What Is Databricks CustomerLake? Agentic CDP

Databricks CustomerLake is an agentic customer data platform (CDP) built inside the Databricks lakehouse — and as of October 19, 2026, its Campaign Agents are in Beta, its Profile Agents are in Private Preview, and Databricks has published early results from five named customers. The product Databricks announced at Data + AI Summit on June 16 has moved from architecture to production claims: audience builds measured in hours instead of months, and a projected cut of up to 40% in CDP total cost of ownership at HP.

This guide covers what CustomerLake is, what the October beta actually proves, where it fits in the CDP category’s evolution, and what remains unproven.

Databricks CustomerLake at a Glance

No peer-review mix exists: CustomerLake has no G2 review base and Campaign Agents only reached Beta in October 2026, so the row above describes who can adopt it — organizations already running the Databricks lakehouse under Unity Catalog — not measured usage. The cross-vendor comparison and the methodology behind these splits are in CDP vendors.

What Is CustomerLake?

CustomerLake is Databricks’ first marketing product — an Agentic CDP that unifies customer data, AI models, identity resolution, audience building, and activation on the same lakehouse where organizations already run analytics and ML workloads.

Since October 19, 2026 the product ships as two independently adoptable tracks: Campaign Agents in Beta and Profile Agents in Private Preview. A team can run campaign workflows on customer data it already has, or use Profile Agents to normalize and unify records first; neither track requires the other. Access runs through Databricks account teams. The company has published no general availability date, no rate card, and no published Beta eligibility criteria (CMSWire, October 2026).

The core concept is what Databricks calls Infinity campaigns — a shift from static, one-off campaigns to continuous agentic loops where AI agents constantly analyze customer signals, decide next-best actions, and optimize engagement in real time. As CEO Ali Ghodsi framed it: “Marketing stops being a series of campaigns and becomes a continuous loop — agents that constantly analyze, decide, and act on every customer in real time.”

CustomerLake is built on three design principles:

• Embedded — built directly on existing Databricks data foundations. No data duplication into a separate CDP
• Democratized — marketers operate through agent-first interfaces on trusted data, reducing ad hoc requests to data teams
• Autonomous — agents analyze customer signals, recommend next-best actions, and optimize engagement around business goals at 1:1 scale

CustomerLake follows the pattern Databricks established with Lakewatch (its security analytics product): building vertical application layers directly on the lakehouse rather than leaving those markets to third-party tools. This pattern signals that CustomerLake is unlikely to be the last category Databricks enters. A Microsoft announcement at the same summit introduced Azure Databricks CustomerLake, so availability is not confined to AWS deployments.

Team lineage matters: CustomerLake’s leadership includes Tasso Argyros and Justin DeBrabant (both from ActionIQ, an enterprise CDP acquired by Uniphore) and Katy Yuan (from Census, a composable CDP pioneer). The people who spent the past decade building packaged and composable CDPs are now inside the lakehouse, building something architecturally different from either.

Profile Agents

Profile Agents handle the data engineering side of CDP functionality — transforming raw customer data into business-ready Customer 360 profiles directly inside Databricks. They entered Private Preview on October 19, 2026.

Databricks describes three connected tasks:

• Source normalization — field mapping and semantic normalization bring data from different sources into a consistent structure and meaning for downstream use
• Agentic Identity Resolution — deterministic, probabilistic, and LLM-based matching connect fragmented records; golden-record survivorship rules determine which attributes the unified profile keeps
• Stateful, incremental processing — micro-batch updates keep profiles current as data changes, with manual overrides for data stewards applying human judgment

At launch, Databricks also listed an identity provider marketplace with Acxiom, Epsilon, LiveRamp, TransUnion, and Adstra for one-click access to third-party identity graphs, and said teams can bring their own identity rules and models alongside Agentic Identity Resolution.

The mechanics answer a question the June launch left open: how the identity resolution would actually work. They also sharpen the latency question. Databricks describes micro-batch updates but publishes no interval, so profile freshness is bounded by a cadence a buyer cannot yet verify — workable for campaign selection, unproven for in-session personalization. The daily audience syncs are a separate constraint: they govern activation cadence, not profile freshness.

For data engineers, Profile Agents reduce the manual pipeline work that typically consumes weeks when building customer 360 views from scratch on a lakehouse.

