Artificial intelligence: definition, types and framing
Definition
Artificial intelligence describes methods by which software takes on tasks that would require perception, language or judgement in a person. The term names no product and no fixed method; it spans a range from rule-based systems to learning models. What matters is not the word but the question of which task a system solves, on which data and with what provable quality.
Why the Term Matters
Because very different things are sold under the same label, from simple automation to a costly learning model. For an investment decision that is risky. Neutral consulting translates the buzzword into three checkable questions: which task exactly, which data situation, which error tolerance. Who breaks the term down this way does not buy a promise but assesses a concrete use case before the first euro.
When It Is Relevant
- Recurring tasks follow patterns that are hard to capture in rigid rules.
- Enough documented examples exist for a system to learn or infer from.
- A result with residual uncertainty is acceptable as long as it stays checkable.
- The expected value can be tied to a metric, not just to a gut feeling.
When Not
When a task is clear and stable, a fixed rule solves it more cheaply and reliably than any learning model. For high-risk decisions without a control point, you first need a responsible owner and clear review steps before a model assists at all. Without a data basis the term stays an empty shell.
Industries That Use It
- Logistics and freight: estimate routes and arrival times without handing dispatch over entirely.
- Insurance: pre-sort incoming claims so adjusters keep time for the tricky cases.
- Energy utilities: forecast consumption and grid load to plan procurement and maintenance.
- Agriculture and agritech: analyse field and crop imagery to intervene early.
Typical Use Cases
- An assistant that answers questions from released internal documents.
- A forecast that surfaces demand, failures or bottlenecks early.
- A classification that sorts and routes documents or requests.
- An extraction that moves fields from receipts or forms into systems.
- A recommendation that gives case handlers context rather than finished verdicts.
Decision Guide
| Signal | Term fits | Alternative |
|---|---|---|
| Task with patterns, many examples | assess an AI approach | fixed rule where logic is clear |
| Clear, stable rule | rather not | formula or standard software |
| High risk, no control | de-risk first | process with a human in the decision |
| Value unclear, no metric | frame it first | readiness check before a project |
Example
An energy utility wants to test AI for load forecasting. In the potential analysis we separate the task from the buzzword: which data exists, how accurate the forecast must be, where the human steps in. Only then is it clear whether a learning model or simpler statistics fit. This describes the method, not an asserted client metric.
Service Connection
The honest entry is the AI readiness check or the AI potential analysis. To assess use cases in a structured way, start with AI consulting for the mid-market.
Sources
This framing describes common practice in AI consulting. Statements on legal duties are general orientation and not legal advice; they must be checked case by case against the currently applicable legal texts.