Commercial analytics and data analysis insights.

AI hasn’t made me an expert in every discipline. It has made me competent enough across several of them to keep building, learning and moving.

Candidates use AI to tailor applications while employers use AI to screen them. Everyone is acting rationally, and the system is getting worse.

I built synthetic analytics tasks with hidden data traps to test whether AI could spot bad joins, missing data, misleading uplift and changing metric definitions before making confident recommendations.

At Bakers Delight, bread campaigns produced limited incremental growth, while pizza and scone promotions changed transaction size and delivered stronger uplift.

AI can speed up analysis, coding and exploration, but a plausible answer is not automatically a trustworthy one. This is the technical validation process I use before relying on AI-assisted analytical work.

A practical guide to diagnosing commercial performance using distribution, rate of sale, price, product mix, customer behaviour and multiple data sources.

A practical framework for deciding whether data is safe to use, from definitions and reconciliation to comparability, context and the limits of the evidence.

How to design dashboards, reporting systems and internal tools around the decisions people need to make, rather than the data that happens to be available.

AI can accelerate coding, exploration and review, but using it well creates two separate responsibilities: validating the technical output and applying human judgement to the decision.

A practical guide to using Australian Census and city data across changing geographies, definitions and time periods to understand local demographic change.

Sales falling is a result, not an explanation. Start with distribution, rate of sale, price, product mix and the wider category.

Before choosing a metric or chart, name the decision the dashboard needs to improve.

A granular projection produced an estimate of 52 million Australians. Rolling the result back up to the national total exposed the problem.

AI can write code, test hypotheses and speed up analysis, but it cannot see all the domain knowledge, internal context, relationships and organisational constraints that determine whether a recommendation makes sense.

Repeated questions, changing inputs, and valuable context are signs that an analysis may deserve an interface.

Before checking nulls and duplicates, make sure the metric, population, and time period mean what you think they mean.

Two years of City of Melbourne pedestrian-sensor data show foot traffic recovering unevenly: strongest at the residential fringe, after dark, and around the events calendar, while cafe and restaurant seating still clusters around the old daytime pattern.

The third city in the series. Brisbane's demographic profile is structurally different from the southern capitals, and the data shows a city in transition.

The second city in the multi-city census series. Sydney's migration footprint is different from Melbourne's in ways the data makes clear - from China's sustained lead in birthplace to Arabic as the standout language.

This is the first in a multi-city Census series. Melbourne is diversifying quickly: India has overtaken England as the largest overseas-born population, while Christianity remained narrowly ahead of secular identification in 2021 and the linear projection places secular identification ahead by 2026.

Before I build a dashboard, automate a report, or start analysing trends, I try to answer one question first: can this dataset actually be trusted? Here's what I check before relying on data for decisions.

Australian census geography changes over time, which makes suburb and LGA trend analysis much harder than it looks. Here's how I used SA1 data, ABS allocation files, and correspondence files to build a consistent 2011–2021 time series and a 2026 projection.

Since 2005, Lego has tripled the number of sets it launches each year, while the catalogue has shifted towards larger, higher-priced adult and licensed products. Brickset data makes the range change visible, although it cannot show which shift contributed most to revenue growth.

A core product range declined 48% in one month. No single dataset could explain why. Here’s how combining scan data, consumer panels and custom research identified the root cause and helped return the range to 18% growth the following year.