Campaign Agents

Campaign Agents handle the activation side — moving from audience creation to channel delivery with AI assistance at every step. They entered Beta on October 19, 2026, open to more customers who meet Databricks’ technical requirements.

What Campaign Agents provide:

• Goal-driven campaign generation — agents use business goals and customer context to recommend audiences, messaging, timing, and channel selection
• Genie-powered interface — marketers interact through Databricks’ natural language interface (Genie) to build audiences, define campaign briefs, and set guardrails — without writing SQL
• Pre-launch simulation — campaigns can be simulated against a sample of customer profiles before going live, providing visibility into how decisions will play out
• Built-in guardrails — marketing opt-out enforcement, suppression for customers with unresolved support tickets, and frequency capping are native to the decisioning layer

The Beta arrives with named production deployments, which is the first evidence that the marketer-facing workflow survives contact with real marketing teams:

• Zé Delivery (AB InBev) — the CRM team builds audiences in CustomerLake and syncs them to Braze without engineering support
• Philadelphia Union — hundreds of audience syncs a month to Salesforce Marketing Cloud after replacing its previous CDP
• Wine.com — 100 active audiences syncing daily to Braze across 6.5 million profiles, with production email campaigns live 23 days after kickoff

The pattern is consistent: CustomerLake assembles audiences inside the lakehouse and hands them to an engagement platform. Where cadence is published, it is batch: Wine.com syncs daily and the Philadelphia Union runs hundreds of syncs a month.

Databricks also ran Campaign Agents on its own marketing. For Data + AI Summit, the agents used each in-person attendee’s engagement history to personalize session-recommendation emails for 30,000 attendees, and Databricks reported a 10-point conversion lift against a control group, a 10% click-through rate, and a 60% open rate. That is a Databricks-run campaign on Databricks’ own data — an existence proof for agent-personalized email, not a benchmark for another company’s data estate.

Early Customer Results

Databricks frames the beta against a market shift: 74% of consumers use AI to discover, compare, or purchase products (CI&T, cited by Databricks, October 2026). The customer evidence it published is narrower than that framing but more concrete than anything available at launch.

Every figure is customer-reported or Databricks-reported; none is independently audited. Read the table as existence proofs — the capability has run in production at named enterprises — rather than as benchmarks.

Two patterns hold across the set. The time savings cluster in audience assembly and governed data access, the work that happens before activation. And the activation itself is sync-based: daily audience pushes, monthly sync volumes, micro-batch profile updates. None of the published numbers measures in-session personalization latency, which is why the real-time question stays open below.

Architecture

CustomerLake’s architecture reflects a fundamental bet: the data lakehouse should be the system of record for marketing, not a data source that feeds a separate CDP.

Key architectural components:

Activation partners at launch include Adobe, Meta (Audience and Conversions API), The Trade Desk, Braze, Bloomreach, Iterable, Snapchat, Magnite, Twilio, IAS, and Unity. The October deployments add Salesforce Marketing Cloud to the destinations customers actually sync to.

Pricing model: Databricks describes CustomerLake pricing as a consumption model — “a flexible, value-aligned alternative to traditional software licensing” on its product page — and has published no rate card. Martech Therapy’s Matthew Niederberger reports that it carries no platform fee, with Databricks earning on the compute used to build profiles and resolve identities. As he put it: “You cannot win a price war against a company that does not need your product to make money.”

HP’s estimate that consolidation could cut CDP total cost of ownership by up to 40% is the first customer-side cost signal (Databricks, October 2026) — but it is a projection about HP’s estate, not a rate card. This creates a structural asymmetry: standalone CDP vendors must recover their entire R&D and infrastructure costs from CDP revenue alone. Databricks can treat CustomerLake as a compute driver for its existing platform — even subsidizing the CDP layer if it increases lakehouse adoption. For CDP-only companies, this is an existential pricing challenge, not a feature gap.

For budgeting, treat CustomerLake as a compute line item rather than a license: no platform fee is reported (Niederberger, 2026), and the incremental cost scales with the compute used for profile builds, identity resolution, and agent runs. Ask Databricks for an estimate at your own profile count and campaign cadence, then stress-test it at double — continuous campaigns remove the ceiling that per-seat licensing imposes.

