Deep Dive 21 minutesFree Report

RetailTech: Leading-Edge Loss Prevention—Tackling Theft and Fraud Through RFID, Video Surveillance and More


A major challenge for retailers globally remains retail theft and fraud, which severely impact profitability and appear to be increasing at an alarming rate. In the US, recent headlines list shoplifting-related store closures in cities including Chicago, San Francisco and New York City.

In this report, we explore the current and future technologies available to retailers for preventing and reducing shrink. We primarily focus on the US retail environment, although the technologies we explore are applicable globally.

This report is sponsored by Sensormatic, a global leader in retail systems, including radio-frequency identification (RFID), electronic article surveillance (EAS), anti-theft tags and labels, detachers, inventory management and retail traffic solutions.

Market Scale and Opportunity

Retail shrink amounted to $94.5 billion in 2021, up 4.1% from $90.8 billion in 2022, according to the September 2022 Retail Security Survey by the National Retail Federation (NRF). The shrink rate was 1.4% of sales, which has remained consistent over the past five years. Although it is facile to conflate shrink with theft—and this report will likely make the same generalization for convenience, the NRF found that the two biggest sources of shrink in 2021 were theft-related—organized retail crime (ORC) and employee/internal theft—which accounted for a combined 65.5% of all shrink, as shown in Figure 1.

Furthermore, surveyed retailers saw a 26.5% year-over-year increase in ORC incidents. Even more alarming, four in five surveyed retailers reported violence and aggression associated with ORC incidents, according to the NRF.

Figure 1. Inventory Shrink by Source, 2021


Source: US Census Bureau/NRF


However, dated industry shrink figures do not do justice to retail companies’ current concerns around shrink and do not account for the recent surge in shoplifting.

Target CEO Brian Cornell, in an interview following the retailer’s 1Q24 (first quarter of fiscal 2024) earnings release, stated that shrink would reduce its full-year profitability by $500 million in 2023. Based on reported 2022 figures, we estimate that shrink reduced gross profit by $753 million in 2022, representing a shrink rate of nearly 3%—far exceeding the NRF figure.

Apart from the social causes (which are beyond the scope of this report), the rash of shoplifting has been fueled by high felony larceny thresholds, difficulties in prosecuting offenders and an increase in ORC. With police not responding to misdemeanor larceny complaints (i.e., with total theft generally below $1,000—and thieves know this), retailers need help to file a felony complaint. Many retailers do not pursue thieves due to the perceived damage to their image and reputations from the unsightly process of detaining a thief.

Many retailers have responded by locking up merchandise, which Dollar Tree recently termed “defensive merchandising,” or by closing stores completely.

A picture containing indoor, bottle, plastic, blue Description automatically generated

Items locked up at New York City drugstore
Source: Coresight Research


These measures are likely to have an effect on consumers. A recent Coresight Research survey asked US consumers about their opinions of retail in the current shoplifting-prevalent environment (see the Methodology for more details) and found that:

  • 48% of consumers are concerned (extremely, very or moderately) that stores serving their community may be closed due to retailers experiencing high levels of theft.
  • 26% would shop elsewhere and 26% would move online if their local store put items under lock and key.
  • 75% are concerned (moderately to extremely) that retailers will raise prices to cover the cost of increasing retail theft.
  • 77% are concerned (moderately to extremely) with the ability of their local governments to enforce the law.

Retailers are highly motivated to reduce losses and alleviate their customers’ concerns about shopping in their stores.

Loss-Prevention Technologies: Coresight Research Analysis

We identify five key technologies that can drive benefits for retailers in preventing loss, as we outline in Figure 2 and discuss in detail below.

Figure 2. Loss-Prevention Technologies and Benefits

Source: Coresight Research


1. Item Tags Give Retailers the What, When and Where of Theft and Fraud

The quote, “you can’t manage what you can’t measure” attributed to management thinker Peter Drucker rings true in loss prevention, especially with regard to the sources of shrink provided in Figure 1. Shrink represents unaccounted-for inventory, which must be counted in the first place.

Item tags enable retailers to ensure inventory accuracy, which is the precondition for popular services such as ship-from-store, curbside pickup and BOPIS (buy online, pick up in-store). Inventory accuracy also enables better control of pricing and minimizes working capital stranded in inventory.

Types of Item Tags

There are several item-tagging technologies, which generally comprise a tag (some type of item identification) plus a means of detection.

  • EAS—Acousto-magnetic tags are clamped on to products in a store and set off alarms when they pass through reader antennas stationed at the store entrance. Sensormatic is a provider of these types of tags.

