Your Company Is Already Using AI. Now What?

Nobody really rolled out AI. It just… arrived.

Someone in an organization gets a ChatGPT subscription. Someone else starts using Copilot. A third person figures out a surprisingly good way to prepare for client meetings. Marketing has its own tools, and sales has others. A few people are miles ahead, while others barely use AI at all.

Eventually, the company is “using AI.”

Then someone asks a few basic questions, and things start to get murky. What are we paying for? What company information is being typed into these tools? Which ones are connected to our systems? Which uses actually save time? And if the person who built this brilliant workflow leaves tomorrow, does the company still have it?

This is the awkward middle of AI adoption: AI is already inside the business, but the business hasn’t really implemented it yet. What started as experimentation has quietly become part of how the company works, and nobody designed it this way.

Employees Are Adopting AI Faster Than Companies Are Organizing It

Ask businesses about AI and you get one picture. Ask employees and you get another.

In its Business Trends and Outlook Survey, the U.S. Census Bureau found that 19.8% of businesses used AI in a business function.

The same month, Gallup found that 52% of U.S. employees had used AI in their role at least a few times a year, up from 27% two years earlier. Only 30% used it a few times a week or more. And just 25% said their organization had communicated a clear plan for integrating it.

These surveys ask different questions, so the numbers shouldn’t be compared directly. But together they describe something most leaders will recognize: for many companies, experimentation is the AI strategy.

This isn’t a bad thing. Experimentation is how people discover where AI is genuinely useful. The trouble starts when it never ends. A few people get very good at it, and others never start. Useful workflows stay locked in personal chat histories. Departments pay for overlapping subscriptions. Nobody is quite sure what data can go where.

And it becomes hard to answer a simple question: what is all this experimentation actually changing?

Your Best AI Work Lives in Someone Else’s Account

That question is hard to answer for a simple reason: most of the experimentation is happening where the company can’t see it. When AI arrives one person at a time, it usually arrives in personal accounts, and that’s where most of the awkward questions come from.

The bill nobody can explain: AI spending rarely shows up as one line item. It’s a ChatGPT plan here, a Copilot add-on there, a note-taker someone expensed in March. Each charge is small enough to approve without a second look. Together, they add up to a number nobody has seen. Try this: search your company card statement and expense claims for the big AI names (OpenAI or ChatGPT, Anthropic or Claude, Microsoft or Copilot) and count the charges.

The information nobody is tracking: In a personal account, every employee makes their own call about what’s safe to paste in: a client contract, a salary spreadsheet, a customer’s financial details. Many tools also connect to email and shared drives in a couple of clicks, with nobody approving it. For any business that holds client information, “we think people are being careful” isn’t an answer you want to give a client.

The work the company doesn’t keep: Chat history, saved prompts, and uploaded documents live with the person, not the business. When they leave, all of it leaves too.

The fix isn’t glamorous. One business plan the company owns, with everyone signing in through their work email. Then a few rules people can actually remember: which account to use, what never goes in, and which connections need approval.

It’s the difference between people using AI and the company owning it.

Getting People to Try AI Was the Easy Part

For many companies, convincing people to try AI is no longer the hard part.With the foundations in place, the harder questions are about the work itself:

Which uses are actually valuable? Which should become standard? What should AI be allowed to access? Where should a human stay in control? And how do you turn one person’s clever workaround into something the company keeps?

That last question matters more than it sounds.


Picture someone on your sales team who uses AI before every client meeting. They search their inbox for the latest thread, find the proposal, reread their notes, check the CRM, and ask AI to pull it all into a one-page briefing.

It’s genuinely useful. It’s also a personal habit, built on their own login, their own prompts, and their own judgment about what’s safe to paste in.

Now picture the same workflow set up properly. It connects to the company’s approved CRM and email, with permissions that let it read account history but never send anything on anyone’s behalf. Anyone on the sales team can run it. The salesperson still decides what to say in the meeting; AI simply handles the legwork beforehand. And when that employee leaves, the workflow stays.

This is the shift worth making: from people using AI to the business knowing how it uses AI.

How to Pick Your First AI Workflow

Many AI initiatives stall because they start with the question, “What can we do with AI?” It’s an enormous question, and rarely a useful one.

