Agent
Use Your Agent to execute deep dives into your data.
If you haven't already, read What is Dimension Labs? first for the bigger picture.
What is an Agent?
Agents are a class above chatbots.
When you use ChatGPT, you're talking to a model. You type a question; it answers. But it can't reach beyond the chat box. It can't query a database, run a statistical analysis, or take any action in the world.
Using the best AI model available (as we do) is like driving a Cadillac. It's a powerful machine but you are always the driver. You decide where to go. An Agent is the self-driving software behind the wheel. Give it a direction like "Get the oil changed," and it checks which navigation app it has, searches for shops nearby, then looks through your email and finds a receipt from the place you've used twice before. It routes you there instead, even though that shop sits a few miles further away. The car does the driving. The agent decides where, when, and why.
In practice you have a fully automated system for exploring your data. It still requires direction, but it move you to the position of an executive—identifying opportunities, presenting you with options, and drafting the analysis.
Causal Intelligence
The Agent can execute statistical testing to identify real and meaningful differences in the data you bring it, separating genuine patterns from the kind of variation you'd expect by chance. This gives you the combined expertise of a market researcher, a statistician, and an industry expert working together as a single collaborator. As you begin working with the Agent, it draws on these skills automatically when the analysis calls for them, and applies them directly when you ask for a specific test or comparison.
Getting Started
New to the Agent? The 1-2-3 below are a good place to start.
Agent Modes
It's helpful imagine the Agent as having two modes of interaction: Ask vs. Plan. Most conversations will involve elements of each. However, conceptually they represent different aspects of what an Agent can do.
Ask Mode
Ask in plain English. The agent translates your question into queries, joins your structured business data with your enriched conversation data and used when a quick answer is needed or you want to develop an answer iteratively over multiple turns.
Example questions:
Top drivers of churn among enterprise accounts?
Satisfaction difference: annual vs. monthly plans?
Billing complaint trend by region, last six months?
Which product issues drive negative NPS among enterprise accounts?
How do billing complaints differ by plan tier and region?
Every answer is grounded in row-level data. Every chart is drillable to the source conversations. Every result is reproducible.
A typical session runs two to six turns. When you have what you need, ask the agent to compile the conversation into a deliverable: a markdown summary, an HTML report, a CSV of the underlying rows, or chart files you can paste into a deck.
Report Mode
Define a business objective. The agent builds an analysis plan, executes it against your current data, and produces a polished report.
Reports include:
- Executive summary with the key findings.
- Methodology and statistical analysis.
- Visualizations and segment breakdowns.
- Recommendations grounded in evidence.
Reports are repeatable. Define the objective once and run it monthly. Export as HTML, PDF, or slides.
Bring supporting files when you have them: a plan file describing the report, a design file defining the visual language, a plan-support file with calculations and prior-period comparisons, or a template the output should conform to. The agent will use what you give it and ask for the rest.
What the agent does in a session
"Create a dimension that captures billing-related frustration."
It samples the data, generates a definition, and applies it across 100% of historical and future records. The result is a new structured field, not a one-off analysis.
How the agent is different from a BI tool
You don't need to know SQL or your schema. The agent figures out which fields to pull and how to join them.
It adapts to your custom dimensions and prompt configurations. Its answers reflect how your team defines the problem, not a generic template.
When your data has a gap, the agent says so and suggests how to fill it. Often that's a new dimension you can create on the spot.
It explains why a finding matters, not just what the number is.
What makes the agent yours
The agent is generic until your prompts make it yours. Custom prompts define the dimensions it reasons over. System prompts shape how it interprets and responds.
→ Configure dimensions in Custom Prompt Writing.
What the agent will not do
The agent answers questions your data can support. If a concept is not yet captured by a dimension, the agent tells you and proposes how to add it. It does not guess.
It works best with clear, specific business questions. Open-ended prompts like "tell me everything about my customers" return less useful results than scoped ones like "which customer segments are showing rising cancellation intent this quarter?"
For platform issues — login, billing, account configuration — the agent will direct you to the support chat widget. That is by design. The agent reasons over your data, not your account.