AI solutions for business: the four categories that actually exist, and how to choose

The market for AI solutions for business looks endless, but it boils down to four categories. Assistants answer questions and draft text. Agents carry out tasks inside your software. Workflow automation chains predictable steps. AI on internal data answers by citing your own documents. Most serious projects combine two or three of these building blocks.

Then comes an architecture decision: rent a cloud service, often American, or deploy sovereign AI that runs on your premises or at a hosting provider in your region. That choice matters as much as the tool itself.

This is the best-known category: ChatGPT Enterprise, Microsoft Copilot, Claude, Gemini. An employee asks a question. The assistant answers, summarizes, translates or drafts. The gain is real for writing and research, and deployment takes days.

The limit is what the assistant knows about your business: only what the employee feeds it or what the connectors expose. It does not take initiative and does not write to your ERP. Pricing is generally per user, per month. Our honest comparison explains when that subscription is enough.

An agent receives a goal rather than a question. It reads an invoice that arrives by email, matches it to the right project, prepares the accounting entry and submits it for a person’s approval. In other words, it acts inside your systems, within limits set by management.

This is the category that frees up the most hours, because it takes on whole tasks rather than fragments. It is also the one that demands the most discipline: precise access rights, an action log and approval thresholds. Our article What is an AI agent? explains how they work.

Businesses were automating long before generative AI: ERP rules, software robots (RPA), tools that connect applications. These solutions shine when every step is predictable: if this field says X, do Y.

Their weakness shows up with a document that does not follow the template, like an invoice from a new supplier or an email phrased differently. AI fills exactly that gap by reading unstructured content. According to Statistics Canada, robotic process automation was used by only 5.0% of Canadian businesses using AI in 2026, against 34.5% for text analytics. For a list of candidate processes, see our 12 admin tasks to automate first.

This category, often called RAG (retrieval-augmented generation), indexes your procedures, contracts, quotes and files. An employee asks a question. The system finds the relevant passages and answers with citations. A new hire finds the procedure without interrupting three colleagues. A project manager pulls up the quote for a similar job in seconds.

It is often the foundation for the other categories: an agent preparing a quote needs to find past prices and projects. Quality depends directly on the state of your documents. Our AI glossary for executives defines the terms worth knowing.

The table below sums up what each category does well, and what it requires.

Each category comes in two forms. In the public cloud, you rent access to a model run by a provider, often American. The CLOUD Act requires such a provider to disclose data in its control whether it is stored in the United States or elsewhere (18 U.S.C. § 2713). A server in Montreal does not change its operator’s jurisdiction.

Sovereign AI runs on a server on your premises or at a hosting provider in your region, with an open model you keep a copy of. Under Law 25, section 17 requires a privacy impact assessment before personal information is communicated outside Quebec. When the data stays in Quebec, that assessment gets simpler. Our Law 25 compliance support and our sovereign AI vs public cloud business case go further.

Three questions usually rule out half the options.

At Les Industries Romy in Granby, all four categories live side by side on a server inside the company. AI on internal data finds past projects, agents prepare quotes and enter invoices, and every commitment goes through a person. According to management’s estimate, accounting went from 12 hours to one hour a week on the targeted tasks. The details are in the Romy case study.

To see what these solutions look like in your industry, visit our pages for manufacturers, law firms, healthcare and real estate, or read about our approach. Available programs are on our Funding page.