Embedded analytics requirements checklist

The strongest answer enforces tenant filters in the analytics layer and in the database, not only in the front end, so a dashboard cannot return another customer's rows even if a filter is edited. Ask the partner to show where the filter is defined, how it is bound to the signed-in identity, and what happens when a user has no matching attribute (the correct result is no rows, not all rows). Request a test script that attempts cross-tenant reads from a guest or SSO session and the logged output. In Metabase, row and column level permissions are listed on the pricing page as a Pro and Enterprise feature, so confirm the edition in the proposal. Superset's row-level filtering should be verified the same way in your own environment.

A sound answer names the embedding method, who issues the token or assertion, how long it lives, and how the product's session maps to dashboard permissions. Guest tokens are signed by your server and carry the tenant or customer attributes, while SSO embedding relies on your identity provider's groups or claims. Ask which protocol is used for each option, for example JWT or OIDC for Metabase, and confirm that the edition you plan to buy includes it, since the Metabase pricing page places JWT, OIDC and SAML in Pro or Enterprise. Request the signing key handling procedure, the expiry settings, and a demo where a revoked user loses access without a manual step in the dashboard tool.

Start from the vendor's current pricing page and ask for the plan in writing against each feature you need. On Metabase's pricing page, SSO, row and column level permissions, multi-tenant data segregation and usage analytics and auditing are included only in the Pro and Enterprise plans, not in Open Source or Starter. Starter and Pro list their included users and per-user additions, and Enterprise is priced by quote starting at $20,000 per year. Apache Superset is an Apache-2.0 project with no paid edition, so its cost sits in infrastructure and engineering. Ask the partner to map each requirement to an edition, plan, or self-built component, and to state which items recur each year.

The right answer separates the analytics workload from transactional systems with a replica, a read-only analytics store, or a warehouse such as ClickHouse, and it caps query concurrency and timeouts. Ask how the dashboard tool's connections are pooled, which role it uses, and whether heavy dashboards refresh on a schedule against cached or precomputed tables. Superset's documentation notes that async queries, alerts, reports and thumbnails need a caching layer and Celery workers, so ask how those are sized. Request a load test that shows query latency on the production replica during peak embedded traffic, plus the configured timeout values.

The answer depends on how you distribute and modify the software, so ask for a written opinion from your counsel, and ask the partner what it has seen in comparable setups. Metabase ships its open source edition under the AGPL and its commercial editions under the Metabase Commercial Software License, so confirm which license covers each component you deploy. Ask whether your team plans to modify Metabase source, whether the modified version is offered to users over a network, and whether any enterprise directory code is included. Apache Superset is Apache-2.0 with no paid edition, which changes the obligation profile. DeployOpen does not give legal advice, so treat the partner's technical notes as input to your counsel's review.

A credible answer lists the events captured (logins, embedded views, query text, export actions, and the identity or tenant behind each one) and where they are stored and for how long. Ask whether usage analytics and auditing is available in your chosen edition, since Metabase limits it to Pro and Enterprise. If you operate under the EU AI Act, note that Article 12 concerns automatic event logging for high-risk AI systems, and Article 19 requires providers to keep those logs for at least six months, so confirm whether that applies to any AI features in the product. Request a sample audit export and a retention setting that you can change without vendor help.

Yes if your team owns the runbooks, and the proof of concept should show that before you commit. Upgrades need a backup of the application database, a staging test, and a read of the release notes; for Superset, the documentation says to back up the metadata database, run the upgrade steps, and check UPDATING.md for breaking changes. Metabase applies application database migrations automatically, and its docs recommend a single node during major upgrades and upgrading release by release from versions older than 40. Ask a partner to document these steps, support the first upgrade, and state whether ongoing support is a fixed-scope project or a retainer.

The largest drivers are the licensing tier, the number of users who see dashboards, the infrastructure, the analytics database and the embedding and security work. Apache Superset is licensed under Apache-2.0 and has no paid edition, so its software cost is zero, while the rest of the bill comes from servers, the database and labor. Metabase is different: its Open Source edition is free, but SSO, row and column level permissions, multi-tenant data segregation and usage analytics are included only in its Pro and Enterprise plans, according to its pricing page. Ask every vendor which features in your design sit behind a paid tier.

According to the Metabase pricing page, Pro is $575 per month, or $517.50 per month billed yearly, with the first 10 users included, and additional users cost $12 per month. Starter is $100 per month, or $90 per month billed yearly, with the first 5 users included, and additional users cost $6 per month. Enterprise is priced by quote starting at $20,000 per year. These are list prices from the vendor's page at the time of publication, so confirm current figures directly with Metabase before you budget.

Many teams can, provided they have engineers who know the platform's architecture, identity integration and data modeling. Superset, for example, has four components to operate: the application, a metadata database, an optional caching layer and optional Celery workers, and the caching and worker components are needed for alerts, async queries and thumbnails. Operating it also means backups, upgrades and security patching on a recurring schedule. Assess whether you have that skill set on staff for the long term, not only for the first launch.

Duration depends on the number of data sources, how many dashboards must be migrated or rebuilt, the complexity of row-level security, the identity provider integration and the number of review cycles your security team requires. Metabase's upgrade guidance, for instance, expects staged upgrades and a database backup before each change, which adds time to any rollout that touches production. Ask a partner for a timeline tied to a written scope, with milestones for staging, security review and go-live. Do not accept an estimate that is not based on that scope.

Plan to supply three kinds of input. Identity owners must define how users map to roles or groups, which can be done through OAuth or LDAP group mapping such as Superset's AUTH_ROLES_MAPPING. Data owners must confirm which tables, columns and tenant keys drive row-level filters and which datasets may be exposed. Security reviewers must approve the authentication design, token lifetimes, secret handling and logging, and the review should start early because it often sets the schedule.

Ask for a fixed scope that lists each platform component, each data source, each embedding pattern and each identity integration. Request the one-time costs (design, build, testing, documentation, handover) separately from recurring costs (hosting, database, licenses or plan fees, upgrades and support), and ask which plan tier each feature requires. Also request assumptions about user counts, staging environments, backup and restore testing, and what happens to work outside the scope. Compare quotes on that basis rather than on a single total.

Recurring costs typically include hosting, the analytics and metadata databases, backup storage, paid plan fees if you choose a commercial tier, and staff time for upgrades and patching. ClickHouse, when used as the analytics database, has its own release cadence and support windows, with stable releases roughly monthly and LTS releases supported for a year, according to its production guidance, so the upgrade effort is recurring. Budget for these lines explicitly, and ask a partner whether they offer ongoing support as a separate retainer.