Your best people are funding AI out of their own pockets. Your AI budget is still under review.

If someone on your team pays personally for an AI tool, ask which task it helps and what company information goes through the account.

The company had not offered anything. And they needed something that worked, and twenty dollars a month felt like nothing compared to what it was giving them.

A signal of demand, not a corporate business case

The purchase shows perceived individual value. Verify output quality, time actually released, data exposure and the full cost of an approved company setup before extrapolating it.

They do not need to justify that to anyone. They just pay and keep going.

This is the thing that changes the conversation for every CEO still waiting for a business case.

The receipt is a useful starting signal. It does not prove net ROI, adoption by colleagues or safe corporate use.

For twenty years, major technology decisions at the company level went through a cycle. The CIO evaluated options. The CFO modeled costs and projected returns over three years. There was a vendor presentation, a scoring matrix, a sign-off process. Implementation followed approval. Adoption was measured against projections. Someone was responsible for showing it worked.

Procurement duration varies with risk, data, contract and integration scope. A bounded test can answer specific questions before a larger commitment.

Bring the individual experiment into view and test whether its benefit survives an approved setup.

The result is that most mid-sized companies right now have a split reality. At the leadership level, the discussion is should we invest in AI. At the individual level, AI is already embedded in how work gets done. Your best analyst is summarizing research in a fraction of the time. Your best sales rep is using it to prepare for client calls. Your ops lead is using it to draft process documentation that would have taken days.

They are not doing it because you gave them a roadmap. They are doing it because they figured it out themselves.

The next question is what evidence the business needs to approve the use.

The bottleneck is everything that comes after: how do you deploy this at scale, with your data, with proper access controls, with the compliance structure that keeps your legal team out of a difficult conversation? How do you take what your best people are doing individually and make it available to the rest of the team? How do you build on improvised individual solutions rather than letting that productivity leak out into personal subscriptions that the company can never audit or build on?

Those are harder questions. They are also the right questions.

Finance needs a measured baseline, total costs and a credible use for released capacity. The employee’s experience helps build that evidence.

Support learning across the team rather than treating the first users as a separate class. Give colleagues time to test the method on their own work.

There is also a less comfortable version of this problem.

The people working around the company to access these tools are not always doing it with your data and your clients' data in ways you would be comfortable with. The productivity is real. The exposure is also real. When you eventually move on AI, you are not starting from zero. You are cleaning up a set of informal practices that have been running for months or years, with data governance that nobody designed.

That cleanup is easier when you are the one initiating it.

Ask why the employee chose that route, then decide what a safe company alternative must provide.

And the more uncomfortable version of that: what else in the company are people working around because the institution is slower than the individual?

Treat that question as a diagnostic. The same dynamic appears wherever good people outpace slow institutions. They find another way, and the company eventually discovers the gap between its official story and daily reality.

The companies that are ahead on AI right now did not start with a better strategy. They started with the decision to stop waiting for certainty and start building the infrastructure that turns individual productivity into institutional capability.

For the data and permission review, use the shadow-AI checklist. For transfer to colleagues, use the internal-champions method.