Playground

Don’t take my word for it. Drive it yourself.

Six of the models behind my case studies, stripped back and made interactive. Move the inputs and watch the numbers move — the arithmetic is written out beside every result, so you can check my working rather than trust a headline.

Every demo runs entirely in your browser on synthetic data. Nothing you type is sent anywhere.

Automation ROI

What is manual reporting actually costing you?

The first thing I work out on any automation engagement. Put your own numbers in — the formula is shown underneath, so nothing is hidden in a black box.

per week

salary plus overhead

be conservative — I would rather under-promise

given back per year
returned to the team
to pay back the build
Time spent today
Cost of that time
Saving per month
In working days

The arithmetic

This is the model behind the automation & BI practice, where a −80% cut in manual effort came out of exactly this sum.

KPI explorer

One dashboard, eight questions.

How a real executive dashboard behaves: one chart, a metric switcher, and a written reading of what the series actually says. Hover or tab through any point for its value.

The same pattern as the logistics control tower — one screen a leadership team can actually run a meeting from.

SQL console

Ask the warehouse a question.

A sandbox with four tables — orders, customer_monthly, inventory_health and campaigns. Run one of the examples, or write your own SELECT and see what comes back.

Try one

Queries run against an in-browser fixture — there is no database behind this page, and nothing you type leaves your machine.

Churn scorer

Which account is about to leave?

A weighted scorecard, not a neural network — which is the point. Four signals, visible weights, and a verdict a customer-success lead can act on the same morning.

Signal weights:

risk score

Read the full build in the churn early-warning case study.

Forecast horizon

How far ahead can you honestly see?

Push the horizon out and watch the confidence band widen. This is the conversation worth having before anyone commits to a number twelve months out.

Point forecast
80% interval
Uncertainty
Implied growth

Method

The production version of this sits in the retail demand-forecast case study.

ETL pipeline

Watch a nightly run, end to end.

Five stages, safe to re-run, with rejected rows quarantined rather than quietly dropped. Start a run and each stage reports in as it finishes.

Stage: Idle
  1. Extract Warehouse read replica, from the last watermark.
  2. Validate Schema contract checked; bad rows quarantined.
  3. Transform Currency, labels and grain normalised.
  4. Load Idempotent upsert into the reporting mart.
  5. Publish Dashboards refreshed, run logged, team notified.
Rows read
Rows loaded
Quarantined
Tables
Run time

The same five stages run behind every pipeline in the case studies.

Next step

Want one of these running on your own data?

These are toys with the real logic inside them. The production versions read from your warehouse, refresh on a schedule and get monitored — tell me which number you are trying to move and I will tell you what it takes.

Ask the AI Advisor