Most-Detailed Granularity

If you are recording trend data on inventory, you may wish to define a metric on the most-detailed level of granularity; that is, finest level of data granularity for a problem (e.g. Ops Costs by Department and by Building) that you would ever need. In this way, the historical data you would need for any view or analysis would always be in the Metrics Trends table, and you could always define a view aggregating this data by any criteria. No matter what you needed to summarize by in order to find the source of anomalies, the data you would need to find them is present.

Preferred Granularities

Dimensions (time, location, organization) and Levels (Business Unit, Division, Department) are separate axes, and per the above, sometimes you may wish to drill into them in sequentially deeper levels of resolution.

However, each metric has particular dimensions and levels that are the most telling, that is that are the natural divisions along which particular metrics cleave. We call these natural divisions the “preferred granularities” for that metric.

For instance, energy is most often metered on a building level, and so spikes in usage often appear at the building level. Energy is most often governed on a building level, and so anomalies in management, maintenance and control (e.g. of the HVAC system) will also most likely appear on the building level. As such, the building level is a preferred granularity for energy data. Examining the top and bottom performers for energy use metrics per building is most likely to reveal the success stories and problem areas.

Avoiding Loss in Aggregation

The preferred granularities often tend to highlight issues that are normalized out if you aggregate the data by coarser levels of detail. For instance, suppose an illicit machine shop is siphoning off energy from a particular building. That extra energy usage would show by spiking that building’s metrics for energy use when examined on a per-building level. However, when summed together with other buildings for a site or city, the extra energy usage would not be large enough that it would be visible. That is to say, you can find this scale of anomaly by looking at the bottom performers, but not by drilling into the data top-down.

The preferred granularities do not cover all conditions. For instance, some buildings do have separate energy meters for different suites or labs. At other sites, groups of buildings may be served by a common steam generation plant. Or a co-generation facility in the basement of one building could skew its numbers out of the norm. However, the majority of outliers at a particular preferred granularity will represent anomalies.

Combining Levels and Dimensions

Dimensions and levels are separate axes f data; however, commonly the actionable information is often on the Dimension-plus-Level pair, since this pair connects the measured performance with the organization responsible for that performance. There tend to be only a few dimension-plus-level pairs that make sense for each metric, since there are only a few of these pairs that would produce actionable outliers. As such, it seems much easier to group the Dimension-plus-Level combinations together as a single preferred granularity.

Example: Op Ex Metric

For example, for Ops Ex you might have these as the preferred granularities (i.e. the dimensions and levels):

In terms of display, users will often want to see only one of the levels of granularity (that is on one Dimension-plus-Level pair at once) based on their purview, as that one condition will give them the only actionable data they need for running their business unit or building.