HP CustomerLake Brings the CDP to Its Existing Data Foundation

HP says it became ready to build marketing audiences in two hours with CustomerLake, versus an expected two to three months. The HP CustomerLake story centers on reusing an existing data foundation for a customer data platform, or CDP, which organizes customer information for marketing. The tension is whether that reuse translates into better daily execution, beyond an encouraging setup session.

Databricks published the HP marketing architecture case study on October 20, drawing on Luis Alonso’s presentation at Data + AI Summit 2026. Alonso leads HP’s data strategy and engineering for marketing technology and AI enablement. HP also estimates potential CDP ownership savings of up to 40% against its former composable approach.

The comparison matters because composable CDPs, which connect marketing capabilities to existing data infrastructure, already promise less duplication. Hightouch represents that alternative. HP’s choice raises a more specific question: should marketing intelligence sit inside the data platform or in a specialized application connected to it?

HP CustomerLake Starts With Work Already Done

The reported setup gain concerns readiness to build audiences, rather than a completed transformation of marketing performance.

That distinction gives the announcement its proper scale. Configuring available information for a marketing application is an operational milestone. Proving that campaigns reach suitable customers, respect preferences, and improve results requires additional evidence. A short setup window cannot establish all those outcomes.

HP’s account describes CustomerLake using existing schemas, data models, and governance. Schemas define how information is organized; governance determines who can use it and under what conditions. Reusing those elements offers a plausible explanation for the reported speed.

The broader inference is that implementation time depends partly on where earlier preparation happened. A team with approved datasets and agreed definitions starts from a different position than one still reconciling customer records. Comparing those teams through installation time alone would hide the difference.

Consider the work behind a seemingly simple request to contact likely buyers. Someone must define the eligible population, identify exclusions, choose relevant attributes, and determine permitted destinations. A marketing interface can simplify that request without eliminating its underlying decisions.

When those decisions already exist in a usable form, software can inherit their value. When they remain unresolved, an interface may expose the gaps more quickly. Either way, the quality of the starting foundation matters to the result.

This makes HP’s experience useful as an architectural example. It suggests an enterprise can recover more value from prior data investments by connecting them directly to operational work. It does not establish a universal deployment timetable.

Buyers should therefore read “audience-ready” as a defined milestone with boundaries. They should ask what information was already available, which approvals preceded setup, and what remained before campaign launch. Those details determine whether the comparison applies to their organization.

The same distinction affects how teams evaluate HP marketing data. Availability means a dataset can be reached. Usability means a marketer can interpret it correctly and apply it to an approved purpose. Those conditions deserve separate evaluation.

An audience builder might show a field immediately while leaving its business meaning unclear. A field called “active customer” could describe recent purchasing, product use, or an account relationship. Each interpretation would produce a different audience.

The story’s practical significance comes from shortening the distance between prepared information and a usable marketing capability. That is valuable even before revenue effects appear. It also creates a clearer obligation to measure what happens after setup.

Marketing and Data Teams Must Agree on the Same Definitions

Moving marketing work closer to enterprise data shifts pressure toward shared ownership of business meaning.

A central platform does not automatically create agreement. Marketing, sales, and finance can read identical records while interpreting them differently. If an AI assistant answers their questions, those differences become part of the product’s behavior.

A semantic layer, which records business definitions and relationships, can make those interpretations explicit. For example, an organization might define which purchases qualify for a revenue measure. The definition must also specify exclusions, timing, and ownership.

The important organizational question is who can change that definition. Giving every team independent control encourages divergence. Requiring central approval for every adjustment risks rebuilding the same queue that self-service tools are supposed to reduce.

A workable division gives domain specialists responsibility for meaning and platform teams responsibility for dependable access. Marketing operations then owns how approved information enters campaigns. This is an analytical model for evaluating the architecture, rather than a claim about HP’s internal reporting lines.

Under that model, engineering work changes shape. Engineers spend less effort translating recurring requests when marketers can safely reuse approved definitions. They still need to manage reliability, permissions, changes, and exceptions.

Marketing leaders face a corresponding responsibility. Self-service requires them to define acceptable audiences and outcomes precisely enough for software to apply those choices. Asking an agent for “better prospects” leaves the central business decision unresolved.

Genie, Databricks’ natural-language analytics interface, illustrates how that responsibility reaches the user experience. A conversational answer can sound clear even when the underlying business question is ambiguous. The interface needs context that distinguishes possible interpretations.

Databricks’ guidance on Genie evaluation describes benchmark questions and a review process for checking answers. These mechanisms treat generated analysis as something to evaluate and correct. They do not make every answer reliable by default.

For enterprise buyers, that creates a more useful question than whether an assistant supports ordinary language. Can the organization explain why a particular answer is correct? Can a user identify who approved the definition behind it?

A proposed audience should carry an understandable explanation of its selection criteria. A performance answer should expose the measurement choices that produced it. Otherwise, self-service merely moves interpretation work from analysts to people with less visibility.

This has a connection to everyday knowledge management. Information becomes useful when people can recover its meaning, source, and context. Marketing automation adds the consequence that an interpretation can trigger customer-facing action.

