installation prompt), creating a package anyone can install without help. I also redesigned the account structure to 1 account per person. For meeting notes, I built my own local speech-to-text pipeline that files them automatically.
Results
A 123-page vault (measured at the end of September) and 50-plus ingests in total. I turned the 1:1 setup experience into a 22-step installation guide, and the company decided to publish it. In October, I carried out the switch to Claude’s Team plan myself and finished issuing and connecting accounts for all employees. The meeting-notes pipeline (recording → local transcription → AI summary → Slack post) is used for every weekly meeting, and the path for sharing meeting notes was adopted as a company-wide operating rule. The installation package does double duty as the manuscript for an external training curriculum and as the internal onboarding playbook, and it became the template for a business proposal to build “AI wikis” for SME CEOs. Transcription processes a 56-minute recording in 20 minutes.
Lessons Learned
“We did this much” and “we saved this much” carry very different weight. Only the latter justifies a customer’s adoption and an internal budget. The next step after building is quantifying the value it creates, and I have already identified what to measure.