This is based on chapter 19 of the second edition of “The Facility Manager’s Guide to Information Technology”. The book is a collection of, maybe we can call them essays, by about 30 different authors on various FM IT topics. I collaborated with a gentleman named Simon Davis to write chapter 19 on data-driven facility management.
This book was put together by IFMA’s Technology Community. Buying a copy of the book will help support the Technology Community’s mission to bring facility management best practices to the FM industry.
You can purchase the book at Amazon.com. If you buy the book after following my link here, I will get a small referral fee from Amazon for sending you over (at no additional cost to you):
I was at a conference (this is unrelated to the book) and the speaker started talking about using data to make facility management decisions. Someone near me asked, kind of rhetorically to the people near them something like “how else would you make decisions?” This is a good question. One answer is that you follow the smoke to the fire and make whatever decision most expediently puts the fire out. For a facility of any significant complexity, there will be plenty of fires to keep everyone busy without any analysis of any kind of data. Of course fire-fighting mode has Its limitations. Most notably, a lack of predictability in budgets or labor requirements or space requirements or what things coming down the line might have a negative impact on business operations. Also when we’re always in firefighting mode, we have no idea if what we are doing is efficient or in any way optimal. It’s also stressful and not that much fun.
We’ll never be able to completely retire the firefighter’s hat. Some things are not predictable, or maybe just not worth planning for. For all of the other things, data can help us do better.
From the book:
“Data-driven facility management is the use of data and data analysis as the basis for decision making in the operation of facilities. From a business perspective, this approach enables users of the information to move from a tactical focus on resolving issues to a strategic perspective on how operations can be enhanced and improved based on both historical analysis and predictive analytics…”
So, to enable effective decision making we need data, and we need to analyze the data.
Some examples of things that can benefit from data-driven decision making
-
Energy management.
-
MRO (maintenance, repair, operations) and capital budget forecasting.
-
Optimization of running building systems, fast detection and diagnostics (we can see potential problems before they are real problems and find the root cause quickly), automated commissioning (automatically and continuously verifying that our building systems are operating to their design specifications).
-
Space management and optimization (having the right amount of space and using it the right way).
-
Labor force management and optimization (having the right number of employees with the right skills)
-
Creation of benchmarks, standards, and other references by external vendors and industry organizations.
Metrics, benchmarks, and KPI’s (key performance indicator)
-
A metric is something that is measured (or maybe it’s the measured value). We measure things when we care about how they might affect our business. There’s the oft-quoted adage by Peter Drucker: “what gets measured gets managed” I believe is that quote. There’s also the common paraphrasing: “you can’t manage what you can’t measure.” Drucker might have said or written that too. There are 2 bits of wisdom in there. The first is maybe the most obvious: if you want to manage something or improve something or whatever, you have to measure it. The other, maybe less obvious bit is that if you’re measuring something, someone’s probably going to be expending resources to optimize or improve that thing. So you better make sure you are measuring the right things (see KPI’s). I’m getting sidetracked.
-
A benchmark is something to measure against. It is a standard or a reference value that serves as basis for comparison or for evaluation of a metric. The term “benchmark” is usually used when we are comparing our performance to the performance of others. The benchmark could be the performance of your competitor or an industry standard or average. You might also use the term “benchmark” when comparing your performance to your own organization’s past performance or to describe a target for future goals. Industry organizations such as IFMA and EnergyStar do the research necessary to aggregate and publish facility management related benchmarks.
-
KPI is short for key performance indicator. The term KPI is used to describe a metric when that metric is tied to an important business process. An example of a KPI for a property management company might be occupancy rate. This would measure what percentage of managed space is leased. Too low of an occupancy rate would indicate the property manager is at risk of losing clients. An occupancy rate that is similar to the average for the market would indicate that the property manager is running a healthy business while an occupancy rate that is much higher than the market average would indicate the PM is a high performer. Important KPIs for a maintenance organization might be wrench time percentage or maintenance backlog. Either of these things being far out of range indicates problems in the maintenance organization that need to be solved. FCI is important for facility management organizations.
Data collection and storage
I said before that in order to do data-driven facility management, we need data. We probably want to let the computers do some of our data analysis. For this, we need our data to be sufficiently structured, in a machine readable format, and economically feasible to acquire
-
Sufficiently structured means each data point has to have some meta-data so the computer can put it into the context of the rest of the data. We want to represent all points of the same type in the same way so they can be automatically analyzed.
-
We have to get the data to the machines. Data that is generated by the machines is likely already in a machine-readable format. Other sources of data may need to be converted after they are created. In many cases, where computerized analysis is valuable, it might even make sense to modify the processes that generate the data so that the original form of the data is machine-readable.
-
Economically feasible to acquire – if, in order to get useful analysis, we have to have somebody read the value from a thermometer and hand enter it into the computer every five minutes, it’s probably not going to happen. Labor is expensive and this type of process is prone to error. If instead we can install a sensor that the computer can read, the data entry is almost free after the initial setup (making it economically feasible vs the manual data entry). The same can be said about data from paper work orders, etc that has to be hand-entered by maintenance technicians. If the process is too laborious, it isn’t likely to happen. A process where the computerized version of the data is the primary version and paper can be automatically generated if necessary is more likely to be economically feasible.
