Uneven Tracking Loss Biases Experiments and Causal Estimates

Client-side JavaScript has long been the default way to collect behavioural data because it is simple and works everywhere. Its reliability has declined, and the losses now undermine the analysis built on top of it.

Ad blockers, browser privacy features such as Intelligent Tracking Prevention (ITP) and consent management platforms each remove part of the picture. The visible symptom is a persistent gap between GA4 and internal databases, which teams then spend time reconciling.

The less visible problem is worse. Causal inference assumes that outcomes are measured the same way in every group being compared. When treatment and control groups are subject to different levels of data capture, part of the measured difference reflects tracking rather than the intervention. This applies to A/B tests, difference-in-differences designs and other quasi-experiments alike.

Many inconclusive or contradictory A/B tests trace back to data integrity rather than experimental design. Before attributing a result to the treatment, compare capture rates across groups against a source that does not depend on the browser, such as server logs or backend transactions.

Server-side collection reduces the problem at source. Controlling the data stream before the browser can interfere improves completeness and consistency, allows precise event definitions and enrichment with internal data, and delivers events to destinations such as GA4 and its BigQuery export. Solutions such as Robin Signals capture this telemetry with privacy built in.

The gain is not simply more data but more consistent data, which is what causal methods need. Models built on server-side pipelines tend to be more stable and more predictive than those built on fragmented client-side data, and the same pipeline can join the customer journey to CRM and operational data.