CDP Evolution: Where CustomerLake Fits

The CDP category has evolved through three stages, each driven by pressure to close the Customer Intelligence Loop faster:

The evolution from Packaged to Composable to Agentic reflects a deeper shift. The 2020–2025 era settled the question of data gravity — where customer data lives (answer: the warehouse/lakehouse won). The emerging question is decision gravity — who controls where and how business decisions are made on that data. CustomerLake is Databricks’ bid to own not just data storage but the decisioning layer: the place where AI agents analyze, decide, and act on customer data in real time.

CustomerLake is a Stage 3 entry — but it arrives from a different direction than most Agentic CDPs. Where traditional CDP vendors are adding AI agents to existing platforms, Databricks is adding CDP capabilities to an existing data platform. The result is architecturally similar but organizationally different: CustomerLake’s buyer is likely the data team (who already owns Databricks), not the marketing team (who typically owns CDP procurement).

The October milestone does not change that architecture. It changes the evidence behind it. Two of the named deployments describe replacing an existing CDP, and one of them — HP — names its former approach as composable.

CDP Capabilities: What CustomerLake Covers and What It Does Not

A customer data platform is defined by five core capabilities. Before CustomerLake, a Databricks lakehouse covered none of them natively. Here is where CustomerLake stands after the October release:

The most significant shift is identity resolution: Agentic Identity Resolution’s multi-method approach (deterministic + probabilistic + LLM matching) with golden-record survivorship and a built-in identity provider marketplace directly addresses what was previously Databricks’ weakest CDP gap. The most notable absence is native messaging — CustomerLake activates through partner channels, not built-in email, SMS, or push. This is a deliberate architectural choice: Databricks is building the intelligence and decisioning layer, not the delivery infrastructure.

The October evidence narrows that question rather than settling it. Daily audience syncs and micro-batch profile updates are enough for campaign selection; whether they deliver a profile lookup inside a live session is still unproven.

Competitive Impact

CustomerLake’s entry reshapes competitive dynamics across the CDP landscape, and the October results give the shift its first named evidence.

Composable CDP vendors (Hightouch, Census, GrowthLoop) face structural pressure. These companies built their value proposition on being the activation layer between the warehouse and marketing tools — the position CustomerLake now occupies natively. HP describes its prior approach as composable and projects up to 40% lower CDP total cost of ownership after consolidating; the Philadelphia Union replaced its previous CDP outright. Two named displacements in one announcement is the strongest signal yet that the composable layer is contestable where the customer already runs Databricks. As Niederberger observes, composable vendors won the argument so thoroughly that “the warehouse ended up with a fully paved on-ramp to build its own front door.” That said, composable vendors have mature integrations, established customer bases, and multi-warehouse support (Snowflake, BigQuery, Redshift) — switching costs are real, and Beta status limits near-term displacement.

Packaged CDP vendors (Segment, mParticle, Lytics) face a different challenge. CustomerLake does not replicate their full feature sets (particularly native messaging and event SDKs), but it undermines the data-copy model that packaged CDPs require. For organizations already on Databricks, the value proposition of copying customer data into a separate platform weakens when the lakehouse itself offers identity resolution, segmentation, and activation.

Enterprise suite CDPs (Adobe AEP, Salesforce Data Cloud) are less directly threatened but not immune. Their CDP capabilities are embedded within broader marketing clouds that include content management, journey orchestration, and cross-channel delivery — layers CustomerLake does not attempt to replace. However, CustomerLake’s Lakehouse Federation (which can query Snowflake, BigQuery, and cloud storage without copying data) challenges the data gravity that suite vendors rely on. Notably, Adobe is a CustomerLake launch partner — suggesting coexistence rather than displacement in the near term.

The market-structure argument is now coming from analysts. CMSWire reported in October 2026 that Gartner expects 80% of net-new enterprise CDP deployments by 2030 to be embedded in or composable with data platforms — a forecast that treats this purchase as an infrastructure decision rather than a marketing-software one. David Raab, founder of the CDP Institute, framed the trade-off: “It will simplify life for IT and marketing teams — and be a challenge for many conventional CDP vendors.” He also warned that “IT gains more control, which marketers may not appreciate.” Zé Delivery’s result speaks to that warning: its CRM team builds audiences inside a platform governed by Unity Catalog.

Adweek characterized CustomerLake as “an agentic challenger to traditional CDPs” — framing it as a category disruptor rather than a feature addition.

