The focus of our next benchmarking survey will be data analytics. In our survey of loss prevention practitioners, this topic was ranked second highest after emerging technologies in terms of interest.
Prior to undertaking the survey, we thought it important to first map out what the term “data analytics” may mean to the loss prevention industry—it is certainly a phrase now widely used but is open to a wide range of interpretations and definitions. Equally, the way in which data analytics is performed can vary enormously, ranging from something as simple as an Excel spreadsheet sent to store managers to the development and use of customized systems that integrate data flows from across the entire retail business. Moreover, understanding the range and breadth of data sources that can be used as part of data analytics performed by the loss prevention function is also important.
Working in a Data Lake
In the not-too-distant past, loss prevention was often described as “living in a data desert” with inventory-driven shrinkage numbers, which were only available a few times a year, being the primary driver of most business intelligence. Fast forward to today, and instead of deserts, retailers now refer to having “data lakes,” vast quantities and types of data covering many aspects of the operation. While undoubtedly preferable, moving from a state of data famine to data feast presents its own challenges in terms of prioritization, management, and control of the data.
The term data analytics is typically used to describe the collection, interpretation, and dissemination of data in order to describe, predict, and improve the performance of a business. Analytics can be undertaken in many ways, but it is important to distinguish the difference between it and other forms of data-driven systems that provide routine alerts and responses, such as exception-based CCTV systems and EAS alarms. These types of systems routinely generate “data” upon which individuals may react, such as a security guard responding to an alarm at a store exit triggered by an active EAS tag or a member of staff approaching a customer who has triggered an alert at a smart shelf. However, the process typically lacks any form of analytical interpretation. Certainly though, the aggregation and subsequent analysis of this type of data would be data analytics—the key difference being the steps taken beyond simply responding or reacting to data-based prompts.
It is also important to distinguish the difference between data analysis and data analytics—the former is the interrogation of data sets, while the latter is viewed more broadly as the analysis, interpretation, and use of data sets to make better informed business decisions. In addition, while data analytics can be used on single data sources, it is normally associated with multiple data sources and the use of advanced statistics and predictive models. In this respect, data analytics moves beyond simply answering simple questions from data sources—how many refunds store X performed yesterday—to using the data sources to enable the business to make better and more-informed decisions—is the current refund policy being applied correctly across the business, and if not, what changes need to be made to ensure that it is?
How the process of data analytics is performed varies enormously across the retail industry. Some companies prefer to “build” their own analytics capability internally, using the expertise and resources available within their businesses, while others opt for using third-party technologies. A visit to any of the major retail loss prevention conferences and associated exhibit halls reveals a plethora of providers now offering a wide array of data-analytics packages. Others adopt a blended approach using internal resources for some analytical functionality and external systems for other elements. In addition, the location and availability of the human resource to perform the analysis and interpretation steps of data analytics varies considerably—this may reside in the IT department, it could be a specialist function within the loss prevention team, or it could be a service provided by a third party. It could also be a task carried out by a full-time specialist or be something tagged on to the duties of staff employed to perform a range of tasks.
Preliminary analysis of how LP teams are using data analytics suggests it is being used in at least three ways, based on the type of analysis being performed, the frequency with which it is done, and the type of employees using it.
Operational data analytics—Typically, automated systems based on multiple data sources and computer algorithms designed to push results to a given audience to trigger organizational responses. This can take the form of data dashboards where the intended audience is guided both in the interpretation of the results and how they might best react to them in order to make business improvements. Frequency of use: daily or weekly. Typical user: store managers or LP associates in field or store-based roles.
Tactical data analytics—The use of data streams that can be interrogated by LP teams to resolve specific issues more quickly and efficiently, for instance the investigation of specific anomalies such as above-average rates of refund frauds or out of stocks. This type of analytics is typically pulling data from business systems. Frequency of use: on an ad hoc basis. Typical user: LP investigators or regional management.
Strategic data analytics—The assessment and review of multiple data sources to enable the LP senior management team to develop medium- and long-term strategic decisions for the business. This type of analytics is typically pulling data from business systems. Frequency of use: quarterly or yearly. Typical user: senior LP executives.
One of the key elements in these three types of analytics is the extent to which the systems employed are used to push or pull data to recipients. As can be seen, a functionality increasingly offered by third-party analytics providers is the collation, interpretation, and dissemination of operational data analytics, typically to store managers, that will not only give them data specific to their situation but also potential interventions that may improve their performance. This data is largely generated automatically and is “pushed” out by the system. For some, this is increasingly being described as “prescriptive analytics”—in effect data analytics but with greater emphasis placed on helping the end user improve business performance. At the strategic data analytics level, where the data is more likely to be “pulled” from the system, some companies refer to the use of “predictive analytics”—another variation on data analytics where the various data streams are collated and analyzed with an end goal of making predictions about what future key data indicators are likely to be. For instance, stores may be offered estimates of what their levels of shrinkage, refunds, or cash losses are likely to be based upon historical data and other changes to their environment.
As mentioned earlier, the range and depth of data sources used in data analytics by LP teams can vary enormously depending on the scope and nature of the department, the data streams readily available within the company, and the type of IT hardware and software being used. Detailed below are various data streams most likely to be used in data analytics:
- Security systems and operations—CCTV images; EAS data; access control and alarm records; investigation results; records of internal and external theft; safety incidents; workers compensation; general liability.
- Store operations data—Stock audit results, stock adjustments; product damages data, spoilage data.
- Point-of-sale data—Transaction data, refund information, self-scan audit data.
- Payment data/cash control—Cash over/shorts; deposit shortages; check write-offs; gift card/SVC-losses or fraud; credit card write-off; card payment information; credit card charge backs.
- Employees—Staff turnover; performance reviews; management and employee tenure; staff engagement/morale survey results; number of full-time versus part-time staff.
- Crime data—Local police crime statistics; third-party risk models.
- Store profile—Type; size; sales volume; mall, strip, or street; layout; number of entrances; age of store.
- E-commerce data—Customers frauds; credit card charge backs; damages; errors; rate of returns.
- Supply-chain data—Unknown losses; vendor short shipping; picking accuracy; in-transit losses.
As can be seen, the potential data sources are many and varied—the data lake is increasingly fed by myriad data tributaries. Undoubtedly, these suggestions are but a snap shot of what some companies are now using, but the power of data analytics is to enable retail executives to begin to make sense of this wealth of data to improve the quality of the decision-making in the business.
Over the next few weeks, we will be reaching out to those companies that have previously contributed to the loss prevention benchmark series to gather information, based upon the framework outlined above, to understand how they are thinking about and using data analytics. In particular, we will be asking about:
- The types of data analytics they are using (operational, tactical, or strategic).
- How they are delivering data analytics in their business (internal, third-party provider, and so forth).
- What human resource they are employing.
- The types of data they are making use of when undertaking data analytics.
As with all the loss prevention benchmark surveys, the results will be made freely available to the US loss prevention community. As we design the survey instrument, based upon the discussion points above, we would very much welcome any comments and suggestions you might have to help us navigate your retail data lake and how you use data analytics. The topic is hard to define but potentially of significant interest as more and more data points become available to those working in the loss prevention function. For as the old sage Edwards Deming put it, “Without data, you’re just another person with an opinion.”