Five disciplines for records held on trust

Telemetry from a planter records engine hours, hydraulic pressures and fault codes, but it also records where a grower sowed, at what rate and on which day. Those are different kinds of information and deserve different rules. I separate machine-health data, which we use for warranty, service and engineering, from agronomic and location records, which the grower controls and shares with a dealer or agronomist only on instruction. The terms are written in plain English and given at the point of sale, with a one-page schedule listing each data class, its purpose and how long we keep it. Legal colleagues advise on our privacy and contract obligations. My part is to make sure the systems enforce what the document says, because a term the database cannot honour is a promise we should never make.

Toowoomba builds complete machines and Dalby builds row units and hydraulic assemblies, so a raw count of defects tells a misleading story about either plant. I prefer a measure a customer would recognise: first-season warranty claims per hundred machines, traced to the plant and build week that produced the failed assembly. Beneath that shared measure, each plant keeps its own detailed defect codes, because a paint run at Toowoomba and a seed-tube misalignment at Dalby need different language. My team maintains the mapping between local codes and the shared customer-facing categories and reviews it with both quality managers each quarter. When the comparison shows a gap, the first question is whether the two sets of records are equivalent. Only after that do we discuss whether the work is.

A failure model earns its place when someone does something different because of it. Before we build one, I ask the service or parts manager to name the decision, its timing and what they would accept as evidence. For our planters the useful decision is usually pre-season: which kits to stock at which branch, and which machines to inspect in August before summer sowing. A mobile alert in the middle of planting is rarely the right moment, and a false alarm then costs goodwill that is slow to recover. I report a model's performance in the units of its decision, such as same-day parts availability or breakdown-to-repair days, alongside the usual accuracy measures. If the decision has not moved after a season, we change how the model is used or retire it.

Independent dealers sell our machines beside other brands, and they will not tolerate a catalogue that sends the wrong seal to a grower at night. I treat each part number as a controlled asset. Engineering owns its creation and supersession, my master-data team owns its description and illustration, and the parts business owns price and availability. We publish a weekly change file into dealers' own ordering systems instead of expecting their staff to search our portal. Product data for sales follows the same discipline: specifications, configurations and compatible options come from the released engineering record, never from a brochure draft. When a dealer finds an error, it is reported through one route and the dealer can see when it is corrected. That loop has done more for accuracy than any single clean-up.

At Dalby, a camera checks seed-tube alignment on planter row units before they leave the assembly cell. Its role is to direct a trained inspector's attention; acceptance stays with the inspector. Before the trial began, we agreed with the plant's workforce representatives that images and results would be used to improve the process and the fixtures, not to rank individual assemblers. Images are deleted after ninety days unless linked to a warranty investigation. I report missed defects and false flags to the quality committee monthly, broken down by row-unit variant, because most training images came from a single variant. Any extension to Toowoomba will depend on evidence gathered under that plant's own lighting, fixtures and product mix, not on the Dalby results alone.