Eighty Three Percent

Stripe launched an internal AI platform called Kai in April. It reached most of the company within two weeks. As of last week, 83 percent of employees use it every week, including nearly all of the go to market staff, and it connects to more than a thousand internal tools and skills.

I've watched a lot of companies roll out AI tools. Here's how it usually goes.

The owner gets excited. Buys seats for everyone. Sends a Slack message with a lot of exclamation points. Maybe runs a lunch and learn. For two weeks, people poke at it. A few power users emerge. Then usage drifts down, and by month three the tool is something two people use daily and everyone else remembers exists when the renewal invoice shows up.

Eighty three percent weekly use, months after launch, is not normal. It's worth understanding why it happened, because the gap between the companies getting real value from AI and everyone else is increasingly not about which tools they bought. It's about whether anyone actually uses them.

I don't have an inside view of how Stripe ran its rollout, and I'm not going to pretend I do. But the numbers they published point at a few things, and they line up with what I've seen work in much smaller companies. Let's go through them.

Start with the failure, because it's so consistent.

The tool doesn't know anything about the business. A general chat assistant is impressive in a demo and mediocre on a Tuesday, because it doesn't know your clients, your products, your prices, your process or your history. So every task starts with five minutes of explaining context, and after the fifth time, people stop bothering.

Nobody changed the work. The tool got added on top of the existing process instead of being built into it. Using it is an extra step, optional, and optional extra steps are the first thing to go when people get busy.

The wins stayed private. The two people who figured out something great kept it to themselves, not out of selfishness, but because there was no place to share it and no habit of sharing.

Nobody measured it. Usage fell and nobody noticed until renewal time.

Now look at what Stripe reported. The platform connects to more than a thousand internal tools and skills. That's the detail I keep coming back to.

That's the opposite of the first failure. A tool connected to a thousand internal systems knows things. It can look up the customer, pull the report, find the policy. It starts every task with context instead of a blank page. That's the difference between an assistant that's impressive and one that's useful.

The second detail: nearly all go to market staff use it. Sales and customer facing teams are usually the hardest group to get onto any new tool, because their day is packed and anything that slows them down gets dropped. Near total adoption there suggests the tool made their day faster, not slower. It saved them time on things they already had to do.

You don't need a thousand integrations. You need the handful that matter for your business. The principle scales down fine.

Here's the playbook for a team of five, fifteen or fifty. It's built around fixing the four failures above, in order.

Fix one: give it the context. Before you roll anything out, build the knowledge base. Your service descriptions, your pricing, your processes, your client list with the basics, your brand voice, your answers to the twenty questions customers ask most. Put it somewhere the AI tool can reach, whether that's a shared project with uploaded documents, a connected drive folder, or a connected Notion workspace.

This is the step everyone skips, and it's the one that determines whether the tool is useful. An assistant that already knows your business turns a five minute task into a thirty second one. An assistant that doesn't turns it into a five minute task plus explaining.

Budget a full day for this. It's the highest return day of the whole rollout.

Fix two: connect it to where work happens. List the five systems your team touches most. Email, calendar, CRM, project management, file storage. Connect as many as your tool supports, with sensible permissions. If your team lives in a platform like Go High Level, that's where the AI features need to show up, not in a separate tab nobody opens. Where there isn't a direct connection, Make.com can usually bridge the gap.

Every connection removes a copy and paste step, and copy and paste is where adoption goes to die.

Fix three: build it into one process per team. Don't say "use AI more." Pick one specific, repeated task for each team member and change the process so the AI step is the default way it gets done.

For sales, maybe it's the call recap. For support, the first draft of every reply. For operations, the weekly status summary. For the owner, the Monday planning review. One task each. Make it the standard way that task gets done, written into whatever you use for process documentation.

Once one task is working, people find the second and third on their own. That's the pattern. You can't mandate curiosity, but you can make the first win so easy that curiosity takes over.

Fix four: make the wins public. Start a channel or a shared doc called something like "AI wins." Every time someone saves time or does something they couldn't do before, they post it. The prompt, the result, the time saved. Two sentences is plenty.