For HP marketing data, the architectural promise is therefore inseparable from stewardship. Reusing a definition only helps if someone keeps it accurate. The platform’s convenience increases the importance of that ongoing responsibility.

Embedded Marketing and Composable CDPs Share a Starting Point

The meaningful competition concerns who owns the marketing workflow after customer data has become accessible.

Databricks introduced CustomerLake on June 16, 2026. Its agentic CDP architecture places customer profiles, audience creation, and activation within its governed data environment. Agentic software uses AI systems to pursue tasks through multiple steps.

Databricks describes Profile Agents for preparing customer information and Campaign Agents for audience and campaign work. Unity Catalog, its governance service, supplies the surrounding control framework. These are product descriptions from Databricks, rather than independently established performance advantages.

The embedded argument is straightforward. If business information and its controls already live in the platform, marketing capabilities can use that foundation directly. Fewer translation boundaries should mean fewer places to recreate definitions or troubleshoot inconsistencies.

The specialized application argument starts from a different ownership decision. An independent marketing layer can connect to enterprise data while concentrating on campaign operations. Its appeal depends on the quality of those workflows and the breadth of systems it connects.

Hightouch’s June 15 agentic marketing strategy argues for customer-owned context and intelligence across platforms. It describes agents using customer information alongside campaign, creative, and strategic context. That position challenges any assumption that useful context belongs entirely inside a data platform.

The disagreement is narrower than a choice between copying everything and copying nothing. Both vendors argue for working with existing enterprise information. The distinction concerns where additional marketing intelligence accumulates and how teams use it.

A company evaluating HP CustomerLake should therefore look beyond storage diagrams. Audience review, campaign coordination, destination support, experimentation, and error recovery all affect daily operations. A shared foundation matters most when it improves those tasks.

For example, an audience can be valid in the source system but rejected by a destination. The operational question becomes who diagnoses the rejection and where the correction occurs. Architecture should clarify that responsibility.

Similarly, a campaign can use the correct customer records while conflicting with another campaign’s schedule. Solving that problem requires visibility into marketing activity. Access to accurate tables alone does not settle the coordination problem.

An embedded product has to show that it understands the work surrounding its data. A specialized application has to show that its added capabilities justify maintaining another operational boundary. Each approach carries an obligation to demonstrate practical value.

That creates pressure on established CDP suppliers without establishing a winner. Vendors must explain which responsibilities they remove, which they retain, and which remain with the customer. Labels such as “native” or “composable” cannot answer those questions.

The Databricks agentic CDP strategy is consequently an expansion of the platform’s role. It asks buyers to entrust more customer-facing workflow to their data provider. Whether that suits an enterprise depends on its existing systems and operating priorities.

Reusing Data Context Explains the Speed Advantage

The strongest explanation for faster setup is avoided reconstruction of existing work.

A marketing capability needs more than a connection to records. It must know which fields belong together, which relationships are valid, and which definitions users expect. Rebuilding those decisions inside another application creates an additional maintenance obligation.

Imagine an approved business-account model that connects organizations, contacts, and purchases. If a marketing tool requires a separate representation, someone must translate that model. Later changes must reach both versions without changing their meaning.

An embedded capability can reduce that translation when it accepts the existing model directly. The benefit comes from reuse across the boundary between data preparation and business action. It should be evaluated through the work actually removed.

That mechanism also explains why the same product can produce different implementation experiences. A prepared organization may spend its initial session selecting available assets. Another may discover that the necessary relationships remain undocumented or unreliable.

Neither experience alone measures the software’s full quality. The first highlights compatibility with prior preparation. The second exposes unfinished work that any responsible campaign system would need to address.

The distinction matters especially for AI-generated audiences. A marketer’s request must become explicit selection logic before it can be reviewed. The system needs enough context to translate business language without silently changing the request.

Take a hypothetical request to contact business buyers who are considering equipment upgrades. The system must distinguish existing customers from prospects and identify what counts as relevant interest. It must also preserve exclusions that the marketer may not repeat in every prompt.

Those requirements explain why technical access and business context belong in the same evaluation. A system can access every permitted record and still select an unsuitable audience. Clear definitions reduce ambiguity, while review catches mistakes that definitions do not prevent.

Identity resolution, which links records believed to describe the same person or organization, adds another judgment. A match can connect useful history, but an incorrect match can contaminate an audience. More connected data therefore needs quality checks alongside broader access.

HP’s reported direction includes bringing identity signals into its governed environment. The architectural implication is that identity can support measurement and activation through shared rules. The report does not establish that every identity challenge has been resolved.

For buyers, the relevant evidence would include how changes propagate. If an account relationship changes, does the audience update consistently? If a record loses eligibility, can the organization trace where that record was previously sent?

These questions extend the setup story into operations. They reveal whether reused context remains useful as information changes. Static compatibility is a starting condition; dependable maintenance determines whether the advantage lasts.

The same reasoning applies to the HP CustomerLake cost case. Avoiding duplicate maintenance creates a plausible source of savings. The amount depends on which systems, integrations, and support tasks the organization can actually retire.