Alternatives to Databricks CustomerLake

Alternatives fall into three groups, and the right one depends on where your customer data already lives and whether activation has to close inside one platform.

If your data sits in Snowflake or BigQuery, or you run more than one warehouse, composable tools — Hightouch, Census (now Fivetran Activations), and GrowthLoop — keep the warehouse as the system of record and add activation on top. If you need the Customer Intelligence Loop closed inside one vendor boundary, with real-time profiles, native messaging, and AI decisioning, an agentic CDP such as Treasure AI’s Complete CDP covers stages CustomerLake leaves to partners. If you need client-side event collection and a mature marketer UX, packaged CDPs such as Twilio Segment and mParticle still own that ground.

The trade-off is explicit. CustomerLake removes data movement and some licensing overhead for Databricks shops, but it adds dependency on Databricks’ roadmap, its consumption pricing, and partner-delivered messaging. A composable stack costs more in integration work and duplicated PII, but it keeps multi-warehouse portability. Migration in either direction is a multi-month project once identity graphs are involved.

For a comprehensive comparison of CDP vendors across all categories, see the CDP Vendor Comparison Guide. For evaluation criteria specific to AI-era requirements, see How to Evaluate a CDP in the AI Era.

→ Compare all CDP vendors side-by-side in the CDP Vendor Comparison Guide

See how independent analysts evaluate CDP vendors — download the Forrester Wave report for a side-by-side comparison.

What the Beta Proves — and What It Does Not

The October release answers several questions that were open in June. Audience assembly on an existing governed lakehouse can be measured in hours rather than months. Named enterprises are running Campaign Agents against live customer data, not just in demos. Profile Agents and Campaign Agents can be adopted separately, so a buyer does not have to take the whole stack at once. And the marketer-side promise has at least one public test: Zé Delivery’s CRM team works without engineering support.

Other questions remain open, and several of them are the ones that decide enterprise renewals:

• General availability — Databricks has announced no GA date. Forrester’s Joe Stanhope expected GA later in 2026 (Forrester, June 2026); the Beta is a step toward it, not a commitment
• Pricing and cost predictability — no rate card exists, and consumption pricing means costs scale with agent compute. HP’s 40% projection is the only public cost signal, and it is a projection about one company’s estate. When Campaign Agents run continuously, forecasting monthly spend is harder than with per-seat or per-profile licensing
• Real-time profile serving — the product page names a Real-Time Profile API, but Databricks has published no documentation, specification, or latency figures. Published deployments sync on daily or monthly schedules and update profiles in micro-batches. CMSWire’s read is that this shows daily and micro-batch updates rather than real-time activation — the gap Treasure AI has argued decides the orchestration layer (CMSWire, October 2026). Whether that matters depends on the use case: batch campaign selection tolerates a daily sync, in-session personalization does not
• Event collection SDKs — CustomerLake’s ingestion relies on Lakeflow and existing Databricks pipelines. Whether it will offer client-side SDKs for web and mobile event collection (comparable to Segment’s analytics.js or mParticle’s SDK) has not been announced
• Marketer UX at scale — the evidence is customer-reported and vendor-selected. One CRM team working without engineering support is encouraging; it is not a peer-reviewed usability benchmark against purpose-built journey builders
• Migration complexity — switching identity resolution providers or moving from an existing CDP’s identity graph to Agentic Identity Resolution is a multi-month effort. Mid-contract transitions are operationally risky
• Multi-cloud neutrality — CustomerLake runs on Databricks. Organizations with customer data split across Snowflake and Databricks can use Lakehouse Federation, but the control plane, governance, and billing flow through Databricks. As Niederberger notes, federation “still runs it across Databricks’ control plane, its governance, its bill, and its roadmap”

Four things belong in writing before a pilot: a GA date or contractual milestone, a compute estimate at your profile volume and campaign cadence, a reference customer in your industry and data-volume band, and a latency specification for the Real-Time Profile API. If any is missing, scope the pilot to campaign selection — the use case the published deployments actually demonstrate — rather than in-session personalization.

These open questions are the material for an evaluation, not a reason to wait. Teams weighing a Beta platform against an established one can use the same scored criteria across every candidate — marketer UX, SDK coverage, real-time serving, GA timeline, and cost predictability are answerable today even where a vendor’s roadmap is not.

