Unstructured data is information that does not come in a fixed, predictable format — the free text of an email, the audio of a call, a scanned letter, a photograph — so a computer system cannot simply read it into rows and columns the way it can a spreadsheet; it has to interpret the content first.
Statistics Canada’s own survey of AI-using businesses shows how much of what those businesses actually do with AI depends on exactly this kind of information. “Among businesses that reported using AI over the last 12 months (19.2%), the most commonly used applications were data analytics (36.6%), followed by text analytics (34.5%) and virtual agents or chat bots (28.2%)”. Text analytics, together with the chat and virtual-agent tools built on it, works almost entirely on unstructured data: the wording of an email, a support ticket, or a client’s own question, not a fixed field in a database.
Most of what a business actually generates day to day is unstructured by default: an email thread, a scanned consent form, a voicemail, a photograph of a document. None of it arrives pre-sorted into fields, which is exactly why the applications built on it — text analytics, virtual agents, chat bots — have to do interpretive work that a structured-data application like data analytics does not: deciding what a passage means before deciding what to do with it.
A brokerage’s inbox of client emails is unstructured data in its rawest form — no two messages are laid out the same way, and the same request can be phrased a dozen different ways. Before an AI tool can act on that inbox at all, something has to interpret the free text well enough to tell those dozen phrasings apart from an unrelated question, which is a fundamentally different task from filtering a spreadsheet column. A scanned consent form sitting in the same inbox is unstructured for a different reason again: the words might be typed in a fixed layout, but the file itself is an image until something reads the text out of it.
See also: what is structured data, what is a dataset, what is natural language processing.
Turning free text into something a system can act on reliably is a build decision — custom-ai-solutions covers what that interpretation step actually requires and where it tends to go wrong.
