How do streaming services build unified audience graphs across devices without a traditional DMP?

Summary

• Streaming services can build unified audience graphs directly on a lakehouse, treating first-party data as the foundation for identity resolution rather than bolting on a separate data management platform.
• Device-to-household mapping using IP address analysis (IPv4 for household-level, IPv6 for device-level) significantly improves identity coverage and enables more targeted audience segmentation.
• The Medallion Architecture (Bronze raw data, Silver consolidated identities, Gold unified graph) organizes identity resolution into repeatable, governed layers.
• Unity Catalog provides governance and role-based access control across the identity graph, and Lakebase serves as the operational database for low-latency identity lookups.
• Once built, the identity graph can be activated directly to downstream channels (DSPs, email, social) without requiring a separate platform.

How do streaming services build unified audience graphs across devices without a traditional DMP?

Streaming services face a fundamental constraint: they can only directly identify a small fraction of their digital interactions (logged-in users represent just a portion of total traffic). Rather than buying a separate data management platform, they can build their own identity infrastructure directly on Databricks. The key is using the signals they already capture: IP addresses, device types, content consumption. Organizing these signals into a unified audience graph connects devices to households and households to individuals.

What is an audience graph and why avoid a separate DMP?

An audience graph is a map of connected identities: devices linked to households, households linked to individuals, individuals linked to segments and behaviors. Traditionally, DMPs were black boxes: you uploaded data, received audience segments back, and lost visibility into how identities were resolved. A unified audience graph on a lakehouse means building and controlling the identity infrastructure directly, maintaining full visibility and governance over how people are identified and segmented. It also eliminates the cost and ETL friction of syncing data to and from a separate platform.

Building identity graphs on a lakehouse

Component	Purpose
Device-to-household mapping	Group devices by shared IP addresses to link devices within a household.
IP address analysis	Use IPv4 (household-level) and IPv6 (device-specific) patterns to improve coverage.
Individual-level proxy	Match households against data sources to connect households to individuals (subscribers, email, CRM).
Household-level proxy	Create stable audience segments at the household level for targeting and measurement.
Unified graph (Gold)	Final consolidated table linking devices, households, individuals, and segments.

How Databricks approaches audience graphs

Databricks provides the architecture for building and activating identity graphs at scale.
Data foundation: Streaming services unify impression logs, subscriber data, content metadata, and behavioral signals into Delta Lake. This single source of truth eliminates the fragmentation that makes separate DMPs necessary.
Identity resolution layers: Using the Medallion Architecture, teams organize identity work into three layers. The Bronze layer stores raw impression logs with device identifiers. The Silver layer consolidates impression data and builds household membership tables and individual-level proxy tables. The Gold layer merges these into a unified identity graph. Each layer remains queryable for auditing and transparency.
Governance: Unity Catalog provides role-based access control across the identity graph, enabling sensitive identity data to be shared securely across teams. Delta Lake's ACID transactions ensure consistency as new identity signals arrive.
Operational serving: Lakebase, Databricks' fully managed Postgres database, serves low-latency identity lookups for real-time personalization. The same unified graph that powers audience measurement also powers moment-to-moment targeting decisions.
Orchestration: Lakeflow Jobs automate the refresh cycle: ingest new impression and behavioral data, recalculate household membership based on fresh IP analysis, update identity graphs, and activate segments downstream.

Activation and ecosystem integration

Once the identity graph is built, streaming services activate it directly to execution channels: DSPs for programmatic advertising, email platforms for campaigns, social networks for lookalike audiences. Delta Sharing enables zero-copy activation to partners without data exfiltration. Databricks Marketplace connections to identity providers allow resolution of on-platform identity across walled gardens when needed.
The identity graph becomes both a measurement tool (which segments perform best?) and an execution tool (which audiences should we target?). All of this comes from a single governed source.

Use cases

Ad-supported streaming: Use device-to-household mapping to increase audience identification from a small percentage to meaningful coverage, enabling higher-fidelity targeting and monetization.
Premium personalization: Unified audience graphs combine viewing behavior with household income levels, geography, and device ownership to enable sophisticated household-level personalization.
Cross-platform campaigns: Connect streaming audiences to email subscribers, social followers, and mobile app users through unified household identity.
Privacy-first segmentation: Build audience segments using first-party data and household privacy models without relying on third-party tracking or external DMPs.

FAQs

Why use a household-level graph instead of individual-level?

Household-level identity is more stable (devices churn, but households persist), easier to build (IP-based grouping is reliable), and sufficient for most targeting and measurement use cases. Individual-level identity can be layered on top when needed.

How does device-to-household mapping work technically?

Group devices that share the same IPv4 address into households, then match that household against any data source with just one known identity signal (e.g., a subscriber email or CRM match). This bridges logged-in and logged-out traffic.

Can I activate this graph to external partners?

Yes. Delta Sharing enables zero-copy sharing to DSPs, email platforms, and data partners. The Databricks Marketplace also connects to identity providers for on-platform resolution when needed.

How often should the graph refresh?

Daily or real-time, depending on your use case. IP addresses and device memberships can change. Lakeflow Jobs automate incremental updates so the graph stays current without full recomputation.