Agent Bricks in Action: Smarter Offer Targeting for Sales Teams

From Sales Dilemma to Data-Driven Action

Even the best commercial offers are only as effective as their delivery. At Databricks, we provide free credit offers to help customers get started or accelerate adoption, but sales representatives face a deceptively simple question: which of my customer accounts are eligible, and which should I reach out to first?

What seems like a straightforward task can be opaque and quickly turn into a time-consuming, multi-team effort, especially when accounts are unexpectedly ineligible for offers. Sales teams often need to dig through documentation, consult Slack threads, and manually investigate accounts with operations teams. This creates unnecessary back-and-forth, slows down momentum, and gets in the way of providing customers with high-value offers. Even when accounts are known to be eligible, it’s not always obvious which should be prioritized.

Building a Smarter System with Agent Bricks

To tackle the problem, our team turned to Agent Bricks — Databricks’ platform for building high-quality AI agents on enterprise data — and built a multi-agent system that delivers clear, actionable guidance directly to sales teams. In less than two days, I created a tool that lets sales reps:

• Quickly identify which customer accounts qualify for credit offers
• Understand the exact reason an account isn’t eligible
• Rank eligible accounts to focus on the highest-impact prospects first

As an intern in Business Strategy and Operations this summer, I had a short turnaround time, so speed and simplicity were critical. Agent Bricks let me quickly build a high-quality solution and provide the enablement sales teams needed.

Designing the Multi-Agent Solution

Using Agent Bricks’ Multi-Agent Supervisor, I designed a system that chains together three purpose-built agents under one supervisor. Like an air-traffic controller, the Supervisor decides which agent to delegate each part of the question to and then stitches their responses into one clear answer.

One Supervisor, Three Specialized Agents

My solution uses three agents: two AI/BI Genie agents and a Knowledge Assistant agent, managed by a supervisor to orchestrate tasks and information flow:

Without needing to research supervisor agent architecture or engage with technical teams, I was able to build a functional AI agent system directly on our customer data and offer program documents.

From Manual Requests to Self-Serve Insights

The multi-agent solution removes guesswork and creates a seamless, explainable experience. By combining structured customer data with unstructured offer program information, the system enables:

• Self-serve eligibility troubleshooting: Instead of routing through multiple teams and Slack threads, sales teams can now quickly understand offer eligibility issues and take informed action directly, thanks to built-in explanations
• More intelligent targeting: Sales teams can focus on high-value accounts based on real growth signals and offer relevance, not hunches, streamlining how they identify high-impact opportunities
• Faster outreach: By increasing offer understandability and reducing manual friction, the response SLA decreases from 48 hours to under five seconds, and sales teams can move more quickly and confidently

Most importantly, the system scales as accounts are added and more offers are created. Customer account and GTM insights update automatically when the reference data in Unity Catalog changes, and new offer programs can be supported by updating the documents in the knowledge base – with no new code required.

Limitations

While the current system is powerful, there are a few limitations to note:

• Agent Overlap: Because the agents can’t directly share context, certain pieces of information needed to be duplicated across them, even though the supervisor “knows all.” For example, the Account Prioritization agent’s data table includes a column indicating which offer – if any – each account is eligible for (already known to the Eligibility agent). It also contains context about the usage eligibility bands for each offer type (already known to the Offer Details agent). This duplication ensures the Prioritization agent can reason about targeting and rank accounts correctly.
• User Workflow Integration: Most sales teams work primarily in Slack and Salesforce, not Databricks. Integrating this system as a Slackbot or into Salesforce would put eligibility details and guidance directly into their everyday workflows.

Conclusion

Commercial offers only work if sales teams know who to target — and why. Before Agent Bricks, this was a manual, multi-team challenge that slowed down outreach and introduced ambiguity into our programs. With Agent Bricks, we were able to build, test, and refine a multi-agent AI system with nothing more in hand than our data and our goal.

Though our system has a few limitations in its current form and isn’t embedded in the tools sales teams use daily, the gains have already been meaningful; it’s made offer targeting faster, more transparent, and more scalable. The real magic lies in the prioritization of accounts: the system automatically aggregates customer data and offer information to intelligently surface the highest-impact opportunities first, and I didn’t even have to tell the agent exactly how to do it. Now that’s data intelligence. 
 

Get started building with Agent Bricks and create your first solution today.