Sigma occupies an interesting position in business intelligence. The pitch is that business users already know spreadsheets, so instead of teaching them a new visual query paradigm you give them a spreadsheet interface that compiles down to SQL against the warehouse. No extracts, no separate cube, the computation happens in Snowflake or Databricks or BigQuery. It is a genuinely sensible idea and it addresses a real failure of the previous BI generation, where governed dashboards were built by a central team and everyone quietly exported to Excel anyway because that is where they could think.
QuickStarts are how Sigma teaches this, and the format is honest about what it is. Each one takes a specific task, walks through it in numbered steps with screenshots, and finishes in under an hour. Connect a warehouse. Build a first workbook.
Create a parameterised report. Set up row-level security. Embed a workbook in an application. As reference material to work through when you have a concrete job to do, this format is efficient and I prefer it to a ninety minute video where the useful thirty seconds is buried somewhere in the middle.
The embedded analytics guides are the standout. Sigma has invested heavily in embedding, where a software company puts Sigma inside its own product and sells analytics to its customers, and the QuickStarts on this are more detailed than free vendor material usually is. JWT-based authentication, row-level security in a multi-tenant context, theming to match a host application. Anyone evaluating embedded BI would get real value from these regardless of which vendor they eventually pick, because the architectural problems are shared.
The spreadsheet-to-SQL explanation is handled well, and it needs to be, because the abstraction leaks. When a user writes something that generates an expensive query against a billion row table, somebody pays for that compute, and the training does address query efficiency and materialisation. Not as prominently as I would like, given that unpredictable warehouse spend is the most common complaint about this architecture, but it is there. What is missing is education in the proper sense.
QuickStarts teach you what to click. They do not teach you what to think. There is nothing meaningful on choosing an appropriate visualisation for a given question, on the difference between a metric that is well defined and one that four departments compute differently, on cohort construction, on why a percentage change on a small base is usually noise, or on how to tell whether the dimensional model underneath your workbook supports the question you are asking. These are the skills that distinguish an analyst from someone who can operate a BI tool, and a business user working through every QuickStart will come out able to build workbooks and no better at analysis than they started.
That may be exactly Sigma's intent. The product's proposition is that business users already have the domain knowledge and only need access, and there is something to that. But the assumption that spreadsheet familiarity equals analytical competence is doing a lot of quiet work, and anyone who has cleaned up after a confidently wrong Excel model knows why that assumption is shaky. The absence of a certification programme is a smaller gap and I am ambivalent about it.
I have spent enough of this project criticising vendors for charging hundreds of dollars for credentials that I am not going to demand Sigma invent one. But some employers do look for them, and Sigma's competitors have them. The AI agent content is the newest material and it reads that way, thinner and more promotional than the core guides. Three point six.
Efficient, free, well suited to its purpose, and quietly leaving the hardest part of analytics to somebody else.