We hit 98% voluntary adoption of AI across 900 professionals in 11 offices. No mandate. No top-down requirement. No "use this or else." I want to explain exactly how, because the playbook is replicable, and most enterprise AI rollouts are failing for reasons that have nothing to do with the technology.
The standard enterprise approach to AI adoption is: buy a platform, send a company-wide email, run a few training sessions, then wonder why usage flatlines at 15%. I know because I watched it happen at other firms before I built our program. The problem is never the tool. The problem is that adoption is a design problem — it requires the same intentionality you'd bring to designing a product — and most companies treat it as an announcement problem instead.
Here's the system I built, broken into the specific design decisions that made it work.
This was a deliberate choice, not a constraint. I could have pushed for a mandate. I didn't want one. Mandated adoption produces compliance, not competence. People log in to check a box. They don't build real workflows. They definitely don't become the power users who transform how a practice group operates.
Voluntary adoption forces you to be good. If people can walk away, you have to earn every session. That pressure made the program better. Every piece of training content, every new skill in the library, every onboarding flow — all of it was built with the assumption that people had full permission to ignore it. That constraint was a gift.
Nobody in a professional services firm cares that your AI tool has retrieval-augmented generation. They care that it can cut their research synthesis time from four hours to forty minutes. Every single training I ran, every rollout communication, every one-on-one — I led with the outcome, not the capability. "This will save you three hours on your next deal memo." "This will let you compare regulatory frameworks across six jurisdictions in minutes instead of days."
I built a library of 40+ skills — structured, reusable AI workflows mapped to specific professional tasks. Each skill was named for what it does, not how it works. People didn't need to understand the architecture. They needed to see themselves in the use case. The moment someone recognizes their own Tuesday afternoon in a demo, they're in.
This is the single most counterintuitive decision in the playbook, and it might be the most important one. Most AI rollouts train junior staff first — the logic being that younger employees are more tech-savvy and will adopt faster. That logic is exactly backwards for professional services.
In an investment bank, in a law firm, in a consulting practice — juniors model senior behavior. If a senior professional is using AI visibly, juniors get implicit permission to do the same. If seniors aren't using it, juniors assume it's not serious, regardless of what the training materials say. Culture flows downhill.
I trained senior leaders first. I sat with them one-on-one. I built workflows around their specific matters and deals. When they started using AI in their actual work — and talking about it in team meetings — adoption in their groups accelerated without me touching it. The seniors became the signal. The juniors followed the signal.
Across 8 practice groups, usage was never evenly distributed, and I stopped trying to make it even. In every group, a subset of users — roughly 40% — became power users who accounted for a disproportionate share of total activity. These weren't always the most senior people. They were the ones who got genuinely excited about reshaping their work.
I identified these people early and invested in them heavily. Extra training. Early access to new skills. Direct line to me for troubleshooting. They became the local champions — the person in the group that others went to before they came to me. This was scalable in a way that my own bandwidth wasn't. I couldn't personally train 900 people on an ongoing basis. But I could equip 30 champions who collectively had relationships with all 900.
The first time someone uses an AI tool at work, they're forming an opinion that will persist for months. If that first session is confusing, slow, or produces a mediocre result, you've lost them — and getting them back costs ten times more effort than getting the first session right.
I obsessed over the first-session experience. I pre-loaded prompts. I built starter workflows that were almost impossible to use incorrectly. I made sure the first thing someone saw was a result that made them say "wait, it can do that?" Not a tutorial. Not a feature tour. An immediate, visceral demonstration of value applied to their own work.
This meant I had to know each practice group's actual work well enough to build those starter moments. I studied their deal workflows, their research processes, their document review patterns. The first-session experience wasn't generic — it was tailored to the work they did that morning. That specificity is what made it land.
I built dashboards that tracked adoption, usage depth, and skill utilization — and I checked them constantly. Not monthly. Not quarterly. Daily during rollouts, weekly during steady state. When I saw a practice group's usage dip, I called the champion that day. When I saw a new skill getting unexpectedly high adoption, I doubled down on similar skills immediately.
Crucially, I reported these numbers upward. Leadership saw adoption metrics regularly. This created a virtuous cycle: when firm leadership sees that 83% of users are actively engaged — not just licensed, but actively using the system — it reinforces investment in the program. It also created healthy competition between practice groups. Nobody wanted to be the group with the lowest adoption rate when the numbers were visible.
Fast measurement also let me self-fund the program. I could demonstrate net-neutral cost management — the AI stack paid for itself — building toward ~$1.5M in broader value. Those weren't projections — they were documented, attributable savings calculated from actual usage data. When you can show a return within weeks, not quarters, budget conversations change entirely.
If I were starting from zero at a new organization tomorrow, here's the sequence:
Most enterprise AI programs fail not because the technology doesn't work, but because the rollout was designed like a software deployment instead of a behavior change campaign. The technology is the easy part. Designing for voluntary, sustained, deepening adoption — that's the actual work. And it's the work that most organizations skip.
If you're planning an AI rollout — or trying to rescue one that stalled — I've been through the full cycle and I'm happy to compare notes. Reach out or find me on LinkedIn.