Tag: Analytics tool selection
All blog posts with this tag.
- 29 Jun, 2026
How to choose an analytics tool: a 15-question governance scorecard
Analytics comparisons often begin with feature lists and end with an overall score that assumes every team wants the same thing. They do not. A B2B small business measuring three sites, sharing reports with an agency and limiting collection has different constraints from an ecommerce group connected to advertising platforms. A technical team willing to operate self-hosted software makes a different trade-off from a marketing team without systems administration. The best tool is not the one with the most features. It is the one whose capabilities, defaults and governance model fit the need. The existing Google Analytics, Matomo and lightweight analytics comparison explains the main families. This scorecard supports an actual procurement decision without relying on a static ranking or prices that will change. Write the problem before scoring tools A serious selection fits on one page before the first demo. Decisions to support List five to ten decisions:Which channels create qualified requests? Which landing pages contribute to conversion? Which content is used? Which sites are growing or declining? Which product events matter? Where did collection break?“Collect all the data” is not a decision. Users Identify marketing, product, management, agency, analyst, engineering, privacy and external-client roles. The key issue is not only user count. It is rights, skills and usage frequency. Constraints Document site count, expected volume, countries, consent requirements, prohibited data, retention, integrations, budget, operating capacity, migration timeline and contractual requirements. This prevents a polished demo from replacing analysis. The 15-question scorecard Score each tool from 0 to 3:0: unavailable or incompatible; 1: possible with a major workaround; 2: covered with acceptable configuration or limits; 3: naturally supported and documented.Apply a weight from 1 to 3 for your organisation. 1. Can it answer business questions without heavy reconstruction? Test five scenarios, such as campaign landing pages, demo requests, three-site comparison, monthly export and traffic-drop diagnosis. A powerful tool that real users cannot operate is expensive. 2. What granularity is genuinely required? Separate aggregate statistics, events, journeys, cohorts, funnels, user identifiers, advertising data and session replay. Each layer adds capability and governance. A team using only pages, sources and conversions should not select primarily for detailed behavioural analysis. 3. What does the default setup collect? Check cookies or identifiers, IP address, full URL, user-agent, geography, advertising IDs, automatic events, query parameters and cross-site signals. Use the URL parameter audit and inspect a test payload. A default aligned with policy scores better because every option to disable is configuration debt. 4. Are strict and extended modes clearly separated? Some teams want minimal measurement by default where the local framework allows it, with extended capabilities after consent. Assess configuration separation, pre-consent behaviour, signal propagation, accidental activation risk, change history and documentation. A vague privacy switch is not enough. 5. Can you explain the data flow? You should be able to state where requests arrive, which transformations occur, where data are stored, which subprocessors participate, whether the vendor reuses data, which support access exists and which transfers apply. Use the data collection summary as the evaluation format. 6. Do location and transfers fit your constraints? Assess hosting region, contracting entity, subprocessors, remote access, transfer mechanisms and self-hosting options separately. “EU hosted” is one fact, not a complete assessment. Self-hosting provides control while transferring operational duties to the team. 7. Can retention, deletion and backups be governed? Ask about available periods, raw versus aggregate data, automatic deletion, property-level deletion, backup purge, exports, contract-end deletion and operational evidence. Unlimited default retention is not neutral. 8. Do multi-site permissions fit? Test per-site roles, groups, agency access, read-only access, exports, admin logs, SSO where needed, offboarding and portfolio views. The multi-site dashboard model turns these into practical scenarios. 9. How does it handle data quality? Review bots, test environments, duplicates, invalid events, unknown parameters, ingestion delay, definition changes, alerts, time zones and cardinality. A simple report without diagnostics may be too shallow. A rich platform with opaque transformations may be hard to audit. 10. Can events and conversions remain governed? Test event creation, change and deprecation. Can schemas be validated? Can fields be forbidden? Are changes versioned? Do historical goals remain understandable? Can an agency change collection without approval? Easy event creation is not always an advantage. Missing guardrails quickly creates an unreadable taxonomy. 11. Is acquisition readable without excessive setup? Test source and medium, UTM campaigns, referrers, direct, landing pages, conversions, custom channels and only the attribution models you genuinely need. Platforms can classify the same journey differently. Ask how rules are defined and changed. 12. Can history be migrated and compared? Check import availability, format, granularity, supported metrics, cost, duration, definition differences, source-data preservation and break-point labelling. An import does not erase differences in sessions, visitors or conversions. 