After looking into crazily complex possible approaches to simulating half-hourly meter readings based on sparse data (Alex mentioned Fourier Transforms!) we've plumped for the KISS principle. We simply use an array of data points from a reference building where half-hourly meter data is available. We can then generate data for a similar building by multiplying each data point by the ratio of the aggregate consumption e.g. if the reference building consumed 100kWh over a year and the similar building 200kWh then we'd just double each data point.
Once we've generated an array for a similar building we can then transform this data to show what the target consumption is based on a statement like "Oxford aims to reduce consumption by 33% by 2020". (Using for instance a reverse compound interest approach).
The difference between target and actual consumption is then the performance.
We can use the mean of the previous say 60 days performance to generate a projected consumption value, for instance for the end of the current month.
We then have all the data we need to create the Dripping Bucket or the Boat Race visualisation.