Rollouts don't fail at installation; they fail at people. This page is the playbook I run: five principles every successful implementation shares, a learning ladder that right-sizes every tool to the person using it, and a 30/60/90 built to land one measurable win early - then scale a proven motion, not a hopeful one.
The playbook
"We rolled out AI and nothing changed" isn't a mystery - it's what happens when one of these is missing. Get all five working and adoption stops being a hope and becomes a system.
Adoption needs a named owner and executive permission. A sponsor, a body that vets tools and sets responsible-use policy, and clear top-down air cover are what turn "people are trying AI" into a program - change management, not a tech rollout parked with IT.
Strategy sets direction top-down, but the real signal comes from the work people already do. The losing move is launching "an AI initiative" in the abstract; the winning move is attaching AI to tasks people already perform - tangible and urgent instead of optional. Discovery is a standing habit, not a one-time survey.
Peer advocates embedded inside teams are the highest-leverage mechanism in adoption - people trust a colleague who does their job over a mandate from above. In one widely cited example, Citi built a ~4,000-person champion network across ~182,000 staff and passed 70% adoption without mandating use. Most organizations already have champions-in-waiting; the job is to find and equip them, not manufacture them.
One curriculum for everyone fails everyone. Adoption climbs a ladder - literacy, adoption, transformation - and the trick at every rung is to right-size a tool to a single job, so a person learns one thing, not a platform. Small changes stick where sweeping ones stall.
You can't manage what you don't measure. A baseline captured before anything changes, plus honest before-and-after on named processes, turns anecdotes into evidence - which justifies the program to leadership and persuades the next hesitant adopter. The harder half is measuring value, not usage.
The learning ladder
Everyone can reach the tool and thinks out loud with it. The single greatest value an LLM adds to a workforce is as a reasoning partner - a habit that generalizes to nearly every role and underwrites everything above it.
Tools embedded in real workflows. This is where first-order use cases live: simple, fast, visibly useful - the quick win that buys trust for everything that follows.
AI stops making old work faster and starts changing what the work is - automations, agents, connected workflows, led by the people who know the domain and built on the fluency of the rungs below.
First-order before second-order, every time. The fast win creates buy-in; the advanced play shows what fluency unlocks. Skip the ladder and you get shelfware with a training deck.
The measurement stack
Before deciding to build or buy a tracking system, it's worth seeing that adoption is a stack. The first three layers are largely given to you — the enterprise AI vendors already ship the telemetry. The bottom layers are where real work, and a real differentiator, live.
The build-vs-buy call follows from the stack. Don't build what's commoditized — usage telemetry is solved, and bolt-on platforms cover proficiency and roadmap reporting. Spend the build budget on the one layer no tool can sell you: stitching your org's specific data together and tying usage to your outcomes. That last mile is where a custom build beats off-the-shelf — and it's the layer that produces the ROI story leadership actually wants.
The plan
Three phases, each with a single job. Month one finds the target and the baseline. Month two lands the win. Month three turns one win into a repeatable system. Every phase ends in something visible.
Job of the month: know exactly where things stand, plug into the governance that already exists, and pick the one team where a fast win is provable.
Job of the month: one team, one measurable, visible result - done right, not done wide.
Job of the month: convert one win into a repeatable program that scales without me as the bottleneck.
What I'd report
Two kinds of metrics, reported differently. Leading indicators show momentum week to week and tell you early if adoption is real. Lagging indicators take longer to move but are what justifies the budget — so I track both from day one and never report one without the other.
Moves in weeks · early signal adoption is sticking
Moves in months · what leadership funds against
Why I can actually run this
Sole owner of enablement for an internal GenAI platform — I drove it from zero to 32% daily-active across 110 staff in 8 units, with road-show workshops, workflow guides, train-the-trainer, and a ~30-person AI community of practice. Adoption was the deliverable, not a side effect.
Three years building RAG systems, agents, and automation on the Anthropic API. The "stitch the sources and tie usage to outcomes" layer that nobody sells you — Python, FastAPI, data plumbing — is work I can ship, not just spec.
Nine-plus years inside a government compliance environment — data classification, audit, access control. The instincts responsible, governed AI deployment demands are already mine.
Turning dense technical systems into the trainings and plain-language guides non-technical staff actually use is the core of my current role — and the exact skill that turns a tool nobody opens into one a team relies on.
Open to implementation, enablement, internal-tools, and applied-AI roles in the Boston area. In-office or hybrid preferred; open to remote.