The ideal scenario for a BI implementation is a team of non-technical subject matter experts that can get answers to any data related questions by themselves. The Dresner 2025 BI tool survey had 60 percent of respondents say self-serve features were crucial for them and every vendor heavily advertises how their tool could help achieve this. In my experience, however, self-serve works best when added on top of a mature data model.
If you’re a small team in a rapidly growing company the typical bottleneck is the data modeling layer (bringing in new data sources, transforming them, creating defined metrics on top and then disseminating that knowledge to the business users). The rapid improvements in coding agents over the past year have flipped this dynamic - code is easy to generate and with a tool like dbt where tests, metrics, relationships, descriptions are all text the only roadblock is a clear set of requirements. Modern BI tools are bringing in AI agents to do something similar for dashboard generation but these are isolated from the underlying data model.
I believe the solution for small teams instead lies in tools such as evidence-dev where the BI layer is also just code. You’d have an integrated code base with the data model, requirements, schema documentation and dashboards all in one place enabling fast AI assisted turnaround times.