Sequential pattern mining is a foundational technique for analysing behavioural data across domains such as digital advertising and cybersecurity. However, traditional algorithms often treat all sequential events as equally spaced in time, overlooking varied temporal proximity that can carry critical semantic meaning. This limitation leads to fragmented or less interpretable results, especially in contexts where timing is crucial. In this work, we propose TSpan, a novel sequential pattern mining method that integrates temporal compactness directly into the mining process. TSpan identifies sequences that are frequent and temporally cohesive, providing a more accurate reflection of real-world behavioural and operational patterns. We introduce two variants, TSpan-Intra and TSpan-Inter, to separately model patterns within shorter temporally-bounded user activity sessions, and across sessions. These allow us to distinguish between short-term and long-term patterns. TSpan is evaluated on two large datasets: a digital advertising dataset containing 780,000 ad observations and a cybersecurity event log. In both datasets, TSpan consistently identifies patterns that are more compact, interpretable, and impactful than baseline methods. Ad-related exposure sequences are uncovered by TSpan in the advertising dataset, shedding light on how such content clusters and unfolds throughout users’ ad journeys.