Embedded Analytics

Showing charts and reports inside your product to customers.

Data Engineer · Senior Backend Engineer · Senior

Also known as: embedded BI, customer-facing analytics, in-product analytics, embedded dashboards

Embedded analytics means putting charts, reports or a full analytics view inside your own product, so your customers see data about themselves without leaving the app or logging into a separate BI tool. You build the experience; the customer may never know a BI product is behind it.

The classic mistake is treating it like an internal dashboard. An internal BI dashboard is usually single-tenant: one company’s data, a handful of trusted users. Embedded analytics is multi-tenant: many customer accounts share one application, and each must see only its own rows. A query that forgets the tenant filter, or a cached chart keyed without the tenant id, can show one customer another’s numbers. That is a data breach, not a cosmetic bug.

What makes it different

Tenancy is the core problem. Every query needs a tenant filter enforced by the server, not the client. Row-level security or a scoped data API is the usual mechanism.
You cannot trust the embedder. Parameters sent from the browser can be changed, so decide what a tenant may filter on server-side.
Performance is part of the product. A slow report becomes a support ticket. Precompute with aggregate tables and cap query cost, since many tenants query at once.
Reuse a semantic layer so the numbers customers see match the numbers your team quotes.
Charts appear in your design, with your colors and language.

Build or embed

You can build charts yourself against a data API, embed a commercial embedded-BI product, or extend an existing BI tool. Building gives control and no per-seat cost but is real product work. Buying gets you there faster but adds a dependency and per-user pricing. The same reasoning as build vs buy applies.

Cautions

Some platforms offer data sharing instead of copying, which can be simpler and fresher than duplicating per tenant.
One heavy tenant should not slow everyone; isolate workloads where the platform allows.
Sensitive columns still need column-level security even inside a customer’s own data.

For data engineers this is a serving concern; for backend engineers it is a multi-tenant security and performance concern. Get tenancy right before adding features.