Data exploration lets you ask open-ended questions about your connected data sources directly in the Matik AI Agent. Rather than knowing exactly what to query up front, you can be curious — ask what's in a source, poke at a trend, or follow a hunch — and the agent does the work of finding the answer and showing it to you inline as a chart or table.
This is more than the agent writing a query for you. Because you can ask broad, exploratory questions in plain language, data exploration is most useful when you don't yet know exactly what you're looking for. A few example use cases:
-
Understand a source you just connected.
What opportunities data does this data source include?to learn more about how data is structured and typical values. -
Discover content worth building in Matik. Explore your data to find the metrics and breakdowns worth showcasing —
What are the most interesting trends in our sales data over the last year?— then turn the strongest results into Dynamic Content to onboard into templates. -
Answer an ad hoc question fast. Get a number or a chart without building a piece of Dynamic Content just to see it —
How did renewals trend by month last quarter?
This article is for the producers and admins who set up and use data exploration for their team. It covers how to sharpen the agent's answers with data context, how to explore your sources, how to read and refine inline visualizations, and how to export or reuse results.
Setting Up Data Context
Data context is the background knowledge the agent uses when it writes queries against your data — which tables matter, what your metrics mean, and the conventions your team follows. Good context is the single biggest lever on answer quality: it's the difference between the agent guessing at your schema and knowing it.
Context is optional. You can start exploring right away without configuring anything — the agent works with what it can infer. But answers get more accurate as you add context, and you can add or update it at any time. Data sources that have no context yet are tagged NEEDS CONTEXT; once a source has at least one table described, it shows as HEALTHY.
Configuring data context requires admin access. You reach the configuration page from the settings control in the AI Agent's data exploration area. It has two tabs — Data Sources and Business Rules — which map to the two levels of context described below.
The two levels answer different questions.
- Use enterprise-wide context for things that are true no matter which source the agent queries — how your company defines a term, how you want queries written.
- Use per-data-source context for facts about a specific source — what its tables contain and which ones to leave alone.
A good rule of thumb: if it describes your business, it belongs at the enterprise level; if it describes a table, it belongs on the source.
Enterprise-wide context
Enterprise-wide context applies across every data source the agent queries. Put anything here that reflects how your organization thinks about its data, regardless of where that data lives. There are two kinds:
-
Custom Instructions: Free-text guidance that shapes the agent's behavior across all data sources. Use it for conventions that always apply — for example,
Always prefer LEFT JOIN when joining dimension tables. Edit it in the Custom Instructions section and click Save. -
Business Rules: Definitions and conventions the agent should apply when generating queries. Each rule is a short free-text definition — for example,
Active customer = has logged in within 90 days. Add rules on the Business Rules tab with Add Rule, and remove any you no longer need. Business rules keep the whole team's queries consistent by pinning down terms that would otherwise be ambiguous.
Both sections can be drafted for you. Regenerate with AI (Custom Instructions) and Enrich Context with AI (Business Rules) have Matik suggest context automatically, which you can then edit — see How Matik drafts context for you below.
Per-data-source context
Per-source context tells the agent about the tables inside a specific data source — the concrete details it can't reliably infer from table and column names alone. This is where you encode what a warehouse actually contains: which table holds deals, what a cryptically named column means, which tables are noise. On the Data Sources tab, each connected source appears as a card with its status and an Open details button. Open a source to configure two sections:
-
Tables and Metrics: Describe the tables and metrics the agent should know about for this source. Each entry has a Schema, a Table, and a description — for example,
Tracks deals; key fields Amount, StageName, CloseDate. Use Add Table to add more. A source is considered healthy once at least one table has a description. - Tables to Ignore: Tables the agent will never query or reference in this source. List the Schema and Table for anything you want kept out of exploration — deprecated tables, staging tables, or data that's off-limits.
To speed up setup, use Generate context (labeled Scan on the source card) to have Matik draft the Tables and Metrics descriptions for you — see How Matik drafts context for you below. The detail page shows when context was last generated. Tables to Ignore is always yours to maintain; Matik never fills it in for you.
How Matik drafts context for you
The Regenerate with AI, Enrich Context with AI, Scan, and Generate context actions all work the same way under the hood. Matik looks at your enterprise's most-used templates — ranked by how many presentations they've generated — and reads the queries inside their Dynamic Content, along with the live schema (table and column names and types) of the data sources those queries touch. It uses that picture of how your team already queries its data to draft context with AI.
