This paper describes the approach that has been used by the Australian Bureau of Agricultural and Resource Economics and Sciences (ABARES) to produce quarterly forecasts since 1948 for Australia’s most important agricultural commodity markets. The Australian Agricultural Forecasting System (AAFS) is comprised of a database, a group of loosely-coupled commodity-specific balance sheets and a system for publishing forecasts. AAFS has evolved from decades of design choices that have revolved around the competing methodological merits of balance sheets and structural models. Balance sheets have emerged as the preferred method because they provide an efficient means of forecasting in their own right, as well as a means of incorporating insights from other forecasting methods and expert judgement. An issue for ABARES has been that the systems attributes of AAFS have at times gone unrecognised and proven to be incompatible with conventional approaches to management. Recognising the systems characteristics of AAFS has allowed the transfer of management principles from a range of literatures that study complex systems.