Anonymization
Anonymization is the irreversible transformation of personal data so that no individual can be identified by any means reasonably likely to be used, which takes the resulting data outside data protection law altogether.Recital 26 of the GDPR says the regulation does not apply to anonymous information, defined by the test of reasonable likelihood: taking into account the costs, the time, the available technology and future developments, can anyone single out, link or infer the identity of a person? If yes, the data is at best pseudonymized and still regulated. The Article 29 Working Party's Opinion 05/2014 on anonymization techniques reviewed randomization (noise addition, permutation, differential privacy) and generalization (aggregation, k-anonymity, l-diversity) and concluded that none is perfect on its own; the robustness has to be assessed against singling out, linkability and inference.
In analytics the word is used loosely. Truncating the last octet of an IP address, as the CNIL's audience measurement exemption requires, reduces identifiability but does not anonymize a visitor who still carries a cookie identifier. Aggregated counts (pages per day, countries, referrers) with no per-visitor record are anonymous; a per-session table with timestamps and a device profile usually is not. Google Analytics' "IP anonymization" was pseudonymization, and the setting disappeared in GA4 because IPs are no longer stored, which is also not the same as anonymous analytics.
The practical value is that properly anonymous statistics need neither consent nor a banner, which is the route privacy-focused analytics tools take when they store no identifier at all. Whether a given tool achieves it is a factual question for each configuration; see analytics without consent.