Chief Data Officer / Australia
Three decisions that narrowed how data was used
- Annual revenue
- A$342m
- Year to June 2026
- Connected machines
- 6,400
- Planters and sprayers sending operating data
- Telemetry renewal rate
- 90%
- Up from 81% after the 2022 grower data terms
Decisions & consequences
01Retiring 21,500 part numbers and delaying a sprayer to do it
In 2019 our dealer parts catalogue carried about 83,000 part numbers, and roughly one parts line in fourteen ordered by dealers came back as wrong or unsuitable. Many numbers were duplicates created when engineering released a revised assembly without superseding the old one. Others described the same bearing under three suppliers' references. Dealers' parts interpreters had developed workarounds, but growers waiting during sowing paid for every wrong guess.
The stakes were commercial as well as practical. Parts earned about a fifth of our revenue, and two independent dealers had told our sales manager that they were steering growers towards another brand's planters because its parts were easier to identify. I proposed a six-week freeze on new part numbers while my master-data team and engineering cleared duplicates and wrote supersession chains. The engineering manager disagreed strongly: a new 36-metre trailed sprayer was due for release before an autumn field day, and the freeze would delay it. The managing director supported the freeze after we showed him returns by dealer.
We retired 21,500 numbers, linked each one to its replacement and rebuilt 1,900 illustrated catalogue pages. We also changed the release process so that no new number could be published without a supersession decision from engineering. The sprayer reached dealers five weeks late and missed the field day; sales estimated that eleven early orders went to competitors.
Within a year, wrong or unsuitable returns fell from 7.4 per cent of dealer parts lines to 3.2 per cent, and back-orders caused by ambiguous numbers almost disappeared from the service log. Engineering now treats supersession as part of design release rather than a task for the parts department.
What I would do differently: I would involve dealers' parts interpreters from the first week. We discovered late that about 2,300 retired numbers lived on in independent dealers' own systems, and some dealers kept ordering against them for months.
Giving growers control of their paddock records before selling insights to dealers
By early 2022 about 4,900 connected planters and sprayers were sending data to us under terms written when telemetry was mainly a warranty tool. Those terms allowed broad internal use. Our sales team had designed a dealer subscription, expected to earn about A$1.1 million a year, that would show each dealer the hectares sown and seeding rates of machines in its territory. A grower group near Moree heard of the plan at a field day and wrote to our managing director asking who had agreed to their paddock records being sold.
The question was fair, and the old terms did not answer it well. I proposed three data classes. Machine-health data would stay with us for warranty, service and engineering. Agronomic records, such as rates and sowing maps, would belong to the grower and be shared only on the grower's instruction. Location would be held at paddock level for service and never passed on without consent. The general manager of sales and one of the fund's directors argued that dealer insights were the main commercial return on our telemetry investment and that opt-in sharing would make the product unviable. They were right about the product, and we withdrew it.
We rewrote the terms with our lawyers, built consent settings that both dealers and growers could see, and reissued the terms to every connected customer through their dealer. The work took nine months, longer than it should have.
Within twelve months, 54 per cent of connected growers had chosen to share agronomic records with their dealer. Telemetry renewals rose from 81 to 90 per cent, and two large grower groups reconnected 230 machines they had switched off. Dealers now receive shared records with each grower's knowledge, which has made their conversations easier.
What I would do differently: I would publish the data classes at the point of sale before building any product on top of them, and bring growers into the drafting from the start.
Moving a failure model from growers' phones to the pre-season parts order
In 2023 our analytics team built a model that predicted failures of the vacuum-fan drive on our precision planters from hydraulic pressure and fan-speed patterns. We sent its alerts to growers through the telemetry app. About four in ten alerts were not followed by a failure that season, and growers began to ignore them. The service general manager wanted to keep the alerts and sell an uptime subscription at A$420 per machine a year.
The stakes sat in a narrow window. Most of our planters work between September and January, and a planter stopped for three days during sowing can cost a grower the moisture he was waiting for. I argued that the model was more reliable across a branch's population of machines than for any single planter, and that its best use was in the pre-season parts order. The service general manager disagreed, believing growers would value individual warnings once accuracy improved. The managing director asked us to run one summer each way at different branches; the parts approach performed clearly better, and the subscription was shelved.
From August 2024 the model set pre-season stock of drive kits at our nine branches and at 23 independent dealers who chose to take part. Branch technicians also used it to decide which planters to inspect during pre-season service visits. The approach carried about A$640,000 of additional inventory, and dealers later returned A$85,000 of unsold kits under an agreed buy-back.
In the 2024–25 summer planting, same-day availability of the drive kit rose from 69 to 91 per cent, and average breakdown-to-repair time for that failure fell from 3.8 days to 1.6. Grower alerts now go out only when a fault is both imminent and specific.
What I would do differently: we spent fourteen months improving accuracy before anyone asked which decision the model should change. I would put that question first and measure the result against it from the beginning.