13. Is cost predictable at your scale? Include subscription by volume, overages, sites, users, modules, storage, hosting, maintenance, support, consent management, configuration time, manual reporting, migration and exit. For self-hosting, include updates, backups, monitoring, security and availability. For SaaS, model traffic growth and plan-gated features. Verify prices at decision time. 14. What operating burden can the team sustain? List installation, configuration, tests, maintenance, access, compliance, alerts, backups, support, training and documentation. A free tool can be expensive to operate. A simple tool can be expensive if every useful question requires manual exports. 15. Can you exit cleanly? Evaluate full export, open formats, API, post-cancellation access, deletion, portable configuration, event recovery, definition history and proprietary identifier dependence. A good fit today can still create future debt through lock-in. A weighting example For a multi-site B2B SaaS company:Criterion WeightBusiness questions 3Granularity 2Default collection 3Strict/extended separation 3Data flow 3Location and transfers 2Retention and deletion 3Multi-site access 3Data quality 2Event governance 2Acquisition 2Migration 2Total cost 3Operating burden 3Reversibility 2Calculate: sum(score × weight) / sum(3 × weight)A percentage is convenient, but a two-point difference is not scientific truth. The criteria discussion matters more than the final rank. Add disqualifying criteria Some conditions cannot be offset:unavailable DPA; no deletion; no export; prohibited data collected without control; incompatible multi-site access; unacceptable transfer; cost beyond budget; impossible operating burden; missing essential capability.A tool can score 85% and still fail one critical requirement. Reading the main tool families Rich analytics and advertising suites GA4 integrates deeply with Google's ecosystem and provides broad dimensions, explorations and advertising connections. This can fit trained teams with real attribution and activation needs. It also requires more event governance, configuration and understanding of scopes. Familiarity alone is not a selection criterion. Controllable and self-hostable platforms Matomo offers cloud and self-hosted options with broad functionality. Umami and other open-source projects take more compact approaches. Self-hosting increases infrastructure control, but the organisation owns operations. Decide who patches, restores backups and monitors access. Privacy-first analytics SaaS Plausible, Fathom, Simple Analytics and related tools prioritise readable reporting and often more limited collection. They can reduce setup for essential needs. Simplicity may limit advanced analysis, complex event schemas or some consolidation patterns. Verify actual capabilities, not only philosophy. Emerging products A beta or launch-stage product may align well with governance, but assess maturity, documentation, support, export, stability and demonstrated roadmap. Never score a roadmap promise as an available feature. Run a two-week pilot Day 1: confirm scenarios Select five questions, three roles and two representative properties. Days 2 to 4: deploy a small scope Install each candidate on a test environment or pilot property with the same pages and events. Days 5 to 7: audit collection Compare network requests, documented storage, consent, URL parameters and access. Days 8 to 10: user tests Ask marketing, product and administration users to complete the same tasks without excessive assistance. Days 11 to 12: test export and deletion Export data, revoke access, change retention and review deletion procedures. Days 13 to 14: score and document Complete the scorecard, list disqualifiers and write down accepted compromises. Common selection mistakes Choosing from a demo A demo shows the best workflow, not routine operations. Choosing on privacy alone Privacy is a major design constraint, but the tool must support decisions. An unused solution does not improve governance. Choosing on features alone A feature that expands collection or requires a dedicated team can be a cost rather than value. Comparing prices without future volume Model twelve- and twenty-four-month scenarios. Forgetting people The theoretical best tool fails when nobody understands its reports or maintains its rules. Conclusion A useful comparison does not ask, “Which tool is best?” It asks, “Which tool creates the best compromise for this organisation?” The scorecard makes visible:expected decisions; default collection; governance; access; retention; multi-site needs; total cost; operating burden; exit capability.The result is not a universal score. It is an explainable, reviewable and documented decision. FAQ How many tools should be tested? Three well-chosen candidates are often enough: a rich reference platform, a more controllable option and a simple privacy-first option. Add a fourth only when it represents a genuinely different model. Is self-hosting always more compliant? No. It increases potential control, but compliance still depends on configuration, security, access, purpose, retention and real operations. Can prices be compared once and reused? No. Plans, limits and prices change. Verify them at decision time and model several volumes. How should a roadmap feature be scored? Treat it as unavailable until it can be used and verified. A roadmap can affect risk, but it cannot satisfy a current requirement. What is the difference between a weighted and disqualifying criterion? A weighted weakness can be offset by other strengths. A disqualifying criterion makes the tool incompatible regardless of its total score. SourcesCNIL, Audience-measurement cookies and consent conditions Regulation (EU) 2016/679, data minimisation, transparency and accountability principles Google Analytics, Analytics account structure Google Analytics, Data retention Matomo, Privacy Plausible, Data policy Fathom Analytics, Data policy Umami, Documentation