From 'BI User' to 'BI Tool Builder',keep going on my way. Thanks Alex Mou. The development of Large Language Models is reshaping the data analysis industry.
As a BI engineer who heavily relies on Tableau every day, I’ve been thinking: instead of worrying about the potential decline in traditional analyst roles, it makes more sense to proactively leverage new technologies to develop underlying automation tools. Since AI is already proficient at writing Python code, we can completely use Python to programmatically generate and modify Tableau workbooks (.twb). This is the core logic behind the automation tools I am currently developing. In a recent discussion with industry veteran Alex, he brought up a very specific and high-frequency pain point in daily development: the precise, local replacement of KPI templates.
In actual business scenarios, we often need to duplicate a complete KPI module (including current values, YoY/MoM changes, trend charts, etc.) and only replace the underlying measure for that specific module—for example, changing "Sales" to "Profit".
When using Tableau's native "Replace References" feature, the system typically executes a global replacement.
This easily breaks shared calculated fields that are actively being used by other views within the same workbook. Through my current programmatic solution, we can parse the underlying dependencies directly at the code level to complete this local variable replacement safely and precisely.
It doesn't just replace the surface-level measures; it dives deep into calculated fields and filters, automatically handling all nested variables while keeping the original formula structure intact.
This approach shortens what used to be a tedious manual modification process into just a few seconds. Currently, this tool primarily runs as a Python script and an MCP Server, which is tailored more toward data engineers with a programming background. To lower the barrier to entry and allow regular business analysts to easily perform such automated operations, my next step is to package it into a Web App.
In the future, users will be able to simply input specific modification commands or configuration parameters on a webpage.
The system will automatically parse the underlying code in the background and directly output the modified Tableau file. Transitioning from "using tools" to "building tools" is a process of constant trial, error, and iteration. Building in public within the community has provided me with incredibly valuable real-world business requirements and feedback. Regarding this automated approach of "refactoring BI dashboards through code," what urgent pain points have you encountered in your actual work? I welcome any thoughts and discussions in the comments.
#Tableau #DataVisualization #ArtificialIntelligence #BusinessIntelligence #OpenSource #DataAnalytics #MCP #AIAgents #DataEngineering #datafam #gemini #google #ai #bi #agent #cwtwb #codex #chatgpt #salesforce