Data tokenization substitutes a sensitive value with a token that has no usable meaning on its own. A payment card number, for example, becomes a string of the same length, and systems that only need to recognize the card can store the token instead. In vault-based tokenization, a token service stores the mapping between tokens and original values. Vaultless approaches derive tokens with keyed algorithms and keep no mapping table. Payments use tokenization widely because it reduces the number of systems that handle cardholder data under PCI DSS.
Tokenization differs from encryption. Anyone holding the key can decrypt encrypted data, while a token maps back to its original value only through the token service or algorithm. The word also has an unrelated meaning in language models, where tokenization splits text into units the model reads. For analytics and AI, format-preserving and consistent tokens let records keep their structure and still join across tables.