Oracle Retail Blog helps retailers stay current on customer successes, hot retail trends and industry best practices.

Retail Science: Machine Learning Advances in Loss Prevention

Randall Fernandes
Principal Sales Consultant

A highlight of working in retail loss prevention (LP) is the evolution of technologies as companies seek the next big thing that will help reduce shrink. Artificial Intelligence (AI) and Machine Learning (ML), unlike some past buzzwords, are living up to their hype. To understand how AI and ML change the LP analytics approach, it’s important to consider the current landscape.

There are two primary AI and ML models; supervised and unsupervised. Recently, Oracle worked with a large home improvement retailer and found interesting results with both models. These results surprised both the loss prevention veterans at the retailer and Oracle’s own subject matter experts.

In the supervised model, we ran our data models on groupings of certain types of high-risk transaction groups, such as line voids;

  • The model identified numerous transactions that didn’t raise traditional Exception- Based Reporting (EBR) flags – the cashiers didn’t have the most line voids, the dollars weren’t the largest, etc. “I had doubts that these transactions were fraudulent.”
  • Once the video was reviewed, fraud was confirmed. “My jaw dropped. How did the algorithms find this theft?”
  • It was deceptively simple – what the AI algorithms found was certain items had very low instances of being line voided.
  • Typically, these items were very large (Sheetrock/lumber) or were kept in secure areas that the customer had to ask the cashier to access (power tools). Prior to being scanned, these items required a customer to take an extra step to purchase, either by placing the item on a dolly or asking a cashier to remove them from a secure location.
  • This extra effort minimizes instances of unsure buyers changing their mind at the register. Customers are less likely to bring a bulky or secure item to the register unless they are going to purchase.
  • AI identified the trend that any cashier who had a history of line voiding these infrequently voided items was likely involved in passing off the merchandise.
  • Once identified, Oracle was able to configure pre-built core reports to identify this type of fraud. 

In the unsupervised model, one where all transactions were included, it identified numerous high-risk price override transactions;

  • Again, a group of cashiers was identified who didn’t raise the traditional EBR flags
  • Once the transactions were reviewed by the retailer, they realized the suspicious price overrides were occurring at their self-checkout registers.
  • Video confirmed that certain cashiers were “hooking up” accomplices with unauthorized and fraudulent price overrides, resulting in numerous fraud cases.
  • The retailer used this information to configure their core reports and alerts to monitor for this going forward, significantly reducing their risk.

Machine Learning in Loss Prevention

Oracle has a dedicated team of data scientists who are developing the next generation of retail analytics that incorporates rules-based approaches along with the industry expertise of our retail community, all leveraging the analytical power of AI and ML. Machine learning here is not a replacement for the rules-based approaches but rather works in concert, and in fact, the rules are an integral part of training some of the machine-learning models.

This fusion of expertise and science is the base of our AI and ML approach with functionality that includes exception-based reporting dashboards. The methods surface the work for loss prevention field users by bringing data anomalies to light, allowing them to focus on what they were hired to do. They aren’t spending their time analyzing raw data but are instead addressing anomalies and using their findings to fine-tune the AI data models - while simultaneously getting the best combination of their own subject matter expertise and the power of AI.

Science with a Purpose

The latest enhancement to Oracle Retail XBRi Loss Prevention Cloud Service leverages the power of Oracle Retail Science by using machine learning to help retailers detect fraud by working in concert and enhancing rules-based techniques. The dynamic functionality and forensic analysis help retailers make more-informed decisions with timely, data-driven answers to business questions and to protect the bottom line. The cloud-based solution offers even greater precision, building on existing knowledge and adding new features and analytics automatically.

With the Oracle Retail XBRi Loss Prevention Cloud Service enhancements, retail organizations can leverage:

  • Science-based Exception Dashboards: Advance science utilizes machine learning for automated alert identification and delivery. The interactive dashboards deliver comprehensive visualization of performance, KPIs, and high-risk areas making it easy for field and executive users to make informed decisions.
  • Data-Import Wizards: Analytics through self-service allows the import of data files for custom dashboard development. The data is summarized and analyzed in XBRi dashboards and reports. Users can import data and visually expose to the team in the field.
  • Automated Exceptions based on Science: Detect anomalies on known and unknown patterns in the data for customer accounts and cashier fraud. This dashboard enables users to analyze data science results that fall outside normal risk parameters.


Learn More About Oracle’s Adaptive Intelligence solutions:



Join the discussion

Comments ( 2 )
  • Johanne Ruel Wednesday, March 27, 2019
    Hi, Thank you for this article. Question for you: is it compatible with Microsoft D365 Retails suite?
  • Rand Fernandes Tuesday, April 9, 2019
    Hi Johanne! Yes, XBRi Loss Prevention is POS agnostic and has been used with numerous software platforms”.
Please enter your name.Please provide a valid email address.Please enter a comment.CAPTCHA challenge response provided was incorrect. Please try again.