We estimate the effect of the Australian JobKeeper Payment COVID-19 wage subsidy on payroll jobs and wages at the employer-level using novel administrative datasets. We find a cost per job-year saved of around $112,819 ($US80,959) over the program period, implying around 812,000 jobs were saved over this time. Weekly payroll wages were almost $1.1 billion ($US761 million) higher on average during the program period, implying wage benefits equivalent to around 60 per cent of program spending. Program effects are persistent, suggesting cumulative benefits will be larger over time. A medium-scale business cycle model featuring heterogeneous households and learning-by-doing in the production technology is derived to map estimates of costs per job-year saved to approximate output multipliers. The model generates plausible output multipliers centred around 1.3, and identifies the extent to which wage subsidies support liquidity constrained workers as a key determinant of program effectiveness.