The AMS Integrated Management Model: A decision-support system for automatic milking systems

[Display omitted] •First decision-support system developed for pasture-based automatic milking systems.•Uses data from 37 farms across Australia, Ireland, New Zealand, Chile and Argentina.•Differences between actual and predicted physical performance ranged from 2 to 14%.•Farmers agreed that the decision-support system is useful and easy to use.•Applications include investment planning and system performance optimisation. There is a significant opportunity to improve profitability and productivity in pasture-based automatic milking systems (AMS). A decision-support system (DSS) is required for AMS that can integrate key mechanics of dynamic biological processes with farm economics. Here we developed and evaluated a web-based DSS named the Integrated Management Model (IMM) designed for assisting AMS farmers and their advisors to evaluate and improve the physical and economic performance of their businesses (https://bit.ly/MilkingEdgeAMSTool). The IMM comprises a series of empirically determined predictive equations derived from the main drivers of productivity and profitability in AMS, together with stochastic simulation and optimisation modelling. The equations and models in the IMM were developed using two data sources available through the Milking Edge project: (a) an annual physical and economic dataset collected from 14 AMS farms across Australia (from years 2015 to 2018), and (b) a monthly physical dataset from 23 AMS farms located across Australia, Ireland, New Zealand, and Chile (years 2015 to 2019). Model predictions were evaluated using two similar datasets: (a) an annual physical and economic dataset from 11 Australian AMS farms (2018 to 2020) and a monthly physical dataset from 20 AMS from Australia, New Zealand, Ireland, and Argentina (2019 to 2020). The DSS was tested by 11 AMS farmers that provided feedback using the technology acceptance model framework. Results from the model evaluation showed the accuracy of the equations and simulations to predict physical variables such as milk harvested (kg milk/robot.d) or milking frequency (milkings/cow.d) was reasonably good (2–14% differences between observed and predicted values). The economic equations, which predicted operating profit margin (OP) (%) and return on total assets (ROTA) (%), could determine the relative changes and direction of profitability when physical variables change. However, the accuracy of these equations to estimate absolute values was low (ROTA: R2 0.26 and MAE 2.0%; OP: R