Data Analyst Apprenticeship EPA
An anonymised analytics case study showing how clinical occupancy data can support better room allocation, scheduling, and operational planning.
< 85%
Average occupancy stayed below the clinic utilisation benchmark, highlighting clear headroom for better allocation.
35.21%
MAE reduction once public holidays were removed from the forecast input.
39.13%
RMSE reduction after treating holiday noise separately in the model.
12:00-14:00
A repeatable midday lull that could inform session spacing and room release decisions.
This project was carried out to test whether sensor-derived occupancy data could provide credible evidence for improving clinical space use in a large public building setting. The aim was not only to measure utilisation, but to show where practical optimisation opportunities existed for scheduling, allocation, and future planning.
The analysis focused on a single-site Occupeye export, with core reporting centred on weekday activity between 08:00 and 18:00. Comparisons were made across anonymised areas, service group patterns, time-of-day demand, and forecast behaviour to show both current utilisation and the operational choices the data could support next.
The project combined lightweight tooling with practical operational analysis, keeping the work usable within a secure healthcare setting while still producing decision-ready outputs.
Primary source for sensor activity exports and occupancy pattern analysis.
Used for initial cleaning, reshaping, and fast preparation of large CSV exports.
Used for aggregation, statistical analysis, and forecast testing across the cleaned dataset.
Used to package findings into clearer visuals and portfolio-ready reporting outputs.
Compared demand across buildings, departments, and room groupings to find underused capacity.
Reviewed weekday and hourly demand patterns to understand where operational pressure concentrated.
Tested future usage patterns and validated model performance with MAE and RMSE.
Framed the outputs around scheduling, space pressure, and next-step data requirements.
Removed incomplete records where needed and presented results in a portfolio-safe format.
Rebuilt from the project export using weekday average activity by hour. Darker cells indicate heavier observed demand and make peak periods easier to spot at a glance.
Average utilisation by area highlights where available clinical capacity may be underused and where redistribution could be tested first.
The daily profile shows where demand is genuinely concentrated and where recurring dips create room for rescheduling or better release of spare capacity.
Weekday variation is relatively steady overall, but it still helps show where demand is lighter or heavier across the week when balancing allocations.
Removing public holidays reduced forecast error materially, which matters if the model is to support planning rather than only describe historic usage.
Differences between anonymised groups help narrow where deeper booking, room-type, or specialist-space review would add the most value.
Some records lacked room identifiers, room type, department, or area fields, which limited room-level matching and reduced the depth of certain breakdowns. The result was still strong enough for high-level utilisation measurement, while also making clear where cleaner metadata and linked systems would unlock sharper optimisation work.
Organisation-specific names, room identifiers, and sensitive operational detail have been anonymised or selectively generalised for this portfolio version. The analytical structure, method, and direction of the findings remain representative, which allows the case study to demonstrate approach and decision value without exposing confidential information.
The analysis did more than measure underuse. It pointed to practical areas where allocation, scheduling, and planning decisions could be improved once supporting operational data is linked in.
Weekday demand was not identical. Lighter periods at the edges of the week suggest some activity could be redistributed without increasing estate pressure.
Late-morning pressure was more consistent than early morning or late afternoon use, which creates scope to spread suitable sessions more evenly.
Low utilisation does not always mean low demand. It can also indicate sessions being placed in rooms with more capacity than required.
Where higher-value treatment or specialist spaces are underused, tighter allocation rules or shared access could reduce idle capacity.
Lower-use zones provide a practical pilot area for reallocation, consolidation, or overflow use before new space is commissioned.