Tag: Deletion

• 06 Jul, 2026
• • GDPR ,
  • Governance

Analytics data retention: what to keep, for how long, and why

“We retain analytics data for twenty-five months.” The statement sounds precise but does not say whether it covers raw events or reports, whether visitor identifiers expire at the same time, whether collector logs follow another period, whether backups are purged, whether CSV exports remain on laptops, whether new activity resets the clock, or whether truly anonymous statistics remain longer. A useful retention policy is not one number. It describes the lifecycle of each data layer and connects every period to a purpose. The CNIL states that personal data cannot be kept indefinitely and that controllers should determine retention from the collection purpose. It distinguishes active use, intermediate archiving and, in particular cases, permanent archiving. Web analytics teams usually need a well-limited active dataset, effective deletion and possibly sufficiently aggregated long-term trends. Begin with purpose, not available settings A platform may offer 2, 14, 25 or 50 months. Those options do not determine the right period. Ask:For how long does this individual-level data remain necessary for the defined decision?Examples:seasonal comparison may require more than twelve months of trends; diagnosing collection failures may require only weeks of logs; a long B2B campaign may need several months; retaining a visitor identifier for years “just in case” is harder to justify; an aggregate monthly total may remain useful longer than detailed events.Retention should be proportionate to granularity and risk. Map seven retention layers 1. Terminal trackers and identifiers This includes cookies, local-storage IDs, session identifiers and other persistent mechanisms. Document initial duration, renewal, domain scope, cross-site use, withdrawal deletion and consent expiry. For certain audience-measurement trackers that may meet the French exemption conditions, the CNIL recommends a lifetime allowing meaningful comparison, such as thirteen months, without automatic extension at every visit. That does not make thirteen months correct for every identifier. It is a benchmark tied to the specific framework and conditions. 2. Raw events Events contain the most detail: timestamp, page, source, device, event properties, identifiers and technical metadata. They support diagnostics, funnels and report reconstruction. Their granularity increases singling-out risk and governance cost. Ask whether raw events are needed beyond the analysis cycle, whether they can be aggregated after 30, 90 or 180 days, which fields can disappear earlier, whether invalid events are stored, and whether schema deletion affects history. 3. Pseudonymised data A replaced, hashed or rotating identifier is not automatically anonymous. If events can still be linked, treat this layer cautiously. Define method, re-identification possibility, table separation, access, rotation, mapping-table deletion and the real need for continuity. Rotation helps only when old periods cannot be reconnected without a justified process. 4. Aggregate statistics Daily or monthly reports can sometimes remain longer, especially when they no longer single out a person. Examples: monthly visits by site monthly conversions by channel weekly page views by content category quarterly conversion rate“Aggregate” is not magic. A cell with one conversion in a small region may still reveal information. Set thresholds, reduce dimensions and test sparsity. If data are truly anonymous, GDPR personal-data rules no longer apply. That conclusion requires evidence. 5. Technical logs Collectors, CDNs, reverse proxies and applications may keep IP addresses, full URLs, user-agents, errors, rejected payloads and request IDs. Logs are frequently missed because they do not appear in the analytics dashboard. They serve separate security, availability and diagnostic purposes. Give them a separate period, restricted access and sensitive-field filtering. 6. Backups A backup is not a permanent deletion exception. Document frequency, rotation, encryption, access, restoration, treatment of previously deleted data, purge timeline and offline snapshots. Immediate editing of every backup may be impractical. In that case, limit access and duration, and prevent restored expired data from returning to active use. 7. Exports and secondary systems Exports create parallel retention:emailed CSV files; spreadsheets; warehouses; BI systems; agency storage; presentations; support tickets; local backups.The vendor's setting does not erase these copies. Every export needs an owner, purpose, approved location, period, access rule and deletion method. Reducing exports is often simpler than governing many copies. Build a retention matrix An illustrative SaaS matrix:Layer Purpose Proposed period Trigger Deletion OwnerLimited visit identifier Trend measurement Up to 13 months under the assessed framework Creation, no automatic renewal Expiry Analytics ownerRaw events Analysis and diagnostics 6 months Event date Automated job DataAggregate monthly reports Historical steering 36 months Month end Aggregate then purge Finance/GrowthCollector logs Security and errors 30 days Receipt Rotation EngineeringBackups Recovery 60 days Snapshot creation Rotation InfrastructureOne-off exports Defined analysis 90 days Export date Owner deletion AnalystConsent evidence Demonstrate choice Defined from risk and limitation periods Choice Dedicated policy PrivacyValues are illustrative unless a specific framework is cited. Another organisation can reach different periods