Purpose: Current industry practices illustrate there is no standard method to estimate the number of hours worked on maintenance activities; instead, industry experts use experience to guess maintenance work hours. There is also a gap in the research literature on maintenance work hour estimation. This paper investigates the use of machine-learning algorithms to predict maintenance work hours and proposes a method that utilizes historical preventive maintenance order data to predict maintenance work hours. Design/methodology/approach: The paper uses the design research methodology utilizing a case study to validate the proposed method. Findings: The case study analysis confirms that the proposed method is applicable and has the potential to significantly improve work hour prediction accuracy, especially for medium- and long-term work orders. Moreover, the study finds that this method is more accurate and more efficient than conducting estimations based on experience. Practical implications: The study has major implications for industrial applications. Maintenance-intensive industries such as oil and gas and chemical industries spend a huge portion of their operational expenditures (OPEX) on maintenance. This research will enable them to accurately predict work hour requirements that will help them to avoid unwanted downtime and costs and improve production planning and scheduling. Originality/value: The proposed method provides new insights into maintenance theory and possesses a huge potential to improve the current maintenance planning practices in the industry.