Sigma

What it is

Sigma is a BI tool that looks like a spreadsheet and runs every click as a live query against your cloud warehouse — Snowflake, Databricks, BigQuery, Redshift or Postgres. Nothing is extracted into a Sigma-owned store, so row-level security and column masking you already set in the warehouse apply to every chart without being rebuilt. What separates it from Tableau or Power BI is the write path: input tables let a business user type forecasts, quota overrides or account notes into a grid that lands as rows in the warehouse, and actions chain those edits into forms, approvals and API calls. That turns a dashboard into a small data app. In April 2026 Sigma added Sigma Agents — no-code agents built inside a workbook that run on a schedule or an event, read governed warehouse data, and write back, post to Slack or push to Salesforce and HubSpot under the triggering user’s permissions. Sigma reported $200 million ARR in April 2026 with more than 100% year-over-year growth and 2,000+ customers, then raised an $80 million Series E at a $3 billion valuation on May 18, 2026, led by Princeville Capital with Databricks, ServiceNow and Workday’s venture arms joining.

Why it shows up in RevOps and CS stacks

• Writeback replaces the shadow spreadsheet. The recurring RevOps pain is a forecast call, territory plan or renewal-risk review that lives in Google Sheets because the BI tool is read-only. In Sigma, an AE’s commit or a CSM’s risk flag is an input-table row next to the CRM and product-usage data it references, with an edit log. Input-table editing needs only the Act license, not a builder seat.
• Governance is inherited, not re-implemented. Agents mirror user permissions — if a user cannot write to a table, neither can their agent. For a CS team querying product telemetry that includes customer PII, this is the reason legal says yes.
• It sits inside the AI tooling you already run. The Sigma MCP server lets Claude, ChatGPT, Codex and Cursor search, describe and query Sigma data models under OAuth. An August 2026 public beta lets you run Sigma’s AI features on your own Anthropic API key and contract.

Pricing reality

No list prices are published. Licensing is four tiers — View, Act, Analyze, Build — plus usage credits and your own warehouse bill. Credits meter specific events: one per input-table row changed in a published workbook (capped at 100 per action), one per API or stored-procedure call from an action, one per unique exported file, and 0.1 per message sent to an agent. Draft edits and CSV uploads to drafts cost nothing.

Buyer data: Vendr puts the median contract at $64,941 a year across 154 purchases, ranging $15,585 to $145,339, with buyers landing 24.6% below first quote. Its bands run $20,000–$60,000 for 10–25 users and $60,000–$150,000 for 25–75. Build seats cost roughly 3–5x a View seat, so the license mix drives the bill more than headcount. Budget $25,000 a year as the floor for a real RevOps deployment, before warehouse compute.

Best for

RevOps and CS ops leads at companies with a modeled Snowflake or Databricks warehouse and 15–100 business users who need both self-serve reporting and a handful of writeback apps — forecast submission, territory and quota planning, renewal-risk review — without standing up a separate internal-tool platform.

Do not buy it without a cloud warehouse; Sigma has nothing to query. Skip it if you run Microsoft 365 E5, where Power BI Pro is already in the bundle and the marginal BI cost is near zero. And skip it if the app is mostly UI and API glue with little analytics — that is a Retool job.

Versus the alternatives

Power BI is the incumbent by installed base, at $14 per user per month for Pro since April 2025. Pick it when you are a Microsoft shop and reporting is the whole job. Pick Sigma when business users must edit governed data, or when your warehouse is Snowflake or Databricks and you want every query live rather than refreshed into a dataset.

Tableau Cloud Creator lists at $75 per user per month. Pick it for heavy visual exploration by trained analysts. Pick Sigma when the audience is spreadsheet-native ops staff and the answer has to write back.

Omni is the fastest-growing entrant, raising a $120 million Series C at $1.5 billion in April 2026 on 4x year-over-year revenue growth. Pick Omni when a governed semantic layer, Looker-style, is the priority and writeback is not. Pick Sigma when writeback and agents are the point.

Retool has a similar median contract ($64,623 per Vendr). Pick Retool when the app is forms and API calls with analytics as a side panel; pick Sigma when the app is a governed table people edit and chart. To push Sigma-defined segments into the CRM on a schedule, pair it with Hightouch rather than building actions per field.

If none fit because finance, sales and CS disagree on what an active account is, you have a modeling problem — see data warehouse vs CDP.

Watch-outs

• Agents are free until November 1, 2026, then paid. Sigma’s billing docs say so, without a published rate. Guard: before building renewal-risk or forecast agents into daily operations, get the post-November agent price and included message volume written into your order form.
• Writeback credits scale with rows, not users. A scheduled action updating 2,000 account rows nightly spends 2,000 credits a night. Guard: model credit volume from your planned actions before signing, do bulk loads in drafts (zero credits), and negotiate included credits plus the overage rate.
• Warehouse compute is not on the Sigma invoice. Every interaction is a live query, and reviewers flag slowness on large unaggregated tables. Guard: point Sigma at a dedicated small warehouse with auto-suspend and a resource monitor, materialize heavy models, and track Sigma-attributed spend for 60 days.
• Writeback support varies by warehouse. BigQuery input tables are in public beta, Sigma recommends ra3.4xlarge or larger nodes for Redshift, and Sigma Tables — shared writeback across workbooks — is in private beta. Guard: confirm your warehouse on Sigma’s supported-features page and design apps around single-workbook input tables until Sigma Tables is GA on your platform.
• The MCP server is read-only. It searches and queries; it does not edit input tables or build workbooks today. Guard: keep write actions inside Sigma workflows where permissions and the edit log apply, and use MCP for analysis only.