after assessment. Define the trigger. Six months from collection differs from six months after last visit, campaign close or contract termination. Understand platform settings GA4: explorations versus aggregate reports GA4 documentation says retention controls apply to user-level and event-level data. Standard properties document options including 2 or 14 months for user and event data. Google also says the setting does not affect standard aggregate reports and applies to surfaces such as Explorations and funnel reporting. A team can therefore believe it reduced all retention while aggregate reporting remains available. “Reset on new activity” can extend certain user-data expiry with each event. Check whether that matches policy. Other vendors Privacy-first SaaS, self-hosted platforms and cloud tools differ in fixed or configurable periods, raw-event availability, aggregate retention, provider backups, site deletion, APIs and contract-end deletion. Do not copy a period from a comparison article. Verify current documentation and contract terms. The CNIL twenty-five-month recommendation For audience-measurement trackers under the French exemption framework it describes, the CNIL recommends a maximum of twenty-five months for collected information, with periodic review. Three cautions:maximum is not the default: less may be sufficient; the framework is conditional: purpose, anonymous statistics, no combining and other conditions still matter; layers remain separate: logs, exports and identifiers need their own assessment.A policy can therefore keep detailed events for six months and aggregate reports for twenty-four, when justified. It should not copy twenty-five months into every database. Test deletion An untested policy is an intention. Test 1: expired data Use test events or a short-period environment. Verify disappearance from storage, APIs and affected reports. Test 2: identifiers Check cookie or identifier expiry, renewal and withdrawal behaviour. Test 3: exports Track a file from creation to deletion, including copies, attachments and recycle bins. Test 4: restored backups Restore into an isolated environment and verify that deletion rules are reapplied or expired data are blocked from use. Test 5: contract termination Ask the vendor about export windows, deletion date, backups, confirmation, legally retained data and subprocessors. Avoid silent extensions Retention can be prolonged by cookie renewal, user-row updates, warehouse imports, sparse aggregates, never-rotated backups, shared-folder copies, reused test accounts and changes applying only to future data. Document whether a change acts retroactively. Retention across multiple sites One policy may not fit a corporate site, authenticated app and help centre. Purposes, data, risks and countries differ. Use a common baseline with approved exceptions. The multi-site dashboard can show collection health, while administration tracks the retention mode and review date of every property. Do not share a cross-site identifier merely to simplify retention. Who validates the period? Business owners explain need. Analytics describes granularity. Engineering confirms deletion. Security owns logs. Privacy reviews the framework. Infrastructure covers backups. Procurement checks contracts. Name a final owner. “Data team” is not enough. The data collection summary and analytics privacy notice should show consistent periods at different levels of detail. A quarterly ten-question reviewHave purposes changed? Are detailed data still used? Can long-term reports be more aggregated? Do identifiers renew? Did deletion jobs succeed? Are there new exports? Are backups rotating? Did vendor policy change? Are public documents aligned? Can an inactive property be deleted?Removing an unused site, export or event is often the strongest risk reduction. Conclusion Retention is not a number in a policy. It is a system connecting purpose, granularity, access, trigger and deletion. A sound policy:separates identifiers, events, aggregates, logs, backups and exports; justifies each period; avoids silent renewal; tests purge operations; documents exceptions; aligns tools, contracts and public information.Begin with the most detailed data. Ask how long it remains useful, then aggregate or delete. Long-term history should not require indefinite individual-level collection. FAQ Does the CNIL always require twenty-five months? No. It recommends a twenty-five-month maximum for collected information under the conditional framework for certain audience-measurement trackers. A shorter period may be sufficient or necessary. Can aggregate statistics be kept indefinitely? Only when they are truly anonymous or another framework justifies retention. Sparse or highly segmented aggregates can still single out people. Does GA4 retention remove all reports? No. Google states that user and event retention settings do not affect standard aggregate reports. Understand which data and surfaces are covered. Must backup data be deleted immediately? It depends on architecture and risk. Backups need limited rotation, controlled access and procedures preventing expired data from returning to use after restoration. When does the retention clock begin? Define the trigger: collection, last activity, campaign close, aggregation or contract termination. Avoid automatic renewal where the framework or policy does not allow it. SourcesCNIL, Data retention periods CNIL, Audience-measurement cookies and consent conditions Regulation (EU) 2016/679, Article 5(1)(e) Google Analytics, Data retention Plausible, Data policy CNIL, Practical guide to retention periods