A better starting point is far less exciting: what does your team do every week that takes too long? Preparing for meetings. Following up afterward. Hunting for information across email, documents, and other systems. Compiling a daily briefing. Processing an intake form. Drafting the first version of a document.

You don’t need outside help to choose. The best first candidates usually pass five tests:

• It happens often: A weekly or daily task pays back faster than a quarterly one.
• It’s mostly gathering and assembling: Searching, summarizing, compiling, and first-drafting are where AI is strongest today.
• A person still makes the final call: Start where AI prepares the work and someone reviews it, not where AI acts on its own.
• The information already lives in systems you control: Email, shared drives, and your CRM are easier and safer to connect than scattered personal files.
• You can time it: If you can’t say how long it takes now, you’ll never prove it improved.

That last test is the one most companies skip. Before changing anything, ask three to five people who do the task to time it for a week or two. A simple tally is enough.

Without a baseline, “AI made it faster” is just a feeling. With one, you get a number you can defend. If a task takes 45 minutes today and 15 a month from now, you can multiply the difference across a team and a year, and decide what to do next with real evidence.

And small differences add up. Give five employees two hours back each week, and over 48 working weeks that’s 480 hours returned to the business every year. That’s capacity for more client work, faster projects, and the kind of thinking only people can do.

AI Jumpstart by Troon: From Experimenting to Implementing in Two Weeks

If you’d rather not run the process alone, this is what AI Jumpstart is built for.It’s a focused two-week engagement for companies already experimenting with AI that want to turn it into something deliberate.

It works with ChatGPT, Claude, or Copilot. If you already have a business plan, we start there. If your team is on personal accounts, we help you choose the right business plan and you put it in place before we begin. Then we connect it to four of your business tools, such as Outlook, SharePoint, or HubSpot.

Here’s how it runs:

1. Map how work actually happens:
   We look at the tools your people use, where their time goes, and which repetitive, information-heavy tasks are worth improving.
2. Set up the company account properly:
   One business plan, owned by you, with a named owner and a backup admin. Your team signs in with work email, personal accounts stop being where company work happens, and the data protections are switched on and tested.
3. Choose two workflows and measure them:
   We apply the five tests above and record a baseline before anything changes.
4. Build them with boundaries:
   Each workflow is connected to your systems at approved permission levels, with clear rules on what AI can access and where a person must approve the result. Read-only access is the default.
5. Write the rules for everyone, not just the workflows:
   A short AI policy your whole team can follow: company work happens in the company account, some things never go in (passwords, banking details, sensitive personal or medical information), and new connections need approval first. Your leadership signs it.
6. Put them to work:
   A small group from across the business uses the workflows on their real responsibilities and helps establish what works. That’s usually 3 to 10 people, one per department, chosen because colleagues already go to them for help. We call them champions.
7. Hand everything over:
   The workflows, configuration record, signed governance policy, runbooks, and admin guide all belong to you. So does a written cost briefing: what we found you were paying for, what the new setup costs, and what to watch as usage grows.
8. Come back and measure again:
   At 30 days we re-time the same tasks. At 60 and 90 days we look at what stuck, what didn’t, and what opportunities have emerged.

The goal isn’t to give AI as much control as possible. It’s to give it enough access to remove unnecessary work, without removing the judgment the work requires.

What it asks of your team: Two 90-minute discovery sessions, under an hour from your IT contact for setup, two hours with each champion, and one executive session to walk through the setup, the governance policy, the permissions map, and the costs. Every date is in the calendar before we begin.

AI Doesn’t Need a Bigger Mandate. It Needs Somewhere Useful to Start

There will always be another model, another feature, and another tool worth exploring.

The companies that get the most from AI won’t necessarily be the ones experimenting with the most tools. They’ll be the ones that get good at spotting where AI belongs, putting it against real work, measuring whether it helped, and making what works part of how the business runs.

It’s a much less dramatic version of AI transformation. It may also be a far more useful one. Because the real measure of an AI initiative isn’t how much AI you deployed. It’s whether your team can point to real work and say: That used to take me an hour. Now it doesn’t.