Agentic BI Reporting Services

Agentic BI is business intelligence run by AI agents. A leader asks a business question in plain words, in the chat app they already use; the agents pick the governed metric, write the SQL, check it before it runs, query your own data and reply with the answer, a chart and its sources. Unlike a dashboard, it answers questions nobody built a chart for.

Each leader asks about their own area and gets an answer to act on: the CEO which regions are behind plan and why, the CFO what pulled gross margin down, the COO where orders get stuck, the CRO which deals may slip, the CMO which channels bring customers who stay, the CHRO which teams are losing people. Every answer comes from the same governed metrics, with the evidence, the decision it supports and the number to watch.

The ones your company already uses: Slack, Microsoft Teams, Google Chat, Mattermost, Webex, Zoom Team Chat, Rocket.Chat and Discord; WhatsApp Business, Telegram, Zalo, LINE and Viber; Lark, DingTalk and WeCom; or email, a web portal and your own in-house chat, connected through an API and webhooks. Every channel asks the same agents and reads the same governed metrics.

Yes, because the agent cannot invent a metric or run unchecked SQL. It picks metrics defined in your governed semantic layer, a validator checks the schema and the joins before any query runs, every answer shows its sources, and answers on regulated metrics go to a person first. Before launch it is tested on an evaluation suite of 80 to 150 baseline questions against a target accuracy threshold.

A dashboard answers the questions someone planned for. Conversational analytics and text-to-SQL turn a question into a query; agentic BI adds the steps that make the answer safe to act on. Six agents plan the question against governed metrics, write and validate the SQL, run it, chart the result and explain it with its sources, and answers on regulated metrics go to a person for approval.

Yes. Van Data Team builds and runs it for you, founder-led: you bring the reports your team rebuilds and the questions they answer, the audit turns them into governed metrics, and the agents answer in your chat app, so nobody in-house writes SQL or maintains dashboards. If the data still sits in separate tools, the same team builds the pipelines first. After launch, upkeep is a fixed monthly scope.

2 to 8 weeks to go-live, depending on the project. Four things set where it falls: how many sources feed the answers and how clean they are, whether your metrics already live in a semantic layer, how wide the first release is, and how formal your security review is. The audit dates your plan before any build starts; after go-live come 30 days of tuning, and large scopes go live one domain at a time.

Van Data Team quotes each agentic BI build at a fixed price after a free 30-minute consultation, once your warehouses, metrics and chat apps are known. One warehouse and one channel is the smallest scope; several warehouses with strict compliance needs is the largest. LLM and cloud usage are billed by your providers, and the proposal estimates them for your query volume.

No. The agents sit next to Tableau, Power BI or Looker, and your dashboards stay where they are. dbt metrics, MetricFlow, Cube or LookML become the agents' semantic layer, Airflow keeps orchestrating and Fivetran keeps loading sources; the build adds an MCP server, the agent runtime and an evaluation suite beside them.

A built-in assistant is often enough when all the data sits in one warehouse, the metrics are simple and people already work in that tool. A custom agent fits when answers span several warehouses or tools, need your own metric definitions and validation, or have to reach people in their chat apps, a web chat or your product.

Yes, when access is set up for it. We work in your cloud with least-privilege access and your warehouse row-level security, so each leader sees only what their role allows; personal data stays in your environment, every tool call the agent makes is logged, and an NDA and a data processing agreement are signed before access is granted.

Regular upkeep: new evaluation cases, prompt tuning, new metrics in the semantic layer, monitoring and incident response. Van Data Team quotes it as a fixed monthly scope sized to the hours you need.

You do. Once the invoice is paid in full, everything written for you, the reporting agents included, is yours; the existing tools we bring in are licensed to you permanently and royalty free, and cloud, API and model accounts are opened in your name. Your data and documents stay confidential during the project and for three years after it, and a data processing agreement is signed before any personal data is shared.

In English, in writing and on calls, on Vietnam time (UTC+7), which covers the APAC working day, the morning in the UK and Europe, and the start or end of the day in the US. The founder who scopes the work also builds the reporting agents, so there is no account manager between your team and the engineer, and progress is visible in your own repository and tools.