Walk into most mid-market companies today and you'll find AI everywhere — and nowhere at once.
Someone is using Copilot to draft emails. Someone else is running prompts through a personal ChatGPT account. A third person built a workflow in a tool IT has never heard of. Nobody is tracking what any of it costs, or whether it's actually working.
That pattern has a name: Shadow AI. Real usage, no shared record, no owner. And it is not a fringe behavior. In a 2026 survey of 1,250 office professionals at companies with $500 million or more in revenue, 66% had used AI tools at work that they believed were not allowed; at companies with 1,500 or more employees, 72%. A third (34%) had entered customer data into public AI tools.1 IT sees the same thing from the other side: 70% of the 2,000 CIOs and CTOs IBM surveyed this year say teams are deploying technology faster than IT can track it.2
The pattern isn't a technology problem
The cost side is just as hard to see. Zylo's 2026 index, built on $75 billion of software spend, found expensed software spend up 267% in a year, with ChatGPT now the most expensed application, and 78% of IT leaders reporting unexpected charges tied to AI features or usage-based pricing.3 85% of technology executives say they lack full visibility into real-time AI spend, and in Gartner's 2026 survey of 1,303 leaders, about 11% did not know what their own function spent on AI last year.24
Returns are thin where nobody owns the work. McKinsey's 2026 survey found that nearly nine in ten organizations use AI regularly, but only 37% attribute any EBIT impact to it.5 The reason is usually ordinary: nobody assigned a workflow, a budget, or a way to measure results. The tools tend to work; what is missing is someone accountable for the rollout.
The companies that do see results treat AI as a change to how work gets done. In McKinsey's data, nearly three-quarters of AI high performers have fundamentally redesigned workflows, compared with about a quarter of other companies.5
For mid-market manufacturing and logistics companies in the South US, this shows up in a specific way. Most teams are already paying for Microsoft 365 E3 or E5. Copilot Chat is part of those plans, and Microsoft is adding more Copilot capability to them through the 2026 packaging changes.7 People pick it up on their own, get some value from it, and the organization ends up with a dozen small, disconnected experiments instead of one plan.
Three questions worth asking before you spend another dollar
Before adding another AI tool, subscription, or pilot, it's worth answering three questions honestly:
- What are we already paying for? Most Microsoft 365 tenants include AI capability that's licensed and unused. Before buying anything new, confirm what's already inside the contract.
- Who owns this, end to end? Not who requested it — who is accountable for the workflow working, being measured, and being secured. If the answer is "everyone," the honest answer is "no one."
- How will we know if it worked? A pilot without a defined outcome doesn't fail loudly. It quietly stops getting used, and the license renews anyway.
Governance from day one, not after something goes wrong
The teams that get past the pilot stage get one thing right early: they treat security and governance as part of the build, not a step that happens afterward. Copilot reads everything a user already has access to. Deciding what that should be is a short conversation before rollout — or an incident report after.
The risk is no longer theoretical. Among large US companies, EY found this year that 47% have skipped their own AI governance process for "urgent" deployments, 26% of those using AI agents cannot detect unauthorized agents running internally, and 36% have already had an AI incident with a materially negative impact.6 The practical starting point is simple: a company-wide policy on which AI tools are approved, a regular check for the ones that aren't, and AI included in how you already review software and data access.
This is the first article in our Switch It On series, where we walk through what it actually looks like to move from asking AI questions to putting AI to work inside a Microsoft 365 tenant you already own.
Next up: what other companies are running into with AI pilots, tools and costs in 2026, and what the ones that scale do differently.
Want to know what this looks like in your company?
Tell us who you are and what you'd like: more information about the Switch It On program, or a meeting with the Bravent team to look at your Microsoft 365 tenant together. Someone from Bravent will get back to you.
Too many AI pilots, too few in production: what companies are facing in 2026 →
What the latest 2026 studies say about the tools, costs and workload piling up, and what the companies that scale do differently.