# Analytics Belongs Where the Work Happens: Four Embedded Demos You Can Try > Source: > Published: 2026-10-06 00:00:00+00:00 [The Wren Journal](https://getwren.ai/blog) Product, Insight # Analytics Belongs Where the Work Happens: Four Embedded Demos You Can Try Nobody leaves your product to go ask a dashboard a question. The question shows up mid-task, about the thing on screen. We built four working demos, in capital markets, retail, healthcare, and insurance, to show what happens when the answer lives there too. Wren AI Product Team Updated: Oct 06, 2026 Published: Oct 06, 2026 An investor is staring at a chart that just dropped 6% and wants to know why. A shopper sees two charges on one order. A GP has a patient in the room and needs the HbA1c trend before deciding on a prescription. A claims agent has a policyholder on hold asking what a roadside add-on would cost. None of these people are going to open a BI tool. They are inside a product, in the middle of a task, and the question is about the thing in front of them. If the answer isn't there, they do what people have always done: guess, escalate, or give up. That is the case for embedded analytics in one sentence. The answer has to live where the question is asked. This week we built four working products that put a governed Wren AI thread exactly there, one per industry, and we'd like you to go click around in them. ## Why "embedded" is not a dashboard in an iframe Most embedded analytics is a dashboard someone built for one audience, bolted into a product page. It answers the questions the builder anticipated and nothing else. The moment the user's question diverges from the layout, they are back to filing a ticket. A conversational thread embedded in the product is a different thing, for three reasons. **The context is already on screen.** The host application knows who is signed in, which account, patient, policy, or stock they are looking at, and what timeframe is in view. When it passes that into the session, "this stock" and "my orders" resolve without the user typing anything. A standalone BI tool knows none of this. **The trust boundary is the product's, not the analyst's.** A dashboard scoped to one team is safe because a person scoped it. A natural-language interface has no such boundary. Anyone can ask anything, so the scoping has to be enforced below the model, inside the engine, per end user. Otherwise "please only show my data" is a suggestion, not a control. We wrote up how Wren AI does this in [How Wren AI enforces role-based access control on AI-generated SQL](https://getwren.ai/post/how-wren-ai-enforces-role-based-access-control). **The answer has to arrive in the product's own voice.** Your users see your brand, your terminology, your definitions. "Gained the most" or "refund window" mean one thing in your product, and the embedded answer has to use that meaning, not a column name the model guessed at. The first point is a product requirement. The second two are why embedding is hard to do well, and why most teams that try to build it in-house spend months on identity plumbing and policy code instead of the experience. ## Model once, embed anywhere The way Wren AI makes this tractable is the same idea behind everything else we build: the context layer is the work, and the surface is configuration. You define metrics, relationships, and access policies once in MDL. Every surface, whether an Embedded Thread in an iframe, your own UI on the Embedded AI API, or an agent calling Wren over MCP, resolves through that one definition. The host signs a JWT with the end user's identity and session properties. Wren Engine applies row- and column-level policies at query time, and every answer carries the SQL and rows behind it so the user can check the number. That is the pattern all four demos share. What differs is the industry, the surface it sits in, and who is signed in. ## Four products, four industries Each demo is a self-contained product on sample data. Open one, pick a suggested question or type your own, and watch how the thread uses what the host already knows. ### Capital markets: an AI analyst next to the chart A stock-watch app with an analyst side panel. The host passes the signed-in investor, their watchlist, and the symbol and timeframe on screen. The panel answers from the same daily prices, volume, analyst ratings, and news events that drive the dashboard. Try: "Which stocks in my watchlist gained the most over the past year?" or "Is COST's trading volume unusual compared to its 30-session average?" What to notice: "gained the most" is defined once in MDL as a return calculation, so the answer uses the app's own definition. The SQL is one click away. ### Retail: a support widget that knows whose order it is A storefront support widget scoped to the signed-in customer. It answers from account, order, payment, and help-center data, shows cards and charts where they help, and hands unresolved issues to a human with the conversation attached. Try: "Where is my order #112-4839201?", "I was charged twice for my last order", or "How much did I spend each month this year?" What to notice: the storefront passes the customer as a session property, so "my orders" resolves to this shopper's orders and nobody else's, enforced inside the engine rather than in the widget's code. ### Healthcare: a consultation co-pilot scoped to the open patient A signed thread inside a clinical workspace. The EHR signs the clinician, practice, and open patient into the token, so the thread can only query that one record. The GP gets lab trends, active medications, and safety flags without leaving the consultation. Try: "Show HbA1c trend and medications" or "Check metformin safety and renal flags." What to notice: every flag comes back with the rows and SQL behind it. The clinician verifies before acting, and there is no second screen to open mid-consultation. ### Insurance: answer and quote while the caller is on the line A claims-agent desktop where the assistant unlocks only after the caller's identity is verified. The desktop then signs the agent role, policy, and claim into the session. The thread retrieves coverage, prices a mid-call quote, and compares against similar claims. Try: "What is the claim status and coverage?", "Quote roadside assistance", or "Compare similar claim review flags." What to notice: limits, deductibles, and rating rules are modeled once, so the quote matches what billing would charge. Comparisons against other claims return aggregates only, never another policyholder's details. ## What the four have in common Strip away the industry and the same shape appears in each one. - **The host supplies the context.** Identity, role, and what is on screen travel into the session. The user never re-types what the product already knows. - **The engine enforces the boundary.** Row- and column-level policies run at query time, per user, regardless of what the model wrote. - **Definitions live in one place.** Returns, refund windows, renal flags, and deductibles are modeled once, so the embedded answer agrees with the report finance or compliance already trusts. - **Every answer shows its work.** SQL and rows are one click away, which is what turns a chat response into something a professional will act on. - **Escalation keeps the context.** When a person needs to take over, the thread hands off the conversation and the record instead of starting from zero. None of this is specific to finance or healthcare. It is what any product needs before it can put a question box in front of real users and real data. ## Go try them, then build yours Start at the [embedded demos page](https://getwren.ai/embedded) and pick the industry closest to yours. Each demo has an About tab with the surface, the session scope, and the suggested questions, and you can open it full screen. When you're ready to build, there are three ways in, all on the same engine: - **Embedded Threads** drops the full conversational experience into your app with one iframe snippet, with JWT identity and white-label branding.[Read the docs](https://docs.getwren.ai/cp/guide/integrations/embedded-threads) . - **Embedded AI API** gives you the raw endpoints, natural language to governed SQL plus Vega-Lite charts, so you build your own UX.[API reference](https://wrenai.readme.io/reference/cloud-getting-started) . - **MCP** lets Claude, ChatGPT, or your own agent call Wren as a sub-agent through your context layer.[MCP docs](https://docs.getwren.ai/cp/guide/integrations/wrenai-mcp) . If you are weighing whether to build this yourself, the [Embedded GenBI build-vs-buy ROI worksheet](https://getwren.ai/download/embedded-roi) walks through the engineering and maintenance cost of an in-house context layer. Or [request a demo](https://getwren.ai/request-demo) and we'll walk you through embedding on your own data. The four demos use sample products and sample data, for illustration only. Answers are read-only and describe what already happened. Wren AI does not place trades, give investment advice, diagnose, prescribe, or make offers of coverage. ### See governed GenBI in action Watch short demos of Wren AI turning business questions into governed SQL, charts, and reusable GenBI apps. [Watch product demos](https://getwren.ai/demos) ## Related posts [Sep 28, 2026](https://getwren.ai/post/how-wren-ai-enforces-role-based-access-control) ### How Wren AI Enforces Role-Based Access Control on AI-Generated SQL Wren AI doesn't replace your data security policy, it enforces it. Row-level and column-level security are applied in Wren Engine at query time, driven by who is asking, even when the SQL was written by an AI. [Aug 02, 2026](https://getwren.ai/post/the-dashboard-was-never-the-asset) ### The Dashboard Was Never the Asset For twenty years we treated the dashboard as the deliverable, but the durable asset was always underneath it: what your numbers actually mean. Dashboards depreciate. Context compounds. [Apr 02, 2026](https://getwren.ai/post/the-missing-context-layer-for-ai-agents-over-business-data) ### The Missing Context Layer for AI Agents Over Business Data Why we rebuilt Wren Engine from a context layer into an open context engine, and what we learned along the way The Wren Journal ## Get the next deep dive in your inbox Practical GenBI guides, product updates, and customer lessons from the Wren AI team. A couple of emails a month, no noise. ### Keep reading Product, Tutorial ### [How Wren AI Enforces Role-Based Access Control on AI-Generated SQL](https://getwren.ai/post/how-wren-ai-enforces-role-based-access-control) Wren AI doesn't replace your data security policy, it enforces it. Row-level and column-level security are applied in Wren Engine at query time, driven by who is asking, even when the SQL was written by an AI. Insight, Trend ### [The Dashboard Was Never the Asset](https://getwren.ai/post/the-dashboard-was-never-the-asset) For twenty years we treated the dashboard as the deliverable, but the durable asset was always underneath it: what your numbers actually mean. Dashboards depreciate. Context compounds. Insight, Product ### [The Missing Context Layer for AI Agents Over Business Data](https://getwren.ai/post/the-missing-context-layer-for-ai-agents-over-business-data) Why we rebuilt Wren Engine from a context layer into an open context engine, and what we learned along the way