Note: Matik analyzes your schema (table and column names and types) and the query definitions in your Dynamic Content. It does not read or send the actual rows or query results from your data sources.
Two things follow from this:
- It needs existing content to learn from. Drafting works by generalizing from the queries in your top templates. If a data source isn't queried by any of that content yet, there's nothing for Matik to draft from, and it tells you so — you'll add context for that source manually.
- Treat the output as a first draft. Generated context is inferred from query patterns, so review and edit it. You know things about your data — edge cases, deprecated fields, what a metric really means — that the agent can't infer from a query.
Regenerating affects each field differently, so save any unsaved edits before you generate:
| Field | What happens when you generate |
|---|---|
| Custom Instructions | Replaced entirely with the new draft. |
| Business Rules | Your existing rules are kept; new suggestions are added and duplicates are removed. |
| Tables and Metrics | Replaced entirely — regenerating overwrites edits you made to table descriptions. |
| Tables to Ignore | Never changed — this list is always maintained by you. |
Exploring Your Data Sources
To explore, open the AI Agent's data exploration area and type a question in plain language. You don't pick a data source first — the agent decides which connected source (or sources) can answer, runs the query, and tells you which source it used on each result.
Start wherever your curiosity is. Exploration works best as a conversation, not a single well-formed query, so questions can be as broad or as specific as you like:
-
Get oriented in a source:
What data is in this source?orWhich tables would I use to analyze revenue? -
Look for what's interesting:
What stands out about our pipeline this quarter?orWhere are we seeing the biggest month-over-month changes? -
Ask a specific analytical question:
What was monthly revenue by region for the last four quarters?orWhich accounts have the highest open pipeline right now?
Follow-up questions build on the conversation, so you can start broad and narrow in — drill into a segment that looked unusual, change the grouping, or extend the date range without restating everything.
Every result includes the exact query the agent ran in a collapsible panel, with a copy button. You never have to manually write a query, but if you work in it, you can verify the query, reuse it elsewhere, or ask the agent to debug or optimize a query you already have.
Visualizing and Understanding Your Data
Results appear inline as an interactive card. Depending on the shape of the data, the card offers up to three views, selectable with tabs:
- AI Insight: A short written summary of what the data shows.
- Chart: A visualization of the result. This tab appears whenever the data can be charted.
- Table: The full result set as a sortable, paginated table.
The agent chooses the chart type automatically based on your data — you don't have to configure it. Supported types include line charts for time series, bar and stacked bar charts for categories, pie charts for small breakdowns, dual-axis charts that combine bars and a line, and single-value KPI tiles for headline metrics.
When a result contains several metrics, a Metric selector lets you focus the chart on one metric at a time. If you want the visualization built around a specific metric, you can ask the agent to rewrite the query for it and it will return a fresh result.
Exporting and Reusing Results
Once you have a result you care about, you can take it out of the conversation in two ways.
Export to CSV
Click Export CSV on a result to download the full, un-truncated data set as a CSV file. This is the quickest way to pull numbers into a spreadsheet for one-off analysis.
Save as Dynamic Content
To turn a one-off exploration into reusable Matik content, save it as Dynamic Content. Open the actions menu on a result section and choose the option that matches what you want to keep:
- Save as AI insight — the written summary.
- Save as chart dynamic content — the visualization.
- Save as table dynamic content — the result table.
- Save as text dynamic content — a single headline value.
Saved Dynamic Content behaves like any other Dynamic Content in Matik — you can tag it into templates and presentations, and it refreshes on its own schedule. When the agent detects that a result is a good candidate for reuse, it also surfaces a This looks reusable prompt with a Save as DC button so you can capture it in one click.
Best Practices
- Describe your most-used tables first. Context has the biggest impact on the tables people actually query. Prioritize describing those, and clear the NEEDS CONTEXT status on your key sources.
- Pin down ambiguous terms with Business Rules. If "active customer," "churn," or "qualified lead" means something specific at your company, write it as a business rule so every query uses the same definition.
- Use Tables to Ignore to keep exploration focused. Excluding staging, deprecated, or off-limits tables makes the agent's queries faster and more accurate.
Comments
0 comments
Article is closed for comments.