Then read it out loud at your weekly meeting. Thirty seconds, one or two wins. This does more for adoption than any training session, because people learn best from a colleague doing a job like theirs, not from a vendor's demo.

FROM THE AI NEWSROOM

The exact systems behind everything in this issue. The audit sheets, the routing logic, the templates and the review cadences, built out step by step so you can copy them straight into your own stack. One time, forty seven dollars.

Get the Blueprint for $47

.....

You can't manage what you don't measure, and most people measure the wrong thing.

Logins aren't useful. Somebody can log in every day and get nothing done. Number of prompts isn't useful either, because a person fumbling through fifteen prompts for one task looks more "engaged" than someone who nailed it in two.

Measure two things.

Weekly active use by person. Not total logins. Did this person use the tool for real work at least once this week, yes or no. That's the number Stripe reported, and it's the right one. Track it weekly for every person on the team.

Time on the tasks you changed. For the one task per person you built AI into, how long does it take now compared to before? You need a before number, so measure it for a week before you roll anything out. If you want hard numbers instead of guesses, a time tracker like Rize will show you where the hours actually go before and after.

If weekly active use drops for a person, don't send a reminder. Have a five minute conversation. Usually there's a specific reason. The tool gave them a bad answer on something important and they lost trust. Or the task you picked for them isn't one they actually do much. Or they're doing it faster a different way. Fix the reason and usage comes back.

Here's the part owners don't love hearing.

If you don't use it, visibly and regularly, nobody else will for long. People take their cues from what the boss actually does, not what the boss says in a Slack message.

That means using it in meetings. "Let me pull that up," and then actually pulling it up with the AI tool, in front of people. Sharing your own wins in the wins channel, including the small ones. Admitting when it gave you something wrong and showing how you caught it, because that teaches the team that checking the output is normal and expected rather than a sign the tool failed.

It also means protecting the time. If you want your team to build new habits, give them a little room to do it. An hour a week, on the calendar, for experimenting with their one task and finding the next one. That's the cheapest investment in the whole plan and the one most owners cut first.

Here's why this matters more than which model you pick.

The price of AI capability is falling fast. Last week the two biggest labs cut flagship prices on the same day, and a crowd of cheaper models landed in the middle of the market. Access to good AI is becoming a commodity. Your competitors can buy the same tools you can, at the same prices, starting today.

What they can't buy is a team that actually uses those tools well, day in and day out. That takes months to build, and it compounds. A team at 80 percent weekly use that's been finding and sharing wins for a year is working in a genuinely different way than a team where two people use AI and everyone else doesn't.

That gap won't close by buying better software. It closes by doing the boring work: context, connections, one process per person, public wins, the right measurement, and an owner who goes first.

There's a reason so many operators are pulling this kind of adoption into their broader growth plan instead of treating it as a tech project. If you want structured help building the operating rhythm around it, communities like Scaling.com are built around exactly this kind of execution discipline.

You've got about three months left in the year. Here's how I'd use them.

October: context and connections. Build the knowledge base. Connect the five systems that matter. Measure your baseline: weekly active use today, and time spent on the tasks you plan to change.

November: one task per person. Pick the task with each person. Build it into the process. Start the wins channel. Read wins out loud every week.

December: measure and expand. Check weekly active use and time saved. For everyone above the line, pick a second task. For everyone below it, have the five minute conversation and fix the reason.

By January you'll know exactly where you stand, and you'll be planning next year with real numbers instead of a vague sense that "we should be doing more with AI."

FROM THE AI NEWSROOM

For operators who would rather build it with someone than build it alone. The full stack, installed alongside you, with the decisions made in the room instead of guessed at. Ninety seven dollars.

Join the Accelerator for $97

.....

Ask every person on your team one question: "What did you use AI for at work this week?"

Write down the answers. Count how many people had a real answer, not "I tried it once."

That's your weekly active use number. It's probably lower than you'd guess.

Then pick the one person with the best answer, and ask them to show the rest of the team at your next meeting. Five minutes. That's the first entry in your wins channel and the start of everything else.

Jordan

The AI Newsroom | Practical AI for people with a business to run.