Databricks Says Internal AI Assistant Tripled Marketers' Use of Data
Databricks says its Marge AI analytics assistant, built on Genie Agents, raised marketing data use 3x with over 85% adoption internally.
Quick answer
What did Databricks announce about its internal marketing analytics assistant Marge?
Databricks says its marketing team built Marge, a Genie Agents-based conversational analytics assistant, on a governed Marketing Lakehouse. The company reports marketers now use data three times more often in decisions, adoption exceeds 85% of the marketing organization, and flagged incorrect responses have dropped 25%.
Key takeaways
- Databricks says it built Marge, a conversational analytics assistant powered by Genie Agents, on top of a governed Marketing Lakehouse shared across marketing, finance, sales and product data.
- The company reports marketers now use data three times more often in decisions, with adoption exceeding 85% of the marketing organization and usage growing 50% quarter over quarter.
- Databricks says Marge handles more than 800 questions per month, has answered over 5,000 total, and flagged incorrect responses have decreased 25% as the system improved.
- According to Databricks, the assistant relies on Unity Catalog for governance, lineage and role-based access control, and on verified metric definitions, example queries and trusted assets reviewed by domain experts for accuracy.
- Databricks says the rollout began with a single use case, email campaign performance, before expanding, and that one BI manager spends about an hour a week maintaining the system.
Databricks says its marketing organization built an internal AI analytics assistant called Marge that has tripled how often marketers use data to make decisions.
According to the company, Marge is a conversational analytics tool built with Genie Agents, grounded in a governed "Marketing Lakehouse" that unifies data from campaign platforms, web analytics, CRM systems, event tools, advertising channels and sales data. Databricks says marketers can ask questions in plain English, such as how an email campaign performed or which programs influenced pipeline in a given quarter, and receive answers generated from governed enterprise data.
Reported results
Databricks reports the following outcomes from deploying Marge internally:
- Marketers use data three times more often in decisions
- More than 85% of the marketing organization uses Marge
- Usage has grown 50% quarter over quarter
- Marge handles more than 800 questions per month and has answered over 5,000 in total
- Flagged incorrect responses have decreased 25% as the system improved
- The company scaled access to insights without adding analytics headcount
How Databricks says it built accuracy and trust
Databricks says it used Unity Catalog to centralize metadata, table descriptions, column annotations, lineage and access controls, so that each user sees only the data they are authorized to access. The company says marketing stakeholders and data experts reviewed and enriched AI-generated descriptions with business context, such as the precise meaning of a marketing-qualified lead and how campaigns map to products and regions.
The company also says it encoded verified answers and example queries for common, high-value questions covering areas like conversion rates, customer lifetime value and event registration, reviewed by domain experts. Databricks says it gave Marge behavioral guidance for interpreting business terminology, such as recognizing that "spend" and "investment" refer to the same field labeled "cost," and instructed the assistant to ask clarifying questions when a request is missing information like time period, channel or region.
According to Databricks, every response allows users to submit positive or negative feedback, which the marketing analytics team reviews through a monitoring dashboard, alongside benchmark questions run against known answers to evaluate performance systematically. The company says this oversight is lightweight, with one BI manager spending approximately one hour per week reviewing feedback and maintaining Marge.
How Databricks says it drove adoption
Databricks says it worked directly with marketers to understand the questions they asked most often and the language they used to describe their work, using that input to shape the data, examples and instructions configured into Marge. The company says it began with a single, narrow use case, email campaign performance, using only the campaign, recipient and engagement data needed to answer those questions, before expanding to include additional data such as account information.
Marketers access Marge through Genie One, which Databricks describes as an AI cowork experience for business users that combines dashboards, Genie Agents, apps and deeper analysis, automatically routing requests to the appropriate agent. The company says it also created focused agents for domains such as web performance, digital analytics and marketing planning, and that an "Agent mode" can evaluate multiple steps for more complex questions.
Source: Databricks Blog, "How Databricks' marketers use data 3x more with Genie, an AI analytics assistant," published September 15, 2026.
Frequently asked questions
- What is Marge?
- Databricks says Marge is its internal conversational analytics assistant for the marketing team, built using Genie Agents and grounded in the company's Marketing Lakehouse.
- What results does Databricks report from using Marge?
- Databricks says marketers use data three times more often in decisions, more than 85% of the marketing organization uses Marge, and flagged incorrect responses have fallen 25%.
- How does Databricks say it kept Marge's answers accurate?
- According to Databricks, the team documented data relationships in Unity Catalog, added verified answers and example queries reviewed by domain experts, defined business terminology, and ran continuous feedback and benchmark evaluations.
- How did Databricks roll out Marge to marketers?
- Databricks says it started with a single use case, email campaign performance, involved marketers in shaping the assistant's data and language, and later expanded to additional data domains.