How I work

From the question to the decision.

Two ways I work, step by step. Tap a step to see what happens there, with a real example.

Dashboards, from question to decision

  1. Step 1

    Ask the right question

    I start with the decision, not the chart. Who reads the report, what they decide, and which number tells them.

    Real example12 business areas

    Leadership needed one view of the business. The first executive dashboard covers 12 business areas.

    See the case
  2. Step 2

    Pull ERP data

    Reports run on live ERP data, the company’s core business records. No exports pasted by hand.

    Real exampleData up to 5 days old → live

    Leadership had been reviewing sales data up to 5 days old. The report now runs on live ERP data.

    See the case
  3. Step 3

    One model

    One shared data model holds the business definitions. Every report reads from it, so a metric means the same everywhere.

    Real example1 model, every production report

    Every production report now uses one shared model. Standard metric names ended conflicting numbers between departments.

    See the case
  4. Step 4

    Build the dashboard

    The first page answers the question. Each figure shows where it stands against plan, and the detail sits behind it.

    Real example10+ dashboards and reports

    10+ Power BI dashboards and reports, from sales and invoicing to inventory and production.

    Open the dashboards
  5. Step 5

    Check against SQL

    Before a report goes out, its totals are checked against SQL on the source data. Numbers get checked before they get published.

    Real example3 to 4 days → under an hour

    The monthly sales meeting report went from 3 to 4 days to under an hour, checks included.

    See the case
  6. Step 6

    Leadership decides

    The report is there for a decision. It shows the gap, who owns it, and what to fix first.

    Real example~25M AED at-risk stock found

    The inventory risk analysis flagged 26% of raw material items as high risk, with ~25M AED of stock at risk, early enough to act.

    See the case

AI automation a person signs off

  1. Step 1

    Plain-language request

    Work starts from a normal request. No special format and no form to fill in.

    Real exampleCustomer change requests

    The pricing engine starts from customer change requests on modified HVAC units, such as a bigger fan or a new casing.

    See the case
  2. Step 2

    AI reads and proposes

    AI reads the request and proposes what to change. It never does the fixed math.

    Real example37 contradictions found

    A review found 37 contradictions in the method documents. The old AI handled about 70% of fixed calculations unreliably.

    See the case
  3. Step 3

    Python does the math

    Every calculation runs in Python. Automated tests check the rules each time the code changes.

    Real example446 automated tests

    The pricing engine now runs on 446 automated tests. It closed a ~97K AED material costing gap.

    See the case
  4. Step 4

    A person approves

    The proposal goes to a person, who approves or rejects it. Anything a person must confirm is flagged.

    Real exampleEvery change signed off

    AI use stays inside company security rules: approved connectors, tools that run locally, and a person signing off every change.

    See the case
  5. Step 5

    Scheduled run

    Recurring work runs on a schedule. Data pull, checks and the finished output, every time.

    Real example3 to 4 days → under an hour

    The monthly sales meeting report runs as a scheduled AI workflow. Prep went from 3 to 4 days to under an hour.

    See the case
  6. Step 6

    Output anyone can audit

    Each run ends in an Excel workbook anyone can audit. Anything a person must confirm is flagged in it.

    Real example10 AI automations

    Manual costing work became 10 AI automations. They pull live ERP data, build Excel workbooks anyone can audit, and flag anything a person must confirm.

    See the case

What you get

Dashboards

  • One definition per metric, used by every report.
  • A first page that answers the question.
  • Live ERP data, not a deck built by hand.
  • Totals checked against SQL before release.
  • Model documentation generated from the model itself.

AI automation

  • Fixed math in tested Python, never in the AI.
  • A person signs off every change.
  • Recurring reports that run on a schedule.
  • Output anyone can audit.
  • AI use inside company security rules.

See it working

Five working rebuilds of reports I build, with made-up data. Change the period, tap a figure, open an area.

Open the dashboards
Interactive demo

Pricing Assistant demo

The AI automation flow on made-up data. AI reads a request, tested Python prices it, and you approve each change.

Built with AI coding agents

Table Extractor demo

My own tool, written by AI coding agents that I directed and tested. Watch a made-up page become an Excel table, cell by cell.

Open the demo