The data pipeline stack, reviewed

Three moves settled the shape of this market. Fivetran acquired Census, widely credited with pioneering reverse ETL, in May 2025 and rebranded it Fivetran Activations. Databricks announced CustomerLake on 16 June 2026 at its Data and AI Summit, an agentic customer data platform embedded in the lakehouse and governed by Unity Catalog, consolidating customer data, identity resolution, audience building, campaign automation and activation, currently in private preview. Hightouch repositioned from composable CDP to agentic CDP in June 2026, having moved from reverse ETL in 2022 to composable CDP and then to agentic marketing platform in 2025.

The CDP Institute's January 2026 Industry Update, covering the second half of 2025, found composable warehouse-native CDP vendors grew employment 7.8 per cent organically against a 1.3 per cent industry average, nearly six times the rate. More than 25 per cent of CDPs now support a warehouse-centric architecture, while delivery CDPs grew 3.3 per cent.

If the company already has a warehouse, the activation layer is an add-on rather than a platform decision, and the 2026 comparison of Segment, Hightouch and Census describes the warehouse-native reverse-ETL pattern as the leading one. Segment remains the straightforward choice where the requirement is event collection and forwarding rather than activation from a warehouse, which is why it appears here as the incumbent being displaced from the activation layer rather than from the collection layer. If it does not have a warehouse, buying a delivery CDP to avoid building one usually means buying a second copy of the customer record, which is the condition that produces a board deck where CRM and marketing disagree.

The choice is therefore less about vendor features than about sequence. Build the warehouse and the models first, then activate from them. That ordering is why data pipeline integration is stage 3 of the framework, after the go-to-market blueprint and the stack, rather than first.

From 15 June 2026 the Google Signals setting in GA4 no longer governs Google Ads data collection, and the Consent Mode ad_storage parameter alone does. A consent banner that was not correctly wired to Consent Mode before that date costs conversion accuracy, remarketing audiences and attribution quality. UTM parameters are read before cookie logic, so tagged traffic still attributes under denied consent.

Chrome, after years of signalling deprecation, kept third-party cookies with user controls in 2025 while continuing Privacy Sandbox alternatives, so the change arrived as gradual erosion rather than a cut-over. The practical design rule is unchanged by either: capture consent state once, store it somewhere authoritative, and propagate it to every tool that reads or writes customer data.