Average Before Average After 700 600 500 400 300 809 808 807 806 804 802 801 800 501 500 431 429 428 418 416 413 411 409 406 403 236 201 130 113 112 111 109 0 108 200 100 106 Average Monthly Consumption (kWh) 900 800 Dwelling Id Fig. 4. Average monthly non-heating electricity consumption (apartments). and in many cases only annual bills. In the absence of more granular information, the approach adopted was to deploy the monitoring infrastructure and to take actual consumption measurements for a few months from each dwelling before any intervention was introduced. An additional issue was found in handling the missing data. The key reasons for missing data were temporary outages in the network connection, and participants switching off the circuits to which the gateway and the broadband router are connected. This resulted in prolonged periods of time when no data was sent from the dwelling to the remote data collection server. Given the nature of deployment, it is difficult to have complete control over the monitoring infrastructure and therefore, any data analysis mechanism should be able to cope with missing data resulting from such outages. To work around this issue, we resorted to the use of cumulative energy consumption figures reported from the various metering devices. The advantage of this approach is that the difference between cumulative readings, at two different points in time, yields the total consumption, which implies that information is not lost. Consumption Reduction (%) 20 0 -10 -20 -30 0 20 40 60 Percentage of Households 80 100 Fig. 5. Non-heating electricity saving across the whole project. 600 Number of Hits 500 400 300 200 100 0 ICT Interventions The Bristol pilot began in January 2012 and lasted until May 2013. IEEE TECHNOLOGY AND SOCIETY MAGAZINE 10 20 40 Time (day) 60 80 Fig. 6. Interface activity over the initial three months of the trial. | FALL 2014 | 75