Skip to main content

Decision brief

Hex Review: What to Settle Before Opening AI Analysis to the Whole Company

Buy Hex when your data team will curate the context its agents read and test it with evals. This review covers seats, credits, and security before business users ask questions.

Reading time
2 min read
Sources
19 cited
In this piece
Hex+65 Strong positive

GILD verdict

Buy Hex when your data team will maintain the context its agents read and test it with evals.

Across models in Hex's own DataBench, agents scored 75% on direct questions and 54% on trap tasks where the obvious answer is wrong. Open Threads to the whole company only after each domain passes its eval suite.

Hex

AI Data Analysis by Hex

+65 Strong positive
Product page

Fit and limits

Who it fits

  • Your data lives in Snowflake, BigQuery, Databricks, Redshift, or another of the 16 supported databases, and the data team writes SQL or Python.
  • Business users wait in a queue for ad hoc questions, and the data team can endorse the tables and projects that should answer them.
  • You already define metrics in dbt MetricFlow, Cube, or Snowflake Semantic Views, which Hex can sync as semantic models on Team and Enterprise.
  • People want answers where they already work: Hex's agent answers in Slack, and its MCP server serves Claude, ChatGPT, Cursor, and Glean.

Where it falls short

  • Threads prioritizes endorsed data and semantic models, and Explorers use Endorsed Mode by default. A domain nobody has curated gives business users little to work with.
  • Every agent run consumes credits, and per-seat grants (30 on Professional, 40 per Team Editor) reset monthly without rolling over.
  • SSO supports OIDC only, with no SAML, and SSO, OAuth warehouse connections, audit logs, and bring-your-own-key AI are Enterprise features.
  • The Notebook Agent is built for technical users who can audit the SQL and code it writes, so analyst review stays in the loop for anything beyond Threads.

Implementation reality

  1. Curate one domain first

    Endorse tables and projects for one data domain. Write workspace rules for the agent, then build or sync a semantic model. Filter each connection's schema to limit the data business users and agents can see.

  2. Prove it with evals

    Write a YAML eval suite with the domain's required questions and a rubric for each. Run it from the Hex CLI, then publish it to run on a schedule. Evals require Team or Enterprise and consume agent credits.

  3. Open Threads to one team

    Give the team Explorer seats or access through Slack, and keep Endorsed Mode on. Context Studio's questions, topics, and warnings show where context is missing. Admins can open individual conversations, and Managers see aggregates.

  4. Cap spend and settle data terms

    Set credit limits by user and group and a monthly ceiling on auto top-ups. Decide whether models that require provider retention are allowed, confirm the AI training opt-out, and mark sensitive connections before turning on Slack and MCP access.

What to measure in your pilot

  • Run one domain's eval suite with and without its semantic model; open Threads on domains without one only if scores stay within 5 points.
  • Log credits per Thread from the Thread inspector for a month, since some heavy builders already use near 500 credits a month against a 40-credit grant. At 1 credit per simple question, 40 questions cost about US$15 beyond a Team Explorer's grant.
  • GILD found no team write-up on splitting work between the Notebook Agent and coding agents using the Hex CLI; track which tool built each published app and which needed analyst fixes.

Sources

Official documentation
19
  1. Hex pricingOfficial documentationhex.tech
  2. Hex docs: CreditsOfficial documentationlearn.hex.tech
  3. Hex docs: Billing FAQOfficial documentationlearn.hex.tech
  4. Hex docs: AI in HexOfficial documentationlearn.hex.tech
  5. Hex docs: AI data privacyOfficial documentationlearn.hex.tech
  6. Hex docs: Data connections introductionOfficial documentationlearn.hex.tech
  7. Hex docs: Introduction to semantic modelingOfficial documentationlearn.hex.tech
  8. Hex docs: ThreadsOfficial documentationlearn.hex.tech
  9. Hex docs: Context StudioOfficial documentationlearn.hex.tech
  10. Hex docs: EvalsOfficial documentationlearn.hex.tech
  11. Hex docs: Agent observabilityOfficial documentationlearn.hex.tech
  12. Hex docs: Hex MCP serverOfficial documentationlearn.hex.tech
  13. Hex docs: Hex Agent in SlackOfficial documentationlearn.hex.tech
  14. Hex docs: Sensitive data connections for external integrationsOfficial documentationlearn.hex.tech
  15. Hex docs: Workspace rolesOfficial documentationlearn.hex.tech
  16. Hex docs: SSOOfficial documentationlearn.hex.tech
  17. Hex: Data securityOfficial documentationhex.tech
  18. Hex: EnterpriseOfficial documentationhex.tech
  19. Hex blog: DataBench, a benchmark for agentic analytics (Aug 13, 2026)Official documentationhex.tech

Keep reading

  1. Buyer guideHex AI Rollout PlaybookHow to curate context, open Threads to business users, and gate each data domain on evals before Hex's agents reach the whole company.
  2. Buyer guideHex Seats, Credits & Spend ChecklistWhich roles need paid seats, how monthly credit grants and the workspace pool work, and where to cap agent spend and compute.
  3. Buyer guideHex Security & AI Data Governance ChecklistSSO, warehouse authentication, sensitive connections, AI training and retention settings, BYOK, and where Hex stores query results.
Read the research