A data lake is a repository that accepts data in any format — structured (relational tables), semi-structured (JSON, XML), or unstructured (images, videos, free text, logs) — without requiring prior modelling. The philosophy is 'store everything now, structure it when needed' (schema-on-read), as opposed to the traditional data warehouse where data must be structured before entering (schema-on-write).
The data lake shines when future analytical requirements are unknown, or when there is valuable data in formats the traditional data warehouse does not handle well — quality-control images, free text from customer chat, system logs. A poorly managed data lake becomes a 'data swamp' — a repository of data no one knows how to use.
In INFOS environments, data lakes make sense in applied-AI implementations that need multi-format data — for example, computer-vision quality control combining images with metadata. For most customers, a well-designed relational data warehouse (via MyBusiness-ITV on Qlik) solves 90% of the needs; the data lake comes in for specific scenarios.