The conditions written beside every measurement

The test bench at Pelverton Pumps sat at the back of a Brisbane workshop, and in February 1997 it became my responsibility as a graduate engineer. I had finished a mechanical engineering degree at Torrenvale Institute of Technology the previous November. Each irrigation pump left the bench with a flow and pressure figure written on a card. When two cards for the same model disagreed, the explanation was usually missing from both: a warmer tank, a different casting batch, a gauge that had been knocked. I started recording those conditions beside each result. It was a small habit, and it is the one I have relied on most since.

In July 2001 I moved to Wyreeda Mills Co-operative, which ran three sugar mills in central Queensland. As reliability engineer I analysed vibration readings, oil samples and breakdown records for the crushing plant. Crushing ran from about June to November, and the major rebuilding work happened in the slack season afterwards. My early forecasts of bearing and gearbox failure were statistically respectable and practically ignored, because the shutdown scope had been agreed months before and the contractors booked. I learnt to bring a prediction to the planning meeting where it could still change the work list. I completed a part-time master's degree in applied statistics at Torrenvale during those years, mostly at the kitchen table after crushing shifts.

Callowdene recruited me in January 2007 to manage field quality and warranty. The company had been founded in Toowoomba in 1968 by a welder who built grain augers, and by then it also made planters, sprayers and chaser bins. Warranty claims arrived from dealers in free text, with part numbers that were sometimes superseded and descriptions that ranged from careful to cryptic. I designed a claim form with a short list of failure modes and asked dealers to photograph each failed part beside its tag. That gave engineering a record they could argue with, which is what a good record should invite.

From 2012 I led engineering data and parts information. It is the least glamorous work I have done and probably the most useful. A dealer's parts interpreter cannot help a grower in the middle of sowing if three numbers describe the same seal and the illustrated catalogue shows an old revision. My team carried each bill of materials from engineering release into the parts catalogue and the dealer ordering system. I learnt how much of a manufacturer's data quality is decided by who is allowed to create a new number.

In July 2016 the managing director asked me to form an analytics team, and in 2018 we began fitting telemetry modules to planters and sprayers. The volume of machine data grew quickly; agreement about who could use it did not. When an agribusiness fund bought 38 per cent of the company in 2020, its directors asked who was accountable for data across engineering, service, sales and the plants. The honest answer was several people, and therefore nobody. The board created the chief data officer role in March 2021, and I was appointed to it.

I now lead 34 people through four managers and report to the managing director. I present to the board's audit and risk committee twice a year on data governance, privacy obligations and the condition of our product and customer records. I am not a director. Decisions about machine design, safety and warranty acceptance belong to engineering, quality and service leaders; my office supplies the records, the analysis and the rules for using them.

People in our industry probably know me for the grower data terms we published in 2022, which some dealers still regard as a commercial mistake. I would make the same decision again, though faster and with growers involved earlier. I am also known, less publicly, for stopping work. We closed a dashboard programme that branch managers rarely opened and withdrew grower alerts that were wrong too often to deserve attention. Stopping something is hard on the team that built it, so I explain the reason in terms of the decision it failed to change, not the effort it consumed.

My teams include engineers, statisticians and two former parts interpreters, and I expect each of them to be able to explain a dataset to the person whose work produced it. An assembler at Dalby should be able to see how a camera check uses images of her row units, and a grower should be able to read what we keep from his planter. That standard is harder to meet than an accuracy figure. It is also what I would bring to a board: a habit of asking who the data describes, who can act on it, and who would be misled if it were wrong.

Open to Board appointments

Languages / English