Solutions / Analytics agent

Natural-language analytics agent

Ask why margin fell in the north region last month, in those words, and get the answer from your own warehouse in minutes — with the query it ran attached, so the number can be checked instead of trusted.

Analytics System integrations Data
1–3 min
from question to answer
Query shown
with every single answer
No queue
for one-off data questions
Who it is for

Teams with data in a warehouse and a queue of one-off questions, where the analyst is a bottleneck and the dashboards never quite answer what was actually asked.

Short answer

You ask a question about your business in plain language and get the answer from your own warehouse in 1–3 minutes instead of waiting 1–3 days for an analyst to free up. Every answer arrives with the query that produced it and the rows it counted, so a number can be verified rather than believed — which is the only way an answer like this is worth having.

The problem

The dashboard answers last quarter's question. Today's question waits for whoever writes SQL.

01

Every real question needs an analyst

The dashboard shows revenue by month. The question is why the north region dropped while the south held, split by product line — and that is a ticket, and the ticket is behind four others.

02

By the time the answer lands, the meeting is over

One to three days is a normal turnaround for a one-off pull. Decisions do not wait that long, so they get made on an impression instead of a number.

03

Nobody trusts a number without its query

Two people pull the same metric and get different figures, because one filtered out cancelled orders and the other did not. The argument that follows costs more than the analysis did.

How it works

From trigger to result, step by step.

01

Start from your data model, not from raw tables

We map the entities your team actually talks about — order, customer, region, margin — onto your warehouse, with the definitions your finance team already uses. This mapping is the project; everything else follows from it.

02

Turn the question into a query

The question is parsed against that model and compiled into SQL. Ambiguity is resolved by asking back — if margin could mean gross or contribution, it asks which, rather than picking one silently.

03

Run it read-only, with limits

Queries run against a read replica under a role that can only read, with cost and row limits. An analytics agent should never be able to write to, or bankrupt, your warehouse.

04

Answer with the working shown

You get the number, a short plain-language reading of it, a chart where a chart helps, and the exact SQL with row counts. Anyone can copy that query and check it in your own BI tool.

05

Learn the vocabulary of your business

Corrections are captured — when someone says active customer excludes trials, that definition is stored and reused. The agent gets more useful in month three than it was in week one.

Before / after

What changes on the ground.

Today, by hand
×A one-off data question takes 1–3 days
×Dashboards answer everything except the question
×Two people report the same metric differently
×Analysts spend their week on ad-hoc pulls
With the automation running
›An answer in 1–3 minutes, in the chat you use
›Follow-up questions cost nothing to ask
›Definitions are shared, so numbers agree
›Analysts work on analysis, not on pulls
What you get

Delivered, not demoed.

›A semantic model of your business mapped onto your warehouse
›Question-to-SQL with clarifying questions instead of silent guesses
›Read-only execution with cost, row and permission limits
›Answers in Slack or Teams with the query, row counts and a chart
›Documentation and a handover session — the system is yours
Built with

We build in your stack rather than moving you onto ours. The list below is what this solution most often connects to — other systems are a scoping question, not a blocker.

LLM (replaceable) Snowflake BigQuery PostgreSQL dbt Metabase Looker Slack n8n
Time to production4–8 weeks
Build priceFixed quote
First stepFree mini-audit
Honest limits

When this is not the right solution.

·If the same metric has three definitions across three systems, this will expose that rather than solve it. Agreeing the definitions is real work and it comes first — we will scope it honestly.
·For numbers that get signed — statutory reporting, audited accounts, investor packs — keep the process you have. Speed is not the priority there, and it should not be.
·If five people ask two questions a month, a Metabase dashboard and an afternoon of someone's time is a better answer than a system, and cheaper to keep alive.

Questions we get about this one

It cannot: every answer comes from a query executed against your warehouse, and the query is shown. If it cannot build a query it says so instead of estimating. Wrong SQL is possible and visible; invented figures are not part of how it works.

Bring us the process that hurts.

The mini-audit is free: we take your version of this process apart and tell you plainly whether automating it pays. If it does, you get a scope and a fixed price.