↓Skip to main content
  1. posts/

Illusion of Progress

·2 mins

Quote from Winning With Data:

Traditional business intelligence systems comprise three layers. An Oracle, or similar, database stores all the company’s relevant data. The middle layer, called the data warehouse, siphons data from the database and aggregates figures to create reports. The last layer visualizes the report data and serves it to end users.

Problems with this approach surfaced very quickly. As businesses change, the questions they ask themselves evolve, and in many instances they change dramatically. Each time a business user sought to answer a new question, the data team needed to create a new report within the data warehouse, which took time, effort, and money. As the data warehouse grew, the company paid more to Oracle and other vendors for additional database licenses.

With such a brittle BI architecture, customers couldn’t ask and answer new questions. Or they would have to wait weeks, or more, for database teams to adjust the data structure and re-architect the database in order to respond to the organization’s new demands. This latency minimized the value that data can have for operations, ultimately dissuading employees from asking pertinent questions and from explaining their actions and making decisions based on real business data.

This has me wondering: How much has truly changed in our modern data stack? Despite the evolution to cloud data warehouses, transformation tools like dbt, and self-service analytics platforms, we still face similar fundamental challenges:

  • Business users remain dependent on “analytics engineering” teams to model data appropriately
  • New questions often require rearchitecting schemas and creating new transformations
  • The time between question and answer still creates friction that discourages data-driven decision making
  • Technical barriers continue to separate business users from direct data exploration

This persistent gap between business needs and data accessibility explains why natural language interfaces to data remain such a promising concept. The ability to simply ask questions of your data—without SQL knowledge or waiting for technical teams—would fundamentally change how organizations leverage information.

But are we anywhere close to delivering on this promise? Or are we still trapped in prettier versions of the same brittle architectures, just with different names and technologies?

What would it take to truly democratize data access in a way that eliminates these longstanding barriers?

↑