Databricks Puts CRM Teams in Charge of Audience Building

Databricks’ October 19 CustomerLake update lets eligible enterprise CRM teams test whether AI removes engineering handoffs from audience building. Smaller marketing teams can use their existing tools to test who can change a customer group and get it into the campaign system.

Campaign Agents, software that helps build and activate those customer groups, has moved to Beta for more eligible customers. Profile Agents, which prepares unified customer records, is entering Private Preview. Teams can adopt the components separately; access still runs through a Databricks account team.

At Zé Delivery, the director of data and analytics, Bruno Smarsaro Bazelato, describes the campaign work.

“Our CRM team now builds audiences in CustomerLake and syncs them directly to Braze without engineering support”.

That statement puts a useful boundary around the promise. A marketer can own an audience change while the data team remains responsible for the information the campaign uses. The purchasing question is how much routine work crosses that boundary after setup.

Follow one customer group into the campaign

Its product description says Genie lets marketers describe audiences in ordinary language using governed customer data. Unity Catalog supplies shared access controls and definitions. That arrangement gives the marketing interface a controlled foundation instead of asking each marketer to assemble a separate customer database.

Consider a proposed trial for a returning-customer email. The marketer specifies which purchase history qualifies a customer and chooses the intended destination. During the trial, ask them to change the qualifying window and inspect the resulting group. This is an acceptance test, not a reported customer incident.

Watch who performs the change. If the team still needs an engineer to interpret each request, the promised independence hasn’t been demonstrated in that workflow. If the marketer can make the change, check whether the resulting customer records match the agreed definition before any email goes out.

That review deserves time in the trial. A natural-language request can sound clear while leaving a business choice unstated. Specify how returns affect purchase history, for example, rather than letting a generated segment decide the campaign’s meaning. The team should approve that definition before comparing speed.

Our Hightouch and Strata coverage examined the work of maintaining a connection after its first successful transfer. CustomerLake raises a related question about campaign ownership: once the connection works, which changes can marketing make without reopening an engineering request?

The campaign destination still matters

CustomerLake was first announced in Private Preview on June 16. The October milestone expands access and adds named customer accounts. It doesn’t make the platform generally available.

The Braze connection also predates this week’s announcement. In its June 16 integration account, updated August 1, Braze described moving governed audiences into its platform and returning engagement data to Databricks. Its chief revenue officer, Ed McDonnell, said:

“CustomerLake represents a real shift in how brands can act on their data”

Braze sells the downstream engagement platform. Its statement helps establish the intended workflow, while its commercial interest limits its value as independent evidence of savings.

For a trial, follow the chosen customer group through to that destination. Compare expected membership with the records the campaign tool actually receives. Assign responsibility for resolving any difference before judging how many engineering requests the new process removes.

The work a campaign manager cares about ends with a usable audience in the sending tool. A convincing audience-building demonstration should therefore include the destination, with a person able to inspect what arrived.

Separate a customer account from a benchmark

Zé Delivery’s quotation is a vendor-selected customer account. It doesn’t quantify initial setup effort or establish that other companies will achieve the same degree of independence. Eligibility and the condition of an employer’s existing data will affect what that employer can test.

The strongest case for this approach is narrower and practical: teams with a governed data foundation can reuse it for campaign work. The unresolved buying question is what preparation and ongoing support their own workflow requires. Neither a vendor demonstration nor a customer’s testimonial answers that for them.

Keep those costs visible alongside the review work described in our Basis survey analysis. For the next marketing-stack evaluation, ask the intended campaign owner to make one real audience change using approved data, then verify the destination records together. Record every technical handoff needed to finish. Use that completed workflow to decide whether control has actually moved to the marketer.