The stepped wedge cluster randomised trial is a relatively new study design that is increasingly used in the evaluation of service delivery interventions. It is an alternative to parallel cluster trials, which are commonly used for evaluating interventions delivered at the cluster level. In a stepped wedge design, clusters are sequentially randomised to switch from control to intervention, with all clusters eventually exposed to the intervention. This design allows for data collection throughout the study, with each cluster contributing observations under both control and intervention periods. It is a pragmatic design that can reconcile the need for robust evaluations with political or logistical constraints. The stepped wedge design has been used in various settings, including the Gambia hepatitis intervention study, which is one of the earliest and most well-known examples. Two systematic reviews have shown that the use of this design is increasing, with applications in areas such as HIV, cancer, healthcare-associated infections, social policy, and criminal justice. The stepped wedge design is particularly suited to evaluations that do not rely on individual patient recruitment. However, it is at risk of selection bias if individual recruitment is not concealed or if the intervention is not blinded. The design may confound the effect of the intervention with underlying temporal trends, so sample size calculations and analysis must account for both the clustered nature of the design and the confounding effect of time. The stepped wedge design is more powerful than parallel designs when there are substantial cluster-level effects or when clusters are large. It is also more efficient than parallel designs with a baseline period when the intra-cluster correlation is large. However, when the intra-cluster correlation is small, a parallel design may be more powerful. The stepped wedge design allows for the evaluation of interventions that cannot be implemented simultaneously in all clusters. It is particularly useful when the intervention is complex and must be rolled out sequentially. The design also allows for the examination of the development of the intervention's impact over time and the exploration of treatment effects between clusters. The stepped wedge design requires the fitting of more complex models than parallel designs and must adjust for underlying temporal trends. It is recommended for evaluating service delivery interventions where outcomes are based on routinely collected data, as it eliminates the need for individual participant recruitment. When individual recruitment is required, special attention should be given to minimising selection bias. The stepped wedge design is a randomised method of evaluating interventions delivered at the cluster level. It is particularly useful when randomisation to either control or intervention arm is precluded but randomisation to a date of initiation is possible. However, the design requires more complex models and must adjust for underlying temporal trends. It is recommended for evaluating service delivery interventions where outcomes are based on routinely collected data.The stepped wedge cluster randomised trial is a relatively new study design that is increasingly used in the evaluation of service delivery interventions. It is an alternative to parallel cluster trials, which are commonly used for evaluating interventions delivered at the cluster level. In a stepped wedge design, clusters are sequentially randomised to switch from control to intervention, with all clusters eventually exposed to the intervention. This design allows for data collection throughout the study, with each cluster contributing observations under both control and intervention periods. It is a pragmatic design that can reconcile the need for robust evaluations with political or logistical constraints. The stepped wedge design has been used in various settings, including the Gambia hepatitis intervention study, which is one of the earliest and most well-known examples. Two systematic reviews have shown that the use of this design is increasing, with applications in areas such as HIV, cancer, healthcare-associated infections, social policy, and criminal justice. The stepped wedge design is particularly suited to evaluations that do not rely on individual patient recruitment. However, it is at risk of selection bias if individual recruitment is not concealed or if the intervention is not blinded. The design may confound the effect of the intervention with underlying temporal trends, so sample size calculations and analysis must account for both the clustered nature of the design and the confounding effect of time. The stepped wedge design is more powerful than parallel designs when there are substantial cluster-level effects or when clusters are large. It is also more efficient than parallel designs with a baseline period when the intra-cluster correlation is large. However, when the intra-cluster correlation is small, a parallel design may be more powerful. The stepped wedge design allows for the evaluation of interventions that cannot be implemented simultaneously in all clusters. It is particularly useful when the intervention is complex and must be rolled out sequentially. The design also allows for the examination of the development of the intervention's impact over time and the exploration of treatment effects between clusters. The stepped wedge design requires the fitting of more complex models than parallel designs and must adjust for underlying temporal trends. It is recommended for evaluating service delivery interventions where outcomes are based on routinely collected data, as it eliminates the need for individual participant recruitment. When individual recruitment is required, special attention should be given to minimising selection bias. The stepped wedge design is a randomised method of evaluating interventions delivered at the cluster level. It is particularly useful when randomisation to either control or intervention arm is precluded but randomisation to a date of initiation is possible. However, the design requires more complex models and must adjust for underlying temporal trends. It is recommended for evaluating service delivery interventions where outcomes are based on routinely collected data.