Human-centered design systems and product experiences
Core services
01UI/UX Design02Web Development03Software Development042D/3D Experiences05App Development06AI Agents & Automation07Website DesignTechnical services
Frontend Engineering
React DevelopmentNext.js DevelopmentWeb ApplicationsEnterprise DashboardsSaaS PlatformsBackend & Infrastructure
API DevelopmentCloud InfrastructureDevOps & CI/CDBackend SystemsCustom SoftwareIndustries we serve
Technology & Software
AI Business Data Analyst Agent Development
A question in words, an answer with its query
An agent that answers business questions from your own data.
An agent that answers business questions from your own data. It uses the metric definitions your company agreed, runs the query against governed data, shows the number with how it was calculated, and refuses to answer from data the asker may not see.
Every team waits for the analyst. A simple question, how many orders came from that region last quarter, joins a queue behind twenty others. Self-service dashboards were meant to fix this and created a different problem: three dashboards, three numbers, because each defined an order differently. A fast wrong answer is worse than a slow right one, and that is the risk with any tool that turns a sentence into a query.
This page describes a workflow we propose to build as a custom project. It does not exist as software yet. It would reuse the access rules from our internal knowledge demonstration, where what the asker may see is settled before anything is retrieved, and the practice of showing the source behind every answer. Connections to a data warehouse, a metrics layer and query generation are not built.
Six jobs it carries through
Question understanding
Works out what is being asked.
- Maps words to your defined metrics
- Asks when a term is ambiguous
- States the period and filters it will use
- Refuses to guess a definition
Governed query
Queries data through your approved layer.
- Uses your metrics layer or approved views
- Read-only access
- Limits on cost and size
- No raw tables outside what you approve
Access control
Applies the asker's permissions.
- Identity from your directory
- Row and column rules from your warehouse
- Refuses data outside their scope
- Says that it refused, without revealing the data
Shown working
Shows how the number was produced.
- The definition used
- The query that ran
- The data's freshness
- Caveats on small samples
Explanation
Explains the result in plain words.
- What the number means
- How it compares with the last period
- What it does not tell you
- A chart where it helps
Reusable answers
Saves good questions for next time.
- Stores approved question and query pairs
- Re-runs them on request or on a schedule
- Flags when a definition changed
- Shared with those allowed
Where the Time Goes Now
The analyst as a bottleneck
Simple questions wait days.
Three numbers for one question
Each dashboard defines the metric differently.
Queries nobody can check
An answer arrives with no way to see how it was produced.
Data people should not see
A question about one team returns another team's salaries.
Confident nonsense
A tool invents a column or joins two tables wrongly.
Reports that are never reused
The same question is answered from scratch each quarter.
One Case, Start to Finish
An example of one case, in 6 steps. The same steps play in the picture at the top of this page.
Question asked
A sales manager asks how many new customers the northern region won last quarter.
Terms resolved
"New customer" has a company definition. The agent states it and the quarter's dates.
Access checked
The manager may see their own region. The query is limited accordingly.
Query run
The agent runs a read-only query through the metrics layer and gets the count.
Answer shown
The number is given with the definition, the query, the data's freshness and last quarter's figure.
Saved
The manager saves it as a monthly report, which will carry its definition every time.
Walk One Real Case Through It
Thirty minutes, one case from your own process. We map it against the workflow on this page and tell you plainly what can be reused from our demonstrations and what has to be built for your systems.
What It Reads and Writes
The systems named here are the kind this agent works with. Which ones it connects to in your case is settled in the scope.
Data warehouse
- Snowflake
- BigQuery
- Redshift
- Postgres
Metrics layer
- dbt
- Looker
- or your semantic layer
Identity
- Your directory and warehouse permissions
Where people ask
- Slack
- Microsoft Teams
- your BI tool
Definitions
- The metric definitions your company agreed
Saved reports
- A library of approved questions
Every Number Comes With Its Definition and Its Query
The agent answers only through the data and definitions your data team published. If a term has no agreed definition, it asks instead of inventing one. Each answer shows the definition, the query and how fresh the data is, so an analyst can check it in a minute.
Access is applied in your warehouse, from the asker's identity. Data they may not see is not queried, and the refusal does not hint at what it contains.
- Only published definitions and views
- Read-only, with cost limits
- Access enforced in the warehouse
- The query shown with every answer
- Ambiguous terms asked about
- No recommendations, only results
What Buyers Ask First
How does it avoid giving wrong numbers?
By narrowing what it may do. It works through the metric definitions and views your data team approved, asks when a question is ambiguous and shows the query for every answer. It does not write queries against raw tables it was never given.
How is this different from the marketing reporting agent?
Marketing reporting produces one recurring report for one function, with fixed metrics. This agent answers ad hoc questions from anyone, about any governed data, within their permissions.
How is this different from the internal knowledge agent?
The knowledge agent answers from documents. This one answers from structured data, with a query. "What is our refund policy" goes to the first, and "how many refunds did we issue last month" to the second.
What exists today?
Not as software yet. This page describes a proposed workflow. It would reuse the access rules from our internal knowledge demonstration, where what the asker may see is settled before anything is retrieved, and the practice of showing the source behind every answer. Connections to a data warehouse, a metrics layer and query generation are not built. A project starts by mapping your process, then proving the first steps on test data before anything touches a live system.
Bring the Ten Questions Your Analysts Hear Most
And the definitions behind them. Where a definition is missing or disputed, that is the first thing to settle, with or without an agent.