Business analytics: BI, product analytics and how an enterprise buys them

BI vs product analytics vs web analytics vs embedded analytics

Google Analytics 4 also records events. Google says an event can measure a page load, a link click, a purchase or an app crash.¹³

Embedded analytics splits by audience. Microsoft's "embed for your customers" model, also called app owns data, serves external users who need no Power BI license. Its "embed for your organization" model, user owns data, requires a Power BI license for each user.^14,15

The semantic layer and metric definitions

A semantic layer holds the definitions that reports share. In Looker, semantic data models are written in LookML. A model describes dimensions, aggregates, calculations and data relationships in the SQL database.^6,16

Google's documentation says SQL expressions are written once, in one place, and Looker reuses them to generate ad hoc queries.¹⁷

Power BI calls its layer a semantic model. A model built in Power BI Desktop is effectively an Analysis Services tabular model.^9,18

The alternative is a layer outside any one BI tool. dbt Labs' Semantic Layer moves metric definitions out of the BI layer and into the modeling layer. When a definition changes in dbt, it changes everywhere the metric is used. It requires a dbt Starter or Enterprise account.^6,19,20

LookML and Power BI semantic models keep definitions inside the BI product. dbt's layer serves other tools, and AI clients such as Claude and ChatGPT through its MCP server, on a paid dbt account.^9,16,20,21

Our pages on business analytics

• Business Intelligence & AnalyticsBuyer Guide
• Product Analytics PlatformsBuyer Guide
• Embedded Analytics PlatformsBuyer Guide
• Digital Adoption Platforms (DAP)Buyer Guide
• Microsoft Power BI vs TableauComparison
• Tableau vs ThoughtSpotComparison
• Qlik vs TableauComparison
• Sigma vs TableauComparison
• Domo vs TableauComparison
• Amplitude vs MixpanelComparison
• PostHog vs AmplitudeComparison
• Pendo vs AmplitudeComparison
• RFP questions: Business Intelligence & AnalyticsRFP questions
• The BI transformation processArticle
• Migrating BI reports between platformsArticle

Self-service BI and governance: who owns the content

Microsoft's adoption guidance names three strategies for owning data, analytics and BI content.

• Business-led self-service. Creators and subject matter experts in a business unit own and manage all content. It gets the least stringent governance.^22,23
• Managed self-service. A central team owns the data, and business users own the reports and dashboards. Microsoft calls it discipline at the core and flexibility at the edge.²⁴
• Enterprise. A central team such as IT, enterprise BI or a center of excellence owns everything. It gets the most rigorous controls.^25,26

Most organizations run more than one. The strategy can differ by solution and by team.^27,28

Microsoft lists the signs of the least mature state. Creators manage content without a strategy. Many semantic models each support only one report. Reports disagree with one another, and people stop trusting content others produce.^29,30,31

Its advice for self-service is to teach creators the techniques IT uses, such as shared semantic models, and to centralize data to reduce maintenance and risk.^32,33

Power BI, Tableau, Looker, Amplitude and Mixpanel pricing

Power BI and Tableau list prices are per user. Amplitude and Mixpanel price by event volume and include unlimited seats.^34,37,44,45

Capacity does not remove per-user licenses in Power BI. Publishing content to a Premium or Fabric capacity still needs a Pro license. Viewers can use reports without a paid license only on P1 or F64 capacity and above.^46,47

GDPR, ePrivacy and consent for web and product analytics

Article 5(3) of the ePrivacy Directive allows storing or reading information on a user's device only with consent, or where it is necessary for purposes the article lists. The European Data Protection Board adopted version 2.0 of its guidelines on that article's scope on October 7, 2024.^48,49

The guidelines cover more than cookies. The EDPB says tracking pixels on websites can link to an entity that collects many such requests and so tracks users' behavior.⁵⁰

France's regulator, the CNIL, says analytics cookies need consent unless they meet every one of these conditions, in which case users only need a way to object.

• Users are told about the trackers and can object to them.^51,52
• The purpose is limited to audience measurement and A/B testing.⁵³
• The data is not combined with other processing, such as customer files.⁵³
• The tracker serves a single site or app publisher.⁵⁴
• The last byte of the IP address is truncated, and trackers last no more than 13 months.⁵⁴

The CNIL adds that most large audience measurement offerings fall outside the exemption whatever their configuration. Its guidance is tied to the ePrivacy Directive, and national positions vary.^55,56

Google Analytics keeps user-level data for 2 or 14 months. Other event data can be kept for 26, 38 or 50 months only on Analytics 360.^57,58

Switching from web analytics to product analytics

Standard Universal Analytics properties stopped processing hits on July 1, 2023. From the week of July 1, 2024, current and historical Universal Analytics data was no longer accessible.^59,60

A product analytics rollout starts with a taxonomy: which events and properties to track, what to call them, and how they fit together. Amplitude says a solid taxonomy prevents data gaps and costly schema rework later.⁶¹

Naming has to be agreed across teams. Amplitude records "Song Played" and "song played" as two separate events, and recommends one naming convention to prevent messy data when several teams send the same event.^62,63

Mixpanel's Lexicon is a data dictionary of events and properties, and it can block events that are no longer useful without a code change.^64,65

Google Analytics 360 raises the standard product's limits. Its exploration sampling limit is 1 billion events per query, against 10 million on the standard product. Its BigQuery daily export handles billions of events, against 1 million.^66,67,68

Where business analytics programs go wrong

In the same 2026 dbt Labs survey, poor data quality was the most frequently reported obstacle, and stakeholder data literacy was a barrier for 36% of respondents.⁷⁰

• Duplicate models. Many semantic models that each serve one report, and reports that disagree with one another, are signs Microsoft lists of the least mature state.^30,31
• Licenses missed in the budget. Below P1 or F64 capacity, Power BI viewers need paid per-user licenses.⁴⁷
• Event volume outgrows the plan. When a standard Google Analytics property becomes large, event-level retention drops to 2 months and older event data is deleted.⁷²
• Consent assumed. In France, most large audience measurement offerings do not qualify for the consent exemption.⁵⁵