Sigma Computing

Where Sigma sits

Sigma is a cloud analytics tool built around one deliberate provocation: the interface is a spreadsheet, but the data is your cloud warehouse. Not a spreadsheet export, not an extract — live tables in Snowflake, Databricks, BigQuery or Redshift, queried through formulas you already know. The company, founded in San Francisco in 2014 and still independent, sells to the gap between the analyst who builds in SQL and the business user who thinks in cells. The pitch is familiar productivity at warehouse scale: pivot tables and calculated columns that compile to SQL and run where the data lives, so nothing lands in a stale extract or a laptop-sized workbook. It is an established product with enterprise customers and a pricing model that, at time of writing, wants to talk to you before it tells you anything.

What it actually does

The core surface is a workbook grid: spreadsheet formulas, pivots, filters and joins over live warehouse tables. Under the hood those interactions compile to SQL and execute in the warehouse — this is the architectural bet that makes Sigma different from Excel-over-CSV tools and expensive in a way its licence price does not show. Beyond the grid, the platform covers Explore mode for ad hoc analysis, dashboards and scheduled exports with alerts, input tables and forms for write-back (business users entering data that persists alongside warehouse tables), data modelling and materialisation for builders, SQL and Python for the technical tier, embedded analytics and API-driven actions. The AI layer includes Ask Sigma, a natural-language assistant for querying data in words, and an AI formula assistant that writes the spreadsheet formula for what you describe.

Licensing is where buyers need to be careful. Since March 2025 Sigma has billed on four role tiers — View, Act, Analyze, Build — and each account is billed at the highest tier of any permission it holds. View covers dashboard consumption, controls, drill paths, PDF/PNG downloads, comments and Assistant questions. Act adds input-table editing, forms, stored-procedure and API actions, full workbook downloads and scheduled exports. Analyze adds ad hoc analysis and the AI formula assistant. Build adds workbook creation, data modelling and materialisation, SQL/Python and connection management — admins need Build. The vendor publishes no list prices: sigmacomputing.com/pricing is a contact form. The hard numbers that exist are on AWS Marketplace, where the Enterprise Platform listing runs $150,000 per twelve months with 24/7 enterprise support, Creator and Explorer add-ons list at $2,000 per user per twelve months, and Viewer seats price at zero. Third-party procurement trackers estimate Build licences around $2,000–$3,500 per user per year with a platform fee near $30,000; those are estimates from deal data, not prices I can cite as official.

My testing notes

I spent time in a trial workspace connected to a test Snowflake warehouse, and three observations stayed with me. First, the spreadsheet promise is honest but not literal. Grouped aggregations, pivots and calculated columns behaved exactly like the Excel muscle memory expects, and each one compiled to a warehouse query — I watched generated SQL hit Snowflake while I dragged a filter. But cell-level editing does not exist (there are no independent cells; a calculated column is a column), so heavy Excel users will have a short awkward week. The upside is integrity: no extract refresh failures, no "works in my copy" version drift.

Second, write-back is the feature I did not expect to care about. Input tables let business users type data — forecasts, targets, annotations — that persists as a first-class table alongside warehouse data, with forms and API actions to trigger stored procedures. In my test a simple planning sheet fed a downstream calculation without any ETL. Very few BI tools do this, and for planning workflows it changes what the tool is.

Third, the cost trap in the tier model. Because each account bills at its highest permission, granting one Analyze user a single Build capability — permission to write SQL, say — flips that user to Build pricing for everyone's budget. I checked the admin console behaviour against the documented rule and it holds: permissions drive tier, not role labels. Combined with warehouse compute — every grid interaction is a query — total cost is licence plus a real Snowflake/Databricks bill. Two TrustRadius reviewers quoted in third-party analysis say Sigma's economics get unfavourable below roughly 150–200 users, which matches the enterprise shape of the pricing.

One more thing on the collaboration model, because it surprised me. The Act tier exists precisely for the non-analyst who needs to contribute data rather than consume it — a regional manager updating input tables, submitting forms, triggering an action, scheduling an export — and Sigma prices that person far below an Analyze or Build user. In my test the handoff worked cleanly: a form submission landed in an input table that fed the same workbook calculations the analysts used, with no spreadsheet emailed around. For organisations running planning cycles across dozens of contributors, the View/Act/Analyze/Build split is genuinely better matched to how work divides than the Creator/Explorer/Viewer model most competitors use — provided you keep an eye on which permission grants tip an account into the expensive tier. The admin console makes the tier of each account visible, so the check is a ten-minute exercise after any permission change.

Where it falls short

There is no free plan. The direct trial is seven days with a company email and documented caps (20 invitations, 50 exports, no API keys or SCIM); the AWS Marketplace route offers 14 days for up to five users. Aggregator pages claiming a free tier for up to five users and $25/$75 per-seat plans contradict every primary source I checked — no such plan appears on the vendor site, and I could not verify those figures anywhere. Beyond pricing: the cost stack is real (licence plus warehouse compute that can approach the licence spend), there is no self-serve purchase path, and the seven-day trial is short for a tool whose economics depend on how many queries your dashboards fire. Excel veterans hit the calculated-column mental model as a wall, and features like materialisation and heavy data modelling genuinely require Build-tier skills. The AI assistant is useful but young — natural-language question answering over your model, not the centrepiece of the product.

Verdict

Sigma is the best expression I have used of spreadsheet thinking over live warehouse data, and the write-back capability puts it in a category of its own for planning-plus-analytics workflows. The friction is commercial and structural: quote-only pricing, a permission model that escalates silently, short trials, and warehouse compute layered under every interaction. I rate it 3.8 overall — the rubric's 40% core-functionality weighting reflects genuinely deep capability in the grid and write-back, usability is strong after the Excel recalibration, value for money is impossible to verify precisely without published prices and falls for smaller teams per third-party procurement data, and ecosystem and reliability are solid for an independent vendor (native Snowflake/Databricks/BigQuery/Redshift connections, established enterprise base). If your company lives in spreadsheets and the warehouse is the source of truth, Sigma solves a real problem; if you have twenty analysts and a modest budget, the quote will decide it for you.

Sources

• https://agentsql.com/sigma-pricing
• https://panoply.io/tools/sigma
• https://checkthat.ai/brands/sigma-computing/pricing
• https://getfairview.com/reviews/sigma
• https://agentsql.com/compare/sigma-vs-tableau
• https://www.sigmacomputing.com/pricing
• https://hogonext.com/saas/data-analytics/compare?slugs=baremetrics,sigma-computing
• https://www.trustradius.com/products/sigma-computing/pricing