Setup Speed Leaves Cost and Campaign Outcomes Open

An implementation result, a financial estimate, and a marketing outcome answer different questions.

The evidence should remain separated because each category has its own comparison. Setup measures the effort required to reach a defined operational state. Ownership cost concerns resources consumed over time. Campaign performance concerns changes in customer behavior.

Converting an estimate into a settled result would obscure that distinction. A projected saving can guide investment decisions while still depending on future adoption and retirement of older systems. Its usefulness does not require presenting it as already realized.

A complete ownership comparison should include work that moves between teams. Removing a marketing application may reduce one budget while increasing shared platform responsibilities. The enterprise benefits only when the overall operating burden improves.

That does not undermine the consolidation argument. It identifies what a serious financial assessment should capture. Buyers need consistent boundaries around support, development, infrastructure use, migration, and ongoing administration.

Campaign outcomes require a different standard. An audience assembled quickly can still produce weak responses. A more elaborate customer profile can still contain irrelevant signals. Efficiency and effectiveness should be measured together without treating either as a substitute.

Incrementality, which measures results caused by an intervention beyond what would otherwise happen, offers a useful frame. A campaign’s reported responses do not automatically establish its added business value. Comparisons need a credible baseline and an appropriate observation period.

The public HP account does not supply an independent campaign-impact evaluation. That leaves the commercial payoff open while preserving the reported implementation evidence. Readers should resist filling the gap with outcomes drawn from another organization.

Product maturity adds another boundary. Databricks’ October 19 CustomerLake beta update places Campaign Agents in beta and Profile Agents in private preview. The company says customers can adopt the capabilities independently.

Those statuses mean the Databricks agentic CDP should be assessed by capability. An enterprise may have access to campaign functionality while a separate profile capability remains restricted. A general product label does not describe every deployment condition.

The update includes customer examples, but vendor-selected accounts remain vendor-selected evidence. They help identify possible use cases and questions for reference calls. They cannot establish that another enterprise will obtain equivalent results.

There is also an operational risk in concentrating responsibilities. When more workflows depend on a shared platform, change management becomes more consequential. Buyers should ask how they can inspect decisions, reverse actions, and recover from failures.

This is not evidence that HP has encountered those problems. It is the appropriate stress test for the architecture’s claims. A credible analysis distinguishes reported achievements from the questions that remain unanswered.

The resulting judgment is bounded but useful. HP offers evidence that prepared data infrastructure can shorten the path to audience-building readiness. The broader claim requires continued evidence about operation, cost, and customer outcomes.

The Next Evidence Should Follow Adoption, Cost, and Results

The most useful follow-up will connect broader product access to repeatable operations and measurable business value.

Over the coming months, the first signal is capability-specific adoption. Watch for clearer eligibility, release documentation, and customer accounts describing sustained campaign work. Wider access would strengthen the case that the product extends beyond closely supported early deployments.

The strongest accounts would explain what happens after the initial audience is built. Who maintains the definitions? How do marketers handle rejected deliveries? What changes still require engineering involvement? Those details reveal whether the operating model holds up.

Evidence that ordinary users can manage recurring work would support the embedded approach. Evidence of extensive hidden customization would narrow its applicability. Either outcome would be more informative than another isolated setup demonstration.

The second signal is realized ownership cost under a clearly stated comparison. Follow-up reporting should distinguish projected savings from observed changes and identify what the organization stopped operating. It should also acknowledge responsibilities absorbed by the shared platform.

A transparent comparison would strengthen the consolidation case even if results varied across businesses. Continued estimates without defined boundaries would leave the financial argument provisional. Buyers need enough detail to compare the claim with their own architecture.

The third signal is campaign effectiveness using a credible baseline. Look for customer-specific reporting that separates audience preparation from actual engagement and business outcomes. Evidence should explain the campaign context rather than presenting an isolated percentage.

A faster workflow becomes commercially persuasive when teams can connect it to useful action. That might mean more timely communication, fewer unsuitable contacts, or improved experimental results. Each claim should have an explicit measure and a suitable comparison.

Competitive responses will provide supporting context for these signals. Specialized vendors can answer through better workflow integration, clearer portability, or stronger operational evidence. Data-platform vendors can answer by showing that shared context improves more than initial configuration.

For developers, the practical issue is how much application-specific logic still needs to be maintained. For enterprise buyers, it is whether consolidation reduces the full workload. For marketers, it is whether greater independence comes with understandable controls.

HP CustomerLake is worth following because it connects an architectural choice to a concrete implementation milestone. Its wider significance depends on whether that choice keeps delivering value once the initial configuration is complete.

Before treating the case as a purchasing shortcut, compare its starting conditions with your own. Identify which customer definitions are approved, which teams maintain them, and which campaign destinations still require separate integration. Ask vendors to trace a representative audience from the original business request through delivery and measurement. Require them to explain exceptions, maintenance, and the ownership of every unresolved step. Then separate the evidence for setup speed, operating cost, and marketing impact. HP CustomerLake provides a reason to examine embedded marketing more closely. Use that reason to demand a clearer demonstration of the work your organization needs done.