Demand forecast accuracy in the service supply chains e.g. spare parts is critical for customer satisfaction and its financial performance. This is a typical logistic network which is affected by irregular demand resulting from contract and non-contract business strategies. Hence, existing forecasting methods that work excellent with smooth and linear demand patterns become less accurate with increasing erratic, lumpy and intermittent demands. Moreover, increasing number of stock keeping units (SKUs) in service supply chains have computational limitations. This is because of the fact that demand keep on fluctuating their demand classes that result in uncertainty and consequently, leads to higher target stock levels (TSL) and lower reorder point (ROP) to ensure higher customer satisfaction. This raises interest in using AI for service supply chains to improve demand forecast accuracy. In this paper, we present a survey of existing forecasting methods used in service and non-service supply chains to select best performing AI methods and performance measures, using ABC classification. Neural network (NN) and Mean Square Error (MSE), are subsequently modelled and used in aircraft spare parts supply chain using data collected from Dassault Aviation, as a function of most commonly used aggregated demand features. The results are compared with frequently and best performing forecast methods for intermittent demand as Croston, Croston SBJ and Croston TSB; and classical methods as moving average (MA) and single exponential smoothening (SES). The analysis and results suggest that NN with higher number of features improve demand forecast accuracy significantly for intermittent demands along with reduction in associated financial implications. © 2017