Sources of data. In the chapter we listed the following sources of data:
-
CMMS – These contain data about maintenance activities, their associated costs, and also past history and future demand.. These systems also might contain the information necessary to determine the condition of the facility (some might even calculate a usable FCI). CMMS additionally may contain asset lifecycle data to aid in capital and maintenance budget planning.
-
CAFM systems contain data about space utilization, moves (churn rate), facility usage and occupants.
-
Building control and monitoring systems can provide current and historical data about the operational status of building systems.
-
BIMs contain descriptive data about the building and associated assets.
-
Utility bills.
-
Core business metrics, benchmarks and other data from business processes and projections.
Data storage and normalization
So we can get all of this data from these different sources If we want to analyze and compare the data, it might be necessary to convert it to some common format that is understood by the analytics system. Hopefully we can do this automatically. There are various ongoing efforts to develop and maintain standard formats for FM data. I won’t go into them here.
I’ll read a little bit more from the book here:
“When leveraged effectively, the data generated and used in the management of facilities has potentially enormous business value. As a consequence, it is important to have a deliberate data management policy that properly identifies each data store or source and makes the appropriate provisions to secure the data from unauthorized access and backup the data to protect against failure of the storage device or service provider”
Data analysis, analytics
Ok we have the data, now let’s talk about the analysis piece:
-
Simplest case, analytics can be a human browsing a data set for a pattern or a deviation from a pattern. Our brains are really good at seeing certain types of patterns if the data is formatted appropriately. An example of this might be that the FM might notice, while reviewing a stack of completed work orders, that one particular maintenance worker spends 50% more time than other workers on a common maintenance task. This analysis might prompt the FM to investigate the reason. Another example might be visual analysis of a line graph of multiple years of energy or water consumption. A significant deviation would stand out and prompt investigate. Unfortunately, relative to computers, humans can only analyze a small amount of data and might only be good at seeing certain patterns.
-
At the other end of the spectrum we have something like machine learning. Machine learning is an artificial intelligence process that involves training algorithms with data to allow those algorithms to make predictions or decisions based on the patterns they find in the data. This is similar to the way a human might observe data and see whichever patterns happen to emerge. The main difference is that the computer can analyze much more data at a much faster rate which allows it to notice patterns that a human might never notice, and in fact machine learning works best with very large data sets. We can also contrast machine learning with a more traditional analytics approach where the model of the system is created by the human programmer. The computer would then analyze incoming data with respect to that predefined model in order to make predictions or decisions. Machine learning and the predefined model approach (there’s probably a name for this but I can’t think of it right now) have their strengths and weaknesses, so one is not necessarily a replacement for the other.
We can think about the extent of “data-drivenness” in our decision making by thinking about what kind of information our data analysis gives us. One model we described in the book has 4 basic categories: descriptive, diagnostic, predictive, prescriptive. We can think about how each of these might relate to the monitoring of a pump motor:
-
Descriptive analytics tell us what happened. The motor failed at 25,000 of the expected 40,000 run hours. A person then has to figure out why it happened and how to prevent it in the future.
-
Diagnostic analytics tell us why it happened. The motor failed early because because it overheated as a result of running at double the design duty cycle. The system may be able to determine that the excessive motor run was due to a higher-than-design demand on the infrastructure.
-
Predictive analytics tell us what will happen. The motor will fail early (after so many hours) due to running at excessive duty cycle.This knowledge can enable proactive correction of the problem to avoid early failure or planning to prevent unnecessary down time.
-
Prescriptive analytics tell us what we should do. The analytics may suggest a change in operating parameters, if possible, in order to bring the motor duty cycle into spec or suggest the addition of an additional pump.
In this chapter we also detail several different categories of FM analytics
-
Historical analytics deal with what has happened in the past. These are things like
-
work order completion against defined targets
-
customer satisfaction
-
planned vs reactive maintenance ratio
-
cost of service delivery per area for things like capital expenditure, MRO, energy, or cleaning.
-
Churn rates
-
Energy consumption
-
Each of these enable decision making about how to proceed going forward.
-
-
Operational analytics deal with what is happening right now. There are things we might want to pay attention to in order to optimize our operations or prevent problems before they happen. Some examples of things that fit into this category.
-
Utilization of office space, or the one I mentioned earlier rate of occupancy for leased space. Conference or meeting room utilization rate is similarly useful.
-
Work in progress or outstanding, open work orders, maintenance backlog, etc.
-
Equipment performance metrics based on real-time conditions such as energy consumption, temperature, sound levels, vibration, fluid levels, etc.
-
-
Predictive analytics deal with what is likely to happen in the future given knowledge of the past and present. We generally get better predictions with more data and better algorithms. Some examples of things that fit into this category:
-
Anticipated life of equipment. I walked about this a little with the motor example.
-
Optimal space utilization rates. How much space will we need in the future? Will we need more or less?
-
Labor demand based on anticipated planned, reactive and predictive maintenance activities.\
-
Future capital and operational budget requirements based on the current facility condition, asset lifecycle data and historical maintenance and operational expenditures.
-