What is analytics in marketing?

The layers

• Collection. Instrumenting events so that behaviour is recorded at all.
• Processing. ETL into a consistent structure, usually a data warehouse.
• Analysis. Cohorts, segments, funnels and tests.
• Presentation. Dashboards and reporting that people actually read.
• Decision. The only layer that produces value, and the one most often missing.

Where analytics practices fail

Rarely at the tooling layer, despite where most of the money goes.

Instrumentation debt. Events added ad hoc, named inconsistently, and never removed. Two years in, nobody is confident what `purchase_complete` includes, and every analysis begins with an archaeology project.

Metrics without decisions. Dashboards that are read and never acted on. If no possible value of a metric would change what anyone does, tracking it is a cost with no return.

Precision without accuracy. Reporting a figure to two decimal places when the underlying measurement has a 30% blind spot — see attribution.

Confusing observation with causation, which is the expensive one, and the reason incrementality testing exists.

What changed

Analytics inside your own product is largely unaffected by the privacy transition — it is first-party data about people using your service.

What degraded is everything that required following a user across companies: attribution, cross-device journeys, third-party audiences. The response has been a shift in emphasis from tracking individuals to measuring effects: more aggregate causal methods, more first-party depth, less faith in the reconstructed path.

For AI surfaces the pattern is sharper still. A large share of influence produces no observable event anywhere in your analytics, so the honest posture is to measure the outcome and accept that the mechanism is not visible.