Conversational analytics
Do you need a dbt semantic layer first?
You can start with one definition, not a finished dbt semantic layer. Mark Your Data Assistant answers from the metrics and saved queries you have, labels everything else ⚠️ Unverified, and shows which definition to add next.
How it works out of the box
Saved queries first, then metrics, then SQL. The assistant looks for a saved query that fits, then a metric request, and writes SQL only when neither measures the question.
The layer grows one question at a time. Each unverified question that matters can become a metric or saved query in dbt.
Checks run where you already work. The validate command runs in your dbt CI, and reports per domain which metrics and saved queries it can use, and which facts it found.
Your labels and values stay in your project. Column labels and currencies come from dbt, and value lists are reviewed by your team.
What it means for you
- Analytics engineer: You can start small, and add definitions based on real questions.
- Manager: You don't need a large modelling project before colleagues can start asking.
- Analyst: You can see which questions already have an agreed definition.
Questions and answers
Do we need a dbt semantic layer before we start?
No, you can start with one definition. Mark Your Data Assistant needs at least one metric or saved query per domain, and your team builds the dbt semantic layer one question at a time. Questions it doesn't cover yet still get an answer, labelled ⚠️ Unverified, and they show which definition to add next.
Which metrics should we define first?
Define the metrics behind the questions your colleagues actually ask. Collect those questions, then run the coverage scan in Mark Your Data Assistant, which reports each one as governed, unverified, asked back or refused. Add definitions for the unverified questions that matter most, and scan again to see coverage rise.
When is a saved query better than a metric?
A saved query is better when an important question must be answered the same way every time, especially in a fixed shape such as an order bridge from gross to net. Mark Your Data Assistant only fills in its period and filters. A metric request combines the metrics, dimensions and filters you listed.
Does Mark Your Data Assistant work with open-source dbt Core?
Yes. Mark Your Data Assistant works with open-source dbt Core: it reads the artifacts your dbt project generates and uses MetricFlow, which is open source, to compile governed SQL. It needs no dbt platform account, only a semantic layer in your project and those artifacts published where the assistant can read them.
How does the assistant know which values a column can have?
Mark Your Data Assistant knows a column's values from accepted-values tests in your dbt project and from value lists your team reviews. A command reads a column's distinct values once, leaves out recognised personal data and refuses sensitive columns. A person reviews that file before the assistant uses it.