Analytics chatbot development on Databricks to reduce data team dependency

Context and objectives

In a leading self-storage provider operating hundreds of facilities across Europe, business users relied heavily on the data team for ad-hoc data requests, limiting their ability to independently access and analyze business data. This created bottlenecks in decision-making and held back the organization's broader data democratization efforts.

The objective was to reduce this dependency through analytics chatbot development, connected directly to the company's existing data warehouse. The project aimed to:

• Deliver a functional proof of concept within a limited, well-defined scope

• Enable a select group of users to query business data in natural language, regardless of technical background

• Produce comprehensive documentation to support the solution going forward

Approach

1. Technology selection

After comparing several technical options, Agilytic recommended a solution directly integrated into the client's existing data platform. This choice balanced the client's need for fast proof-of-concept development against a preference for limiting dependency on any single vendor. They were then able to start testing adoption internally without a lengthy technical setup.

2. Iterative build and refinement

The solution was built and refined progressively:

• A development environment was created and connected to the relevant business data.

• The chatbot was fine-tuned with the metrics and filters used in the client's existing reporting (since much of the business context behind the numbers existed only in those reports and not in the raw data itself).

• An automated deployment process was built, allowing the solution to move smoothly between development, testing, and production environments.

Given time and budget constraints, the team deliberately prioritized depth over breadth, focusing on two key data areas and enriching them with additional business metrics, rather than expanding coverage across many data sources.

3. Controlled testing

The solution included a built-in feedback mechanism, allowing users to rate responses and leave comments, which supported fast iteration on accuracy. Once the underlying business context was properly captured, response accuracy reached a strong level. Testing was carried out with a small group of users, reflecting the proof-of-concept scope agreed upon with the client.

Results

• Four analytics chatbot environments covering development, testing, and production, including one exploring an additional data area

• An automated deployment process integrated into the client's existing pipeline, to move the solution smoothly across environments

• A custom data-transformation solution to support a business analysis need identified during the project

• Comprehensive documentation to support future configuration and maintenance

Response accuracy was strong once business context was properly incorporated, confirming that data quality and structure are a decisive factor in this type of solution.