Most firms approach AI the way they'd approach an ERP migration. Someone commissions a strategy deck, a committee meets for a quarter, a vendor is selected, licenses are bought for everyone, and a launch email goes out. Six months and a meaningful budget later, the partners ask how adoption is going, and the honest answer is that a handful of associates use it for a few things and everyone else opened the tool once. The deck was excellent. Nothing about how the work gets done actually changed.
That's not a failure of the technology, and it's rarely a failure of the people. It's a failure of sequencing. The expensive, slow rollout front-loads everything that doesn't change behavior, the strategy, the procurement, the abstract training, and back-loads the only thing that does: a professional sitting at their own desk, on their own real work, watching the tool do something they actually needed done. You can get to that moment in weeks, not months, if you stop treating AI adoption as a transformation program and start treating it as a skill people practice on the job they already have.
The strategy deck is the slow part, and it changes nothing
There's a comfortable logic to starting with strategy. You want a roadmap, a governance framework, a clear view of which tools to standardize on. But in a professional-services firm, none of that touches the actual bottleneck. The bottleneck is that an experienced lawyer, accountant, or architect has a way of working that's served them for fifteen years, and abstract slides about "AI-enabled workflows" give them no reason and no method to change it. They nod in the session and go back to doing exactly what they did before.
Worse, the long rollout actively erodes momentum. Every week between the kickoff and the moment someone uses the tool on real work is a week for skepticism to set in, for the early enthusiasm to fade, for the project to compete with billable hours and lose. By the time training arrives, it's a generic demo on generic examples that look nothing like a client matter, and the most capable people in the room have already quietly concluded it isn't for them. The strategy was never the constraint. Behavior was.
Start with the workflows they already repeat
The fastest way to change how someone works is to start with something they do over and over and already wish were faster. Every professional-services firm has these. They're not exotic, and they don't need a discovery phase to find. Ask anyone on the team what part of their week they'd hand off if they could, and they'll tell you in one sentence.
- A lawyer summarizing a long contract or comparing a counterparty's changes against the firm's standard positions.
- An accountant pulling line items off invoices, bank statements and CFDIs, then reconciling what doesn't tie out.
- An architect turning past-winning proposals into a first draft of a new RFP response in the studio's own voice.
- Anyone drafting the third near-identical client email of the day, or writing up notes from a meeting they just left.
When the first thing someone tries is the task they were already dreading, two things happen at once. They get a result they can judge immediately, because they know exactly what a good answer looks like. And they feel the time saved on work they actually had to do, not on a sandbox exercise they'll forget by Friday. That's the moment a tool stops being a curiosity and becomes a habit, and you can engineer it on the first day instead of the sixth month.
Hands-on, on their own files, beats any demo
There's a real difference between watching someone demonstrate AI on a fictional case and using it yourself on the matter open on your screen. The demo is impressive and forgettable. The hands-on session is occasionally messy, because real documents are messier than examples, and that's exactly why it works. Your team learns where the tool genuinely helps and where it confidently gets things wrong, on the kind of material they'll actually feed it, with someone next to them to correct the habit before it sets badly.
This is also where confidentiality stops being a slide and becomes a reflex. In a Mexican firm handling client financials, privileged matters or personal data, the rules under the LFPDPPP and professional secrecy aren't optional, and the worst version of AI adoption is the associate quietly pasting a client's contract into a public chatbot because nobody taught them the safe path. Training on real files means teaching, in the same breath, what's safe to put where, which work needs a private setup, and where a human still has to verify and sign. The judgment and the skill are learned together, on the same documents, which is the only way either of them sticks.
Reinforcement is what makes it hold
Here's the part the one-day workshop gets wrong. People learn a new way of working in a calm session and then abandon it the first week real deadline pressure returns, because the old way is faster under stress, even when it's slower overall. A single intensive burst of training, however good, decays. The fluency that survives is the fluency that gets used on live work in the weeks right after, with someone available to answer the specific question that comes up on a specific file.
That's why reinforcement isn't an add-on, it's the half of the program that determines whether the rest was worth it. After the hands-on phase, people need a short window where they can bring the actual problem they're stuck on, the email that won't come out right, the reconciliation that's behaving strangely, the proposal that's due Thursday, and get unstuck on it. A few weeks of that, attached to real deadlines, is what turns a promising first week into a way the firm now works. It costs far less than a six-month program and changes far more.
What this looks like in practice
So the practical answer to "how do we get our team using AI" is almost the opposite of the transformation playbook. Skip the strategy deck. Don't buy a platform for everyone before anyone has touched it on real work. Pick the workflows your people already repeat and already resent, train them hands-on on their own files with confidentiality built into the same lesson, and then stay close enough for a few weeks that the habit survives the first real deadline. That's measured in weeks of focused effort, not quarters of meetings.
The firms that get this right aren't the ones with the most ambitious AI strategy. They're the ones who understood that adoption is a behavior, not a purchase, and that behavior changes fastest on the work people already care about. A license nobody opens isn't adoption, and a deck nobody acts on isn't a rollout. The thing that actually moves a firm is smaller, faster and far more concrete than the six-month version, and it starts with the work already sitting on everyone's desk.