August 2026 Summaries

Growing use of AI agents in customer research and purchasing is increasing demand for customer data systems that provide real-time, comprehensive context rather than batch-based campaign data, according to Databricks, Snowplow, and Gartner, which predicts that 80% of new enterprise CDP deployments will be embedded in or composable with data platforms by 2030. The post argues that standalone CDPs can create duplicated, incomplete, or delayed customer profiles and may be limited by browser tracking restrictions, making them less suitable for AI-driven personalization. Databricks’ CustomerLake, currently in private preview and expected to reach general availability in late 2026 or early 2027, is presented as an agentic CDP built within Databricks that supports continuously adjusted, individually personalized “Infinity Campaigns” under existing data governance rules. Snowplow is positioned as the upstream source of validated, identity-resolved, consent-aware behavioral data streamed continuously into Databricks, while the post advises organizations approaching CDP renewals to assess real-time responsiveness, completeness of customer data within their own platform, and the strategic direction of their current vendor.

Snowplow announced that its AI management system has been certified to ISO/IEC 42001:2023 following an independent audit by Citation ISO Certification Limited, with the certification effective July 21, 2026 and valid through July 20, 2027. The certification applies to Snowplow’s governance processes for AI used in delivering, managing, and customizing behavioral-data infrastructure, including tracking-plan management, identity resolution, and personalized-interaction triggers, rather than certifying individual products, models, or customer AI systems. ISO 42001 evaluates lifecycle controls such as risk and impact assessment, data governance, supplier oversight, and continuous improvement, and complements Snowplow’s ISO/IEC 27001 and SOC 2 Type II credentials. For customers, the certification can serve as supplier-assurance evidence when establishing AI governance programs or preparing for regulations such as the EU AI Act, though it does not extend certification to their own systems. Snowplow states that customers retain control of behavioral data in their own warehouses or lakes, while its Assistant follows existing permissions, cannot access PII, and requires confirmation before data-changing actions; the company also notes that its certifying body is accredited by a private organization rather than the United Kingdom’s national accreditation body.

Databricks has introduced CustomerLake, an agentic customer data platform in private preview that operates within its lakehouse environment, using AI Profile Agents to build governed customer profiles and Campaign Agents to deliver continuous, real-time engagement; general availability is expected in late 2026 or early 2027. The discussion argues that such systems depend heavily on the completeness, quality, identity resolution, and timeliness of behavioral data already collected in the lakehouse, since CustomerLake does not itself capture web or app events. It frames the market shift toward data-platform-embedded or composable CDPs as a response to data duplication, synchronization delays, governance needs, and the demand for in-session decisioning, citing Gartner projections and consolidation among standalone CDP vendors. It recommends assessing data collection against four criteria—context, control, cost, and choice—to ensure events are structured and validated, owned and governable, portable across platforms, and available in real time. Snowplow presents its own collection and identity-resolution tools as a way to supply this behavioral context across multiple destinations, while disclosing that Databricks is an investor in Snowplow.

Snowplow Signals has introduced a Python SDK capability that creates labeled training tables directly from existing attribute groups, aiming to prevent training-serving skew caused by maintaining separate feature definitions for historical model training and live deployment. Users can define a predictive goal, such as a purchase, or provide their own labeled event anchors, after which the builder recomputes point-in-time accurate attributes using only events that occurred before each prediction point. It can execute generated queries or provide SQL for review and versioning, writing datasets to the user’s warehouse schema and returning previews as pandas DataFrames for notebook-based model development. The feature is intended for real-time behavioral propensity applications such as purchase intent, trial conversion, churn, booking inquiries, registration timing, and content next-action prediction. Existing customers can access it by upgrading the snowplow-signals Python SDK, while prospective users can evaluate the product through a 14-day free trial.