Let me start with a number that should take some weight off your shoulders. Most AI projects fail. The estimates land somewhere around 70–80%, and study after study points to the same culprit: it's not the technology. It's the human part — the rollout, the training, the trust.
So if you bought the subscription, dropped the link in a team chat, and watched everyone quietly drift back to the old way, you didn't fail. You skipped a step nobody warns you about. The key isn't the tool itself, it's the workflow. The workflow is the system you build around the tool and that's what decides whether any of this sticks.
Good news, friend: that part is buildable. Let me show you how.
Why your team really routes around AI
It's tempting to read low adoption as resistance. People don't like change, they're set in their ways, maybe they'll come around eventually. The data tells a kinder and far more useful story. In a 2025 Gartner survey, the number one reason employees gave for not using AI wasn't fear and it wasn't stubbornness. It was that they couldn't see how it helped their actual work. They looked at the tool, looked at their Tuesday, and couldn't connect the two.
Stack a few more findings next to that and the picture sharpens. Nearly half of workers say they don't get enough support to use AI well. More than 40% say no one ever handed them a clear policy on what's allowed. And about a third have quietly started using AI tools nobody approved, because they needed help and went looking on their own. Sit with that last one. Your people aren't avoiding AI. Some of them are already using it in the shadows, with no guardrails, because the official path didn't exist.
Read it together and you get something hopeful: your team mostly wants the help. What's missing is a path. And a path is something you can build in an afternoon.
We're going to build one now, using a single example so it stays concrete: the weekly client recap, the one somebody on your team writes every Friday and quietly dreads.
The five steps
Step 1: Start with one real job, not "use AI"
"Start using AI" is not an instruction. It's a wish. It gives a person nothing to actually do on a Tuesday morning, which is exactly why that Gartner finding happens. So get specific to the point where it almost feels too small. Don't roll out AI. Roll out one task. Ours is the Friday client recap. That's the whole assignment this week. Not "explore AI for client communication." Just: use AI to draft the recap.
In week one, this looks like picking the one task your team would hand off in a heartbeat if they could, and naming it out loud. When the job is that concrete, the most common objection ("I don't see how this helps me") disappears, because the help is right there. You're not asking anyone to reimagine their role. You're taking one dreaded thing off their plate.
Step 2: Set the guardrails before you set anyone loose
Here's a fair question your team is probably too polite to ask: what am I actually allowed to do with this? When that answer is missing, the careful people freeze and the bold people improvise, and improvising is how a client's private details end up pasted into some random free tool at 4:55pm on a Friday.
You don't need a legal document. But what you do need is a page explaining what's allowable vs not when using AI (drafting the recap, summarizing notes, first passes). List explicitly what AI never touches (anything you wouldn't send unencrypted, client data, the final send button). And the rule that carries the most weight: a human reads it before it goes out. That's the page.
This is also where doing it right pays off in trust, which matters most. Confidence in workplace AI actually dropped in 2025 as people watched it get shoved into places it didn't belong. A short, clear policy builds the opposite of that. It tells your team you thought this through, and that their judgment is still the final word.
Step 3: Write it down
This is the step almost everyone skips, and skipping it is why "we tried AI" so often becomes "I'm the only one who knows how we do it now." A workflow that lives only in your head is not a system.
So write the simple version down. One page, no jargon. For the recap, that's: the task (Friday client recap), the tool, the exact prompt that produces a usable draft, what a good recap always includes (say, the three things it can't ship without), and who reviews it before it sends. In week one you're really just capturing what worked the first time you did it well. That single page is what turns a clever trick that lives with you into a repeatable process that lives with the team. It is also, not by accident, how you stop being the bottleneck in your own business.
Step 4: Train the team, don't just announce it
Dropping a link in a chat is not training. Telling people a tool exists is not the same as showing them how to use it, and the gap between those two is exactly where most rollouts die. The research here is clear and a little comforting: when a manager actively models AI use, adoption climbs; when they stay quiet, it stalls.
So do it live. Sit down together, pull up the one-pager, and write that Friday recap out loud in front of everyone, including the part where the first draft is mediocre and you fix it on the spot. Then let them try while you're still in the room to field the "wait, what about..." questions. You're not only teaching a tool. You're showing your team it's safe to be a beginner in front of you, and that permission is the thing that actually moves people.
Step 5: Make the new way the easy way
People take the path of least resistance. Every time. If the old recap template is still sitting in the shared drive and the new AI-assisted one takes three extra clicks to find, the old one wins by Friday afternoon.
Close the gap instead. Put the workflow where the work already happens. Archive the old template so it stops being the default. Make the AI-assisted version the obvious one: pinned at the top of the folder, linked right in the recurring calendar reminder. When the better way is also the easier way, you don't have to enforce adoption or nag anyone about it. It just becomes how everyone operates by default.
The mistakes that quietly kill it
Even with the five steps, a handful of avoidable missteps sink rollouts. Watch for these:
- Launching five workflows at once. Adoption buckles under its own weight. Ship one, let it stick, then add the next.
- Skipping the human review to "save time." One bad AI-written client email costs more trust than the workflow saves all month.
- Measuring activity instead of relief. The goal isn't "we used AI 40 times this week." It's "Fridays got lighter."
- Treating the one-pager as finished. Update it every time someone finds a better prompt. A living doc keeps adding value; a frozen one gets ignored.
- Rolling it out top-down with no input. The person doing the task every day knows where it actually hurts. Build it with them, not at them.
The reason this is worth doing right
It would be easy to read all of this as a recipe for squeezing more out of your people. It isn't. It never should be.
A good AI workflow doesn't exist to fill the freed-up hours with more work. It exists to give those hours back. For small teams, the realistic savings on this kind of work land around 15 hours a week, and I refuse to file that under "productivity." That's a Friday that ends on time. A dinner you're actually present for. Room for the people, and the purpose, you started this business to serve in the first place.
So build the workflow your team will follow. Then guard what it gives back. Because that's the whole point.
Two companion reads make this stick: before you build the workflow, it's worth knowing how many AI tools your business actually needs so you're not automating across a junk drawer — and once it's running, the same instinct turns your task list into a to-do list that's actually a system.
If you want help building systems like this so they actually hold, that's what we do at Sixteen36 Co. And if you just want to talk through where yours keep breaking, let's talk. We'll start by listening.
