Exponential smoothing has been one of the most popular forecasting methods for business and industry. Its simplicity and transparency have made it very attractive. Nonetheless, modelling and identifying trends has been met with mixed success, resulting in the development of different modifications of trend models. We present a new approach to time series modelling, using the notion of "information potential" and the theory of functions of complex variables. A new exponential smoothing method that uses this approach is proposed, the "Complex exponential smoothing" (CES). It has an underlying statistical model described in the paper and has several advantages in comparison with the customary exponential smoothing models, that allow CES to model and forecast effectively both trended and level time series, effectively overcoming the model selection problem.