Originally published on LinkedIn on August 18, 2025. Revised in September 2026 to explain a few terms along the way.
Buying an AI tool is easy. Governing it — deciding the rules for how it can be used safely — and making sure it's actually safe is the hard part.
This pattern has been seen again and again: a company spends six figures on a tool like Microsoft Copilot or ChatGPT Enterprise (two well-known AI assistant products used inside a company's existing software), and only days later realizes company data has been flowing out through gaps nobody knew existed.
Forty-five settings later, the gaps still haven't been closed, and now there are five hundred employees using the tool with no clear ruleset for how to use it. If that sounds familiar, this is for you.
Before chasing the next shiny new tool, a few questions are worth asking first
Where are the gaps in how company data is governed?
Who is actually reviewing what employees are asking the AI to do?
Do employees know what's safe to type into an AI tool, and what could get them fired?
If those can't be answered in thirty seconds, the company may not be ready yet.
Start small. Find the one process that eats up the most of the team's time, build AI into that one process first, and then measure the results. Even a ten-percent improvement on a repetitive task adds up quickly once it's happening every day.
A common worry is that top performers will leave once AI shows up. I'd flip that around: AI isn't here to replace human judgment or the ability to build relationships. It's here to take over the repetitive work that causes burnout. Train people in how to use AI well, and they become far more effective, not replaceable.
A useful exercise: list three low-value chores the team dislikes. Then ask, “What if AI handled 80% of that, and we kept the other 20%?” That question tends to free up real hours for the creative, revenue-driving work that actually matters.
One more thing to watch for: the trap of a tool like Copilot. Thirty dollars per employee per month sounds cheap, until it's five hundred employees and the company is locked into that spend indefinitely. Before signing anything, it's worth testing governance first — can it be proven who has access to what? — and comparing the full cost: license fees, integration, training, and compliance, not just the sticker price.
Gartner, a research firm that studies technology trends, found that only 3% of Copilot rollouts succeed without policies already in place before the rollout starts. Once an AI tool is already live across a company, adding controls after the fact is far harder than building them in from the start.
What to do today
-
Run a privacy-first audit on one AI pilot project — map every place company data flows through it.
-
Pick three quick wins to start with: contract summaries, document search, or sorting incoming customer questions.
-
Write a one-page “AI rules” guide so everyone understands what's allowed and what isn't before any tool goes live.
Doing that puts a company well ahead of everyone else still treating AI like just another app.
We hold a monthly in-person event called the AI Breakfast, where local business people swap stories about what's working. Register at https://www.eventbrite.com/o/akron-innovations-111185028001 — or if you can't make it, email me at rob@aicatalystinstitute.com and we can talk AI one-on-one.