Sources: Databricks press release (June 16, 2026), Databricks launch blog (June 16, 2026), CustomerLake enters Beta (Databricks, October 19, 2026), How HP Is Building an AI-Ready CDP (Databricks, October 20, 2026), CMSWire (October 2026), Adweek, Martech Therapy (Matthew Niederberger).

Related Articles

• What Is Hightouch? — The composable CDP CustomerLake competes with most directly
• What Is GrowthLoop? — The marketer-led composable CDP compared
• CDP Pricing: Models, Ranges, and Hidden Costs to Budget For — How consumption pricing compares with CDP licensing
• Customer 360 in the AI Era — What a unified profile has to do for agents
• Reverse ETL vs CDP: What Reverse ETL Can and Cannot Do — The layer CustomerLake absorbs

FAQ

What is Databricks CustomerLake?

Databricks CustomerLake is an agentic CDP embedded in the Databricks lakehouse, with Campaign Agents in Beta and Profile Agents in Private Preview as of October 19, 2026. It combines Customer 360 profiles, identity resolution, audience building, and partner-based activation on governed lakehouse data. Databricks has announced no general availability date and no published rate card; access runs through its account teams.

Is Databricks a CDP now?

Yes — with CustomerLake, Databricks is now a CDP, specifically an Agentic CDP. CustomerLake adds identity resolution, audience segmentation, campaign automation, and multi-channel activation inside the lakehouse rather than in a separate system that ingests copies of customer data. The caveat is maturity: Campaign Agents reached Beta in October 2026, Profile Agents remain in Private Preview, and neither is generally available.

Is Databricks CustomerLake generally available?

No — CustomerLake is not generally available. Campaign Agents entered Beta on October 19, 2026 and Profile Agents remain in Private Preview. Databricks has announced no GA date, no published rate card, and no published Beta eligibility criteria; companies that want access go through their Databricks account team. Treat any 2026 engagement as a pilot with a defined exit.

What results have CustomerLake customers reported?

The first named results are customer-reported: HP says it was audience-ready in about two hours versus two to three months and projects up to 40% lower CDP total cost of ownership; the Philadelphia Union reports a 90% cut in audience turnaround. Wine.com launched production campaigns 23 days after kickoff. None of the figures is independently audited, so read them as existence proofs rather than benchmarks.

How does CustomerLake compare to packaged CDPs?

CustomerLake covers most core CDP capabilities but takes a different architectural approach. Packaged CDPs copy customer data into a proprietary store and build identity, segmentation, and activation on top. CustomerLake operates on the lakehouse under Unity Catalog. It provides identity resolution, Genie-based self-service, and partner activation, but no native email, SMS, or push — organizations needing built-in delivery still require a messaging platform.

What is an Infinity campaign?

An Infinity campaign is Databricks’ term for a continuously running, agent-driven engagement loop that replaces static one-off campaigns. Campaign Agents analyze customer signals, recommend next-best actions, and optimize engagement in real time instead of running discrete campaign cycles. Databricks describes the October beta results as building blocks toward that vision; continuous autonomous optimization remains a roadmap claim rather than a shipped capability.

Is Databricks CustomerLake the right CDP for my organization?

CustomerLake is the right CDP when you already run the Databricks lakehouse and want CDP capabilities without moving data — and the wrong fit today for anything requiring a generally available, real-time platform. Choose it if zero data movement and Unity Catalog governance outweigh Beta status. Skip it if you are not on Databricks, need native messaging, or require sub-second profile lookups.

What are the alternatives to Databricks CustomerLake?

The main alternatives are composable tools such as Hightouch, Census, and GrowthLoop, agentic CDPs with native messaging such as Treasure AI, and packaged CDPs such as Twilio Segment and mParticle. Which fits depends on where your data lives and whether activation must close inside one platform. Organizations not on Databricks, or needing multi-warehouse portability, usually keep a composable or packaged CDP.

Should I replace my CDP with Databricks CustomerLake?

Replace only if Databricks is already your primary data platform and the CDP you run today duplicates pipelines it could absorb. HP and the Philadelphia Union describe exactly that consolidation, but both are large enterprises with mature lakehouse teams. If you depend on native messaging, client-side event SDKs, or real-time profile lookups, keep your current CDP and run CustomerLake alongside it.