Start with the business value.

Pick a use case

Wrong question: "What can AI do for us?" The right questions:

• Where are we losing time, quality, or money — and is AI the right fix?
• Where is the opportunity we're not yet capturing?
• Where do your best people spend time on things that shouldn't require your best people?

The best starting points: repetitive work that consumes valuable time, quality issues that keep coming back, bottlenecks that slow teams down. Start narrow. Pick 2–3 real use cases, not a broad training program. When people see AI solving a real problem they deal with every day, adoption doesn't need to be pushed. It happens on its own.

Create visibility

You can have the right strategy, the right tools, and the right use cases. It won't matter if leadership is invisible. Employees follow behavior — they watch what their leaders actually do, not what gets announced in an all-hands. If leaders talk about AI but don't use it, the message teams receive is simple: this isn't really important.

Define value early

If you launch with goals like "improve efficiency," six months later nobody can answer whether it worked. Not because nothing happened — but because no one agreed upfront on what they were measuring. Agree on four things before you start:

A simple one-pager before launch is enough. But it has to exist — and everyone involved has to agree on it.

Set up governance

The best governance does one thing: it gives people the confidence to move. Clear boundaries are permissions, not restrictions. When your team knows what tools are approved, what data is off-limits, and what good output looks like, they stop hesitating and start working.

• Acceptable use policy — one clear page: approved tools, prohibited data, human review requirements. A living document that grows with you.
• Risk categorization — an internal drafting tool and a customer-facing AI are not the same thing. Apply the right level of oversight to each.
• Data and security basics — know where your data goes before rollout. Where is it stored? Is it used to train vendor models? Who has admin access?
• Make the right path the easy path — if approved tools work well, people use them. Shadow AI is a governance failure.
• Ongoing monitoring — build review cycles in from the start. Models drift, requirements change, new risks emerge.

Automate · Augment · Avoid

Give your team a simple lens for deciding when to use AI — and when not to. Less confusion. Faster decisions. And a shared language across the whole organization.

Embed AI into workflows

The fastest route to real adoption is making AI the path of least resistance — not an extra option.

• Redefine the task — "here's the new way this gets done"
• Remove the old path where possible
• Make the first experience fast and clearly better — the best onboarding is a workflow that immediately saves someone twenty minutes

Managers are the key to adoption

Not IT. Not the leadership deck. Not another all-hands. Employees take cues from their direct manager. If the manager says "this is how we work now," adoption happens. If the manager is uncertain or disengaged, adoption stalls. The good news: managers don't need to be AI experts. They just need to be equipped.

• The why in plain language — a clear answer to "why are we doing this and what changes for my team?"
• A picture of good — what does strong adoption look like on their team, in their workflows?
• A way to handle hard questions — their team will ask them. Simple, honest answers ready to go.
• Permission to start small — run a pilot before going wide. Confidence comes from doing, not training.

Create psychological safety

Teams adopt AI faster when people feel safe to try, fail, and learn in front of each other. That safety doesn't happen automatically — it has to be built deliberately. AI exposes competence gaps in real time, in front of colleagues. That's uncomfortable. And when people feel that discomfort without support, they go quiet.

• Make failure expected, not exceptional — "We're learning. Every mistake is information." That one sentence, said consistently by a manager, changes the whole dynamic.
• Let people practice before they perform — low-stakes tasks first. No one should figure out AI for the first time on a high-visibility deliverable.
• Have leaders go first — if the manager uses AI imperfectly and says "here's what I tried and what I'd do differently," everyone relaxes.
• Separate the tool from the person — "the AI output wasn't good enough" is not the same as "you weren't good enough." That distinction matters more than most leaders realize.

Simplify your process first. Then add AI.

AI is a multiplier. Multipliers work best when the thing they're multiplying is already solid. The most common implementation mistake is automating a broken process. The result is the same mess, moving faster and at higher volume.

Move fast or get it right? You need both.

The goal is sequencing. Speed and control aren't opposites — they're two tools for different moments. Speed without control creates chaos. Control without speed creates stagnation. Strong execution is knowing when to apply each.

Scaling from pilot

The move from pilot to scale isn't a bigger pilot. It's transfer, standardization, and deliberate enablement. Scaling is where the business value is.

• A playbook — what was done, how it was done, what good looks like, and what to watch out for
• Documented knowledge — capture what worked, how it worked, and what you learned. Written knowledge travels. Tribal knowledge doesn't.
• A clear definition of success — agree on what "working" looks like so you know exactly what to replicate, improve, and build on
• Active sponsorship — scaling needs visible leadership support to cross team boundaries and survive competing priorities

Measure AI progress the right way

Activity metrics — licenses activated, tools deployed, training hours completed — tell you what happened, not whether it mattered. You need both. The second one is where the proof lives.

• Pick the right dimension — value shows up as time saved, quality improved, cost reduced, or revenue enabled. Pick the one that matters most for your use case.
• Establish a baseline before you launch — you can't prove improvement without knowing where you started. One week of baseline measurement saves months of arguments later.
• Set a specific target — not "improve efficiency." Instead: "Reduce first-draft time for weekly reports from 3 hours to 45 minutes."
• Review at 30 and 90 days — not to prove success, but to learn. What moved? What didn't? What does that tell you about where to focus next?
• Diagnose the gaps — low adoption usually means one of three things: people don't know how (skill), don't want to (motivation), or can't (workflow). Each needs a different fix.

Build internal capability

The goal of any AI program should be to become progressively independent. Sustainable adoption requires the organization to stand on its own. The rule is simple: external help for specialized problems. Internal ownership for everything else.

Keep critical thinking in the loop

When AI works well, something subtle happens. People stop questioning outputs. They accept faster. They think less. Ironically, when AI is slightly unreliable, people stay more engaged — they check, they challenge, they apply judgment. Outcomes are often better. That's worth paying attention to.

• Keep humans accountable for decisions — AI can inform, draft, and analyze. The decision belongs to a person.
• Design review points into workflows — not as friction, but as deliberate moments where human judgment is applied before things move forward
• Encourage critical thinking, not just speed — the question is not only "did AI produce this faster?" but "is this actually better?"

Stay intentional about access

The instinct to roll AI out to everyone at once is understandable. But without intentional design, broad access creates inconsistent quality, security gaps, and use cases that drift far from what actually matters. The real question is how much freedom should people have — and when.

• Start focused — clear use cases, approved tools, defined boundaries
• Build capability first — skills, habits, and judgment before full autonomy
• Expand access as trust grows — more freedom earned through demonstrated capability and responsible use