HelloFresh reimagines marketing with Genie on Databricks

HelloFresh's paid marketing teams once lost a full week per cycle just reconciling ad spend numbers across media, creative and finance tools. Underperforming creatives also stayed live on Meta for roughly 25 days. To fix that, HelloFresh built an internal agentic marketing platform on Databricks. It combines Genie Agents, Lakebase, Databricks Apps, Unity Gateway and Unity Catalog to unify data, agents and human approvals into one governed system. The result: ad testing that's now 4x faster and creative launches that happen in a single click.

Challenges of fragmented marketing tools and lengthy creative cycles

As HelloFresh’s paid marketing organization scaled, team members naturally relied on distinct, specialized platforms to manage their day-to-day work: media planners in Meta ad spend dashboards, creative teams in Figma and Jira and finance analysts tracking performance in Tableau and Google Sheets.

With performance insights distributed across these domain-specific tools, aligning on a unified set of metrics required significant manual effort every cycle. Teams spent valuable time reconciling figures and cross-referencing datasets before major marketing decisions could be finalized.

This operational disconnect created extended lead times throughout the campaign lifecycle. A media buyer might spend a full day assembling a weekly business review before briefing the creative team on a new ad concept, and then lead a multi-week production cycle.

Once the new creative assets were ready, deploying them involved a specialized offshore team of seven people plus a manager, along with days of coordination to launch across markets. Meanwhile, identifying and rotating out underperforming ads on platforms like Meta could take up to 25 days.

As Alan Khalil, Associate Director of Data Science and Engineering for paid marketing at HelloFresh, put it: "We need to safely and quickly turn data and evidence into automated actions."

Building an agentic marketing platform with Databricks

To address these challenges, HelloFresh built an internal agentic marketing platform natively on Databricks. The goal was to unify performance data, automate actions and keep humans in the loop with governed approvals.

Genie Agents provide a unified view of performance data

Genie Agents gives media, creative and finance teams governed, code-free access to a single governed view of performance data. It replaces the separate dashboards and spreadsheets that used to force teams into reconciliation mode.

Underneath, Lakebase (Postgres) acts as the state store. It manages agent conversations, workflow runs, human approvals and the creative asset library. Databricks Apps expose that same context as human-facing UIs and agent-hosted experiences.

Unity Catalog governs access end-to-end — across data, apps and Genie Agents. That way, every persona works from the same governed source of truth instead of a private copy.

Purpose-built agents enhance marketing efficiency

The platform runs as a chain of purpose-built agents. A Genie Agent continuously hunts for performance signals. A data-scientist agent forecasts issues such as ad fatigue. An analyst agent takes action — pausing underperforming ads via the Meta API and launching replacements from the creative library — and a supervisor agent reports those actions back to a human for approval.

A creative strategist agent watches ad performance alongside creative elements to guide strategy. A "council of models" approach, enabled through Unity Gateway, lets one model generate creative while several others independently QA it for issues like a missing call-to-action.

“Unity Gateway gives us the flexibility to use the best model for each job without having to manage separate integrations for every provider. We can bring new models into the platform quickly and make them available across our applications and agents,” said Alan.

Transforming reconciliation and creative launches with Genie

The implementation of the agentic marketing platform led to clear improvements in speed and efficiency across ad performance, creative optimization and cross-functional collaboration.

• 4x faster ad testing: The closed loop between creative optimization and media-buying automation. "Increased our testing velocity, our ability to test how fast we can test ads in Meta by 4x," Alan said. Ads that used to stay live on Meta for weeks, even while underperforming, now get caught and replaced far sooner: "We used to have ads live on Meta for 25 days. Now, we're down to six days." 

• Reconciliation cut to under an hour: Decision cycles that used to consume a full week across media, creative and finance teams now take less than an hour on Monday mornings. Everyone is querying the same Genie Agent rather than using separate tools with separate definitions. 

• Creative launches now happen in one click: "This used to be a team of seven people and used to be days of back and forth Slack messages before we could go from, a creative is ready, can we launch it please in this market, to now it happens in a message or happens in a few clicks," Alan said.

The platform now covers 8 brands across 16 markets and has scaled from optimizing roughly 65% of HelloFresh's Meta ad spend to 100%. Because the system is deployed as a Databricks App, it runs entirely on serverless infrastructure, and the HelloFresh engineering team no longer needs deep Kubernetes expertise to keep the apps up and running. That frees them to focus on improving agent quality and solving new marketing use cases, while other functions across the business have begun standing up their own Genie Agents.