Using financial data to enhance labour market statistics in Scotland

Results of the study

Phase 1: Coverage and representativeness

The first phase demonstrated that a large, Scotland-specific financial transaction dataset could provide sufficient geographic, demographic and socioeconomic coverage to support the feasibility study. The analysis compared the transaction dataset with population and socioeconomic reference data to assess representativeness.

The outcomes of the first phase are provided in the report Phase 1: Representativeness (PDF, 405 KB). It found broad representation across Scotland and coverage across all Scottish local authorities. The average (mean) local authority coverage was reported as 13.3%, although coverage varied between places and population groups.

Overall, when compared with official population estimates, the assessment found that the dataset provided broad coverage across Scotland and was sufficiently representative to support the development and testing of transaction-based labour market indicators. However, representation was lower for those aged 18 to 24, although it was still considered usable for the feasibility work. This limitation was more pronounced in some city local authorities. In contrast, representation of “hard to reach” areas such as rural areas and island local authorities was well represented in the sample compared to traditional surveys which struggle to represent these areas. This is acutely seen in Na h-Eileanan Siar which had the highest coverage at 19.8%.

Income values calculated from the transactions dataset were lower than reference estimates at comparable percentiles, but the distribution followed a similar pattern to the distribution of the reference data and included substantial numbers across income bands. The proportion of those whose transactions were categorised as benefits was broadly comparable with reference data, subject to differences in categorisation.

Phase 2: Defining economic activity

This phase demonstrated that financial transaction behaviour could be used to distinguish broad labour market groups, providing a proof of concept that labour market status can be inferred from behavioural financial data.

The outcomes of the second phase are provided in the report Phase 2: Clustering and Classification Feasibility (PDF, 3.8 MB). It details how the source data did not include a record of each customer’s labour market status and tested whether patterns in income and expenditure could be used to infer broad labour market groups.

The analysis created transaction-based features using rolling periods, combined income and expenditure measures, frequency patterns, and change over time. Following refinement, 26 features were used in the final clustering. The study combined two approaches: expert-designed rules based on features associated with economic activity, and unsupervised k-means clustering to identify groups with similar financial behaviour.

Three financial-behaviour clusters were identified. When the rules-based labels were compared with the financial-behaviour clusters, a different economic activity label was clearly dominant in each cluster.

The phase concluded that the method was feasible for further development. It also emphasised that expert choices in the rules were influential and would require significant scrutiny and monitoring if the method were taken forward for statistical use.

Phase 3: Population scaling, time series, and uncertainty

This phase successfully produced stable Scotland-level estimates of “employment”, “economic inactivity” and “unemployment or benefits reliance” across the analysis period, demonstrating the potential of transaction data to generate recurring labour market indicators.

The outcomes of the third phase are provided in the report Phase 3: Geodemographics and Uncertainty (PDF, 8.9 MB). It details how weights were applied to adjust for differences between the customer sample and the Scottish population by local authority, age band and sex. It then examined weekly estimates between March 2025 and March 2026, together with local and demographic patterns and the robustness of the classification method.

For the week ending 8 March 2026, the study estimated that 77.1% were “employed”, 12.9% were “economically inactive” and 9.9% were “unemployed or benefits-reliant” among the 18 to 64 year old transaction data population. The report calculated narrow sampling confidence intervals at national level, reflecting the large sample.

The largest difference from established labour market statistics concerned the division between unemployment and economic inactivity. The transaction-based rules classified some people receiving Universal Credit or Jobseeker’s Allowance, without identifiable salary income, as “unemployed or benefits-reliant”. Official unemployment estimates instead depend on whether a person is out of work and actively seeking work and available to start, or is waiting to start a job. Financial transactions alone do not provide the necessary information to calculate unemployment in the same way as the official unemployment measure.

Results by local authority, age and sex highlighted areas where the method may perform differently across population groups. Wider confidence intervals indicated greater uncertainty for smaller local authorities and the youngest age group, while weaker cluster assignment among women and some age groups suggested areas for methodological refinement.

Tests of clustering stability and rule sensitivity were broadly reassuring and provided confidence that the methodology could form the basis for further development and evaluation as a labour market evidence source. However, the report noted that confidence intervals captured only part of the uncertainty. They did not fully account for non-random selection to the customer dataset, classification error or differences between the study concepts and official definitions.

Phase 4: Recommendations and roadmap to ongoing data delivery

The roadmap developed in this phase demonstrated that the study had progressed beyond initial feasibility and identified a credible pathway towards a recurring analytical data source capable of producing regular labour market insights.

The report Phase 4: Project Overview, Recommendations and Roadmap (PDF, 816 KB) brings together the findings from the earlier reports and proposes a route from feasibility work towards a recurring data source. It envisages a monthly transfer of aggregated data containing weekly observations by age band, sex, local authority and economic activity label, together with information needed to monitor the classifications.

The roadmap proposed a phased approach, beginning with further development of the data and classification method before moving to regular production and ongoing quality assurance. A subsequent governance review is required to determine whether the data and methodology meet the standards required to contribute to official statistical reporting. These are proposed next steps rather than activities completed as part of the feasibility study.