Inherent errors associated with machine tools (MTs) significantly hinder its performance. Amongst all the MT errors, thermal error (TE) contributes majorly to the part tolerance band. However, researchers have adopted several solutions to deal with TE, which need (a) significant design upgradation, (b) high computational resources for simulation, or (c) extensive experimental trials for empirical modelling of TE prediction. A thermally stable MT design is an effective solution but requires considerable capital investment; other compensation approaches need real-time temperature input and specially designed electronic modules to predict and compensate for thermal error based on extensive simulation or experimentation data. Hence, this work proposes a novel thermal compensation strategy based on the reliability concept to avoid the complexity of existing approaches in dealing with TE. The proposed approach uses experimental data containing the TE variation with time for particular machining parameters. Then, an allowable value for the TE was defined, and failure events (when the TE crosses the integer multiple of the allowed value) were identified. Furthermore, the time-to-failures is fitted with Weibull distribution to find the most probable failure times, which was used to develop a TE prediction model. The developed prediction model does not require real-time sensor input and avoids a specially configured electronic module for the prediction and compensation. The proposed approach was demonstrated on a turning centre for three spindle speeds, namely 1000, 2000, and 3000 rpm. The TE prediction model showed more than 99% accuracy in prediction to keep the TE under the allowed value.