It is 4:30 p.m. and the quote has to go out today. A salesperson copies the customer’s draft contract into a free chatbot and asks for a summary. The answer comes back in seconds, the work gets finished on time, and nobody does anything wrong.

The question is where that contract text is now. This is called shadow AI, and almost every company has it.

Shadow AI: good news in a poor disguise

Shadow AI means employees using AI tools on their own accounts, in their own ways, without shared ground rules. One person drafts customer letters with a free chatbot, another runs contract text through a translator, a third generates images for a presentation, each in their own bubble.

In a way this is good news: people have already worked out that AI helps them. Nobody has to be pushed toward a new tool. The bad news is the flip side: company information ends up on personal accounts, the quality of the output varies with how well each person writes prompts, and nobody sees the whole picture. When an employee changes jobs, their chat histories and the working habits they built up leave with them. At the moment nobody can even estimate how many trade secrets move from one company to another along with employees’ AI accounts.

What managed, enterprise-level AI means

Enterprise-level AI does not mean a more expensive license. It means three things:

• Control: company-managed accounts, agreed tools and clear rules about what information may be given to AI. Company data stays within the company’s contracts, not out in the world or inside other companies as training material.
• Context: AI is connected to the company’s own information: instructions, process descriptions, price lists, earlier work. Then it answers like an in-house expert rather than a consultant off the street.
• Ways of working and values: AI is taught how things are done at your company: how customers are addressed, what is promised and what is not, in what order things get approved. A well-briefed AI also knows to say “you should check this with a person” or “that is not how your company does things”.

The difference is the same as with a new employee in their first month and in their sixth: the same skills, but the one who has been shown around knows how the place works.

One principle, three different tools

These three points apply no matter which AI is in use. In a small or mid-sized company there are three forms of it in practice.

Conversational AI: a briefed assistant

Chat is the most familiar form of AI, but in business use its value multiplies when it answers on the basis of the company’s own documents. A new employee asks “how do we handle complaints here?” and gets an answer that follows your quality guidelines, not an internet average. The same assistant advises, walks people through systems and reminds them of company practice.

Agents: AI that acts, within agreed limits

An agent does not just answer; it carries out tasks. It retrieves information, fills in a form, drafts a document and moves the matter along. At the enterprise level what matters is that the agent works the way your process works: it has only the permissions it needs, it asks a person for approval at agreed points, and what it does is logged. Then automation is a managed part of how you operate rather than a black box.

Images and materials: on brand

Generating images and materials is the fastest way to see the difference between unmanaged and managed use. Without direction, everyone produces something that looks different. When AI is given the company’s style guide, meaning colors, tone and what may and may not appear in images, marketing materials, illustrations and presentations stay consistent no matter who makes them.

Two examples from everyday work

We recently helped an accounting firm where shadow AI was already routine: the accountants were using free chatbots on their own accounts. We built a shared, managed environment in its place and briefed the assistant with the firm’s quality manual and its most typical client situations. Now a new accountant gets answers that follow the firm’s own practice, and client data does not travel to personal accounts. According to the firm, questions to experienced colleagues dropped to a fraction of what they had been.

In another project, a construction company wanted to speed up how it handled requests for quotes. An agent reads an incoming request, picks out quantities and requirements, checks prices against the company’s own price list and puts together a draft quote following the company’s own calculation model. A person reviews and approves every quote before it goes out. According to the customer, processing time dropped from days to hours, and the calculation method stays the same regardless of who does the work.

How to get started: five steps

1. Map the current situation. Ask openly what tools people already use and for what. The answers are almost always a surprise, and they also show where AI does the most good.
2. Agree on ground rules. A short, clear guideline: what information may be given to AI, what may not, and how the output is used. One page is enough to start with.
3. Move to company accounts. Company-managed accounts cost little compared with keeping the data and the usage under the company’s control.
4. Bring in the company’s own knowledge. Instructions, processes, price lists, style guide: the better AI knows the place, the better it serves.
5. Start with one process. Pick one recurring task, measure the effect and expand from there. One big bang fails more often than a series of small wins.

Back to that draft contract sent at 4:30 p.m. In a managed environment it would have gone to a company account and stayed within the company’s contracts, and the assistant would have known what could be passed on from it. Same work, same speed, and a record of where the text went.

Shadow AI tells you the tools are genuinely useful. The company’s job is to give them a frame: control, context and the company’s own way of working. Making use of AI is part of our IT department service, and anyone can take these five steps.