ELMA365: Without Corporate AI, Employees Turn to Public Chatbots

Shadow AI works as a workplace norm. If a company has no corporate neural network, employees turn to public ones on their own, and a formal ban has little effect. That is the claim made by Olga Mikheeva, Product Owner of ELMA365 CX (formerly ELMA365 CRM), in an interview with TAdviser. The available announcement contains no statistics and names no specific tools, so what follows is an analysis of the claim itself and the context around it.

What exactly is being claimed

The position is stated plainly. When a company has no corporate AI in place, many employees use public neural networks for work tasks, whether or not they have permission. The interview's headline adds a second part: companies do not control this process.

It is worth considering who is speaking. ELMA365 develops its own platform for business processes and customer-facing scenarios, so the vendor has an interest in selling built-in AI. That does not invalidate the observation: CIOs and information security specialists regularly reach the same conclusion. But the published excerpt contains no figures that would measure the scale.

Where shadow AI comes from

The mechanics are simple. An employee needs to summarize a long email thread, draft a reply to a client, put together a commercial proposal, or parse a spreadsheet. A public chatbot opens in the next tab within seconds. Getting a corporate tool approved takes months. The time savings are visible immediately, while the risk stays abstract.

For CRM scenarios, the problem is sharper than usual. Client names, deal terms, prices, and contract excerpts easily end up in a query to an external model. Managers do not find out: there are no logs, and the question is often asked from a personal account.

Key elements of the story:

• Source: TAdviser interview, October 13, 2026.
• Speaker: Olga Mikheeva, Product Owner of ELMA365 CX.
• Claim: when there is no corporate AI, public neural networks get used regardless of permissions.
• Consequence: the company's control over data and processes is weakened.
• Missing from the announcement: statistics, a list of services, data on incidents.

What companies should do

A ban without an alternative generally pushes AI use into the shadows rather than eliminating it. A more practical approach works in the reverse order.

First, find out what is actually happening. An anonymous employee survey and an analysis of traffic to popular AI services will show which tasks people are already handing to neural networks. That gives you a ready-made list of priorities for rollout.

Next, divide data into sensitivity levels. A draft of a post can be discussed with a public model. The customer base, contracts, and financial data cannot. The rules should fit on one page, or nobody will read them.

Finally, provide a legitimate tool where employees already do their work: in the CRM, email, and document management. The fewer steps it takes to reach corporate AI, the weaker the temptation to open a public one. Here the vendors' argument for built-in assistants holds up, but it should be tested on your own tasks: answer quality, where data is processed, access rights, and logging.

The full TAdviser interview probably contains more detail. But even the claim from the announcement gives IT leaders a clear task: before writing a policy, find out what employees actually use.