Skip to content

Data and analytics

One trusted data model across your ERP, CRM and store, with dashboards and plain-English questions on top, and the query behind every answer.

What this usually looks like before we start.

  • Three reports, three different revenue figures.
  • Every month-end report rebuilt by hand in a spreadsheet.
  • A dashboard nobody opens because nobody trusts it.

What we build.

  1. Modelled on the questions, not the source tables

    We start from what the business cannot currently answer and build the warehouse against that.

  2. A question box over the top

    Ask it the way you'd say it out loud, and get the answer, the chart, and the query it ran.

  3. Every answer shows its working

    The SQL is on screen, so a number can be checked before it is trusted, and read-only access means a question can never change a record.

Migration rehearsal

SetOld systemNew system
Customers4,1824,182
Open invoices917917
Opening balances2,412,6002,412,600

Every set reconciles

  • Modelled on the questions, not the source tables

    We start from what the business cannot currently answer and build the warehouse against that.

    Migration rehearsal

    SetOld systemNew system
    Customers4,1824,182
    Open invoices917917
    Opening balances2,412,6002,412,600

    Every set reconciles

  • A question box over the top

    Ask it the way you'd say it out loud, and get the answer, the chart, and the query it ran.

    Ask the warehouse

    Which product lines grew fastest this quarter?

    • Cartons+92%
    • Sheets+61%
    • Trays+44%
    • Inserts+28%
    Show the query it ran
    select line, round(100.0 * (q3 - q2) / q2, 1) as growth
    from product_quarter
    where fy = 2026
    order by growth desc
  • Every answer shows its working

    The SQL is on screen, so a number can be checked before it is trusted, and read-only access means a question can never change a record.

    Trial balance, August

    AccountDebitCredit
    Sales—2,486,400
    Inventory1,942,800—
    Receivables2,869,500—
    Payables—2,325,900
    Balanced4,812,3004,812,300

How we work on it

  1. 01Describe the question your data should already answer.
  2. 02The model, the semantic layer and what has to reconcile.
  3. 03A working demo every two weeks, answering your own questions.
  4. 04Live, with the model and the semantic layer handed over as documents.
  5. 05We run the pipelines and extend the model as the questions change.
How we work

Retail and e-commerce

A 2.5-million-row warehouse learns to answer in plain English.

Anyone can ask a 2.5-million-row warehouse a business question in plain English.

A read-only AI analyst over a PostgreSQL star schema that answers business questions in plain English and shows its SQL.

Read the case study
A woman working at a laptop, with floating panels beside her showing a database linked to tables, a SQL query, a bar chart and a donut chart.

Tell us the question your data should answer.

Tell us what's hard to run, or what you need built. On the call we work out what's really going on, show where AI fits and where it doesn't, and you leave knowing the next step.

We reply within one business day. If we are not the right fit, we will tell you.