Predictive Analytics For “Low-Tech” Facilities

By Nitin Bangera
From the May/June 2016 Issue

More than 100 years ago, the invention of the internal combustion engine revolutionized automotive travel, and with it gave birth to a new industrial economy. Just as internal combustion once did, the proliferation of data science and analytics is once again reshaping the information economy and transforming how we interact with technology. Business analytics, in particular, is playing a pivotal role in improving operations at large-scale organizations. For example—UPS, a shipping service, saved $30 million in fuel by using predictive analytics. These kinds of cost savings are mouthwatering if you’re a C-suite executive. But what if you’re a facility management (FM) executive? And what if you are a federal or non-profit FM executive with limited resources?

Descriptive Versus Predictive Analytics

Before we look at FM analytics, let’s get a better understanding of the different types of analytics. At the bottom of the value and sophistication chain lie descriptive analytics that summarize recent and historical data. Examples include dashboards, graphs, statistics, trend models, etc. Moving up the chain, by adding data visualizations and drill-downs to these tools, diagnostic analytics allow us to determine the root causes of events. By definition, both these types of analytics are backward looking.

analytics
(Source: Gartner and Booz Allen Hamilton)

At the forward-looking level, there are predictive analytics. These use a variety of advanced statistical, modeling, and machine learning techniques that study historical data in order to make probabilistic predictions. One level up the sophistication chain, prescriptive analytics add tracked outcomes to suggest a best course of action. These last two types of analytics form the basis of much of the information economy. Aside from predicting what you want to buy next, these analytics are also drastically changing the dialogue within the financial, transportation, health, manufacturing, and social sectors.

Are Analytics Changing Facility Management?

The FM industry is saturated with different offerings for descriptive analytics. Many companies that previously buttered their bread by providing design and engineering software suites are focusing on descriptive analytics. This includes building information modeling (BIM), which has recently taken off in sophistication and value due to its partnership with other emerging, cost-effective technologies such as cloud computing, drone mapping, infrared sensing, building automation, visualization, and sensors. FM has also begun to leverage predictive analytics in the form of smart building technology and predictive maintenance systems. Smart systems control everything from air conditioning to lighting. Building sensors generate vast amounts of real-time data and monitor for failures. This combination is allowing modern buildings to become more energy efficient, less costly to operate, and more comfortable. However, many of these benefits can only be realized in modern buildings with savvy owners. As it stands now, most large facility portfolios or campuses do not benefit from built-in sensor technology, networked HVAC components, or even connection to an urban city grid.

Cloudy With A Chance Of Underfunding

When one starts looking at the vast federal real property asset holdings, the situation gets cloudier. The federal government is the largest owner of real property in the United States with over 275,000 owned or leased buildings costing $21.5 billion to operate annually. However, many of these assets are in an alarming state of deterioration and the state of federal facilities has even been listed as “high risk.” The crumbling state of public infrastructure and facilities across the country underscores the fact that public and mission-based organizations lag their for-profit counterparts in using advanced data analytics. Not only do they have to get by with shrinking budgets, they also lack the liquidity necessary to invest in new or immature technologies. Some government agencies are particularly challenged because their assets tend to be larger, older, and, oftentimes, located in rural or remote settings.

Unsensored, Need Insights

There is no “one-size-fits-all” solution at the intersection of FM and advanced analytics. A more customized approach to predictive analytics is needed, especially at organizations that lack sophistication. A recent project by Booz Allen Hamilton for a large U.S. government land management agency showed that a probabilistic approach to portfolio management can improve FM, even without sensor data. The agency was struggling with the consistency of its Asset Priority Index (API) scores, which determine funding prioritization for its 28,000 buildings. Anecdotally, the agency knew managers were introducing bias and subjectivity to the API scores, but how could this be accounted for? The team wanted to know, “What are folks thinking of when they were scoring their assets? Could we find out whose scores were more reliable?”

The agency’s data was limited to field-collected information by facility maintenance staff. However it was still extensive enough (five years of work order and inventory transaction data). The agency was steadfast in the desire to generate predictive insights from this data. So the team decided it needed a probabilistic approach.

Probabilistic Approach To Predictive Analytics

A probabilistic model represents the inherent uncertainties associated with the real world using probabilities. For example, rather than saying the daily morning shuttle flight from New York to Boston is 15 minutes late on average, a better piece of information is that this flight has a 67% chance that it will be delayed at least 17 minutes. If it’s Friday, the probability of delay might go up to 80%.

A graphical representation of such conditional relationships between variables is known as a Bayesian Network. This network is a powerful tool that can represent nearly all the different variables and the conditional probabilities associated with their dependent relationships to generate a probability distribution for a target outcome.

The advantage of using such a probabilistic approach is that it doesn’t get tripped up by blank or erroneous data, which is frequently the case with FM information collected in the field. Probabilistic techniques such as Bayesian Networks can impute missing data and continuously refine its predictions through machine learning. However, FM and organizational subject matter experts are always needed to validate and truth-test assumptions and results.

A facility and data science team collected FM data from the agency’s existing asset management system implemented a decade prior. The data was fed into a desktop application that provides a laboratory environment for creating a Bayesian Network. The application also features optimized machine learning algorithms that can discover probabilistic relationships between a large number of variables.

FM subject matter experts frequently provided validation and course correction where needed. The team was able to determine quickly the variables that were the most influential to API scores, test the reliability of scores across 40 building types, and predict unbiased scores for all 28,000 buildings.

This FM predictive analytics case study demonstrates further potential use cases, which are currently being explored:

  • identifying data entry anomalies in real property data;
  • estimating facility deterioration rates;
  • simulating the likelihood of safety and health risks/facility failures; and
  • better predicting facility operational and maintenance needs.

Little Data Matters, Too

So, the lack of high-tech equipment, sensors, or software does not mean that FM executives are out of luck. It’s not always about big data or acquiring the latest analytical software systems. Sometimes success means finding all of the little data and putting it to use. Almost always, however, analytics is about empowering your staff to make good decisions.

analyticsBangera is a lead associate with Booz Allen Hamilton, located in Washington, DC. He is an expert in infrastructure business intelligence, data analytics, portfolio planning, and facility management strategy services with over 15 years of experience.

Do you have a comment? Share your thoughts in the Comments section below or send an e-mail to the Editor at acosgrove@groupc.com.


2 COMMENTS

  1. Very well articulated Mr. Bangera. But given the cost pressures, and ever evolving Big Data technology landscape, it would behoove companies or IT teams with large budgets to think and run simple irrespective of there they reside in the sophistication graph.

Comments are closed.