AM Alarming 2 Tone SuperTag Pin | Sensormatic

Acousto-magnetic tag
Source: Avery Dennison


  • Internet of Things (IoT)—These tags can include radio transmitters such as Bluetooth or 5G wireless, which transmit identity and whose location can be determined. Qualcomm has provided a 5G reference design for tags that access cellular networks. Israeli innovator Deeyook Technology offers a tracker that can determine an item’s location based on Wi-Fi or LTE (long-term evolution)/5G wireless signals.

A close-up of a white device Description automatically generated with low confidence

5G IoT Tracker
Source: Qualcomm


  • RFID—These tags comprise a chip-antenna module activated by electromagnetic energy, which then retransmits the item’s data to provide identification. A matrix of RFID antennas can triangulate on an item to show its location and movement. Providers of RFID tags, readers and antennas include Arizon Taiwan, Avery Dennison, Checkpoint Systems, HID Global, Honeywell, Impinj, Invengo, Nedap Retail, NXP Semiconductors, Sensormatic, SML Group, STMicroelectronics and Zebra Technologies.

AD-172u7 | Avery Dennison | RFID

RFID tag
Source: Sensormatic


RFID for Loss Prevention

RFID tags offer a low-cost, flexible solution for inventory management and loss prevention due to their ability to provide identity and location data. For loss-prevention efforts, RFID tags provide three key pieces of information:

  • The item’s identification and location—Each RFID tag includes an electronic data capture (EDC) identification code, which uniquely identifies an item by its SKU (stock keeping unit) and item number. With an RFID reader, retailers can know which items are in the store as well as their general location. Associates are able to find an item quickly by searching with a reader device—i.e., using the “Geiger counter” mode in which the device beeps more rapidly as they approach the item. This ability to find items is a major contributor to customer and associate satisfaction, since the consumer does not have to wait while the associate searches through piles of merchandise in the stockroom.
  • The item’s movement (crossing a boundary)—RFID antennas inside the store can show the movement of an item within its bounds, and antennas positioned above entrances and exits can identify items’ entry or exit from the store. A departure via the front door could represent a purchase or customer theft, whereas a departure via the back door could represent an employee theft. Passive RFID tags (which are lower in cost than active tags and represent the majority) typically have a range of less than 30 feet, so their usefulness largely ends when an item has left the store.
  • The item’s existence or absence—If an item resides within the store, the RFID antennas will pick it up; conversely, if an item has been initially scanned but no longer appears to be within the store, then the item can be classified as missing. The inventory tracking system will take notice if a consumer attempts to return an unsold item previously marked as missing.

RFID tags enable retailers to know the following in the event of theft and fraud: which items were taken, the time and date, and whether they left via the front or the back door. These data can be used to find patterns to identify causes of shrink and take action against shrink, such as by modifying the presentation or location of products within the store; retailers can better understand which items are stolen the most, which SKUs in which quantities and at what times of day. For example, unpaid items leaving the store before opening or after closing likely indicate a case of employee theft. Just as important, the store knows the value of the items taken, which is relevant in potentially building a felony case when the value of the items passes the relevant threshold.

Figure 3. The Application of RFID in a Retail Store

A diagram of a store Description automatically generated with low confidence

Source: Coresight Research


RFID tags can also enhance the customer experience. For example, retailer Uniqlo is using RFID tags for self-checkout, where the items to be purchased are placed in a bin (which includes an RFID reader) that automatically scans the items. When the consumer checks out, pays and leaves the store, the inventory view is updated right away.

U-Scan Mini-Express RFID - Fujitsu Frontech North America

RFID self-checkout terminal
Source: Fujitsu Frontech North America


RFID chips from Impinj that use the RAIN standard (derived from Radio-frequency IdentificatioN) offer a “protected mode,” whereby the tags can be made “invisible” to readers (for example, once the item has been purchased) and therefore do not trigger loss-prevention actions when the customer leaves the store; the tag can simply be reactivated in the case of a return. Newer Impinj chips can contain encrypted data that can only be read by a reader with chips that contain the decryption code, which enables them to ensure the privacy of the data saved in the chip.

It is not strictly true that thieves can use aluminum foil to block RFID signals, although metallic conductors do absorb electromagnetic radiation; tinfoil only blocks RFID at long distances. A Faraday cage would block all electromagnetic fields, yet such a device is likely beyond the capabilities of most shoplifters. Still, retailers need to be aware of materials including metals, water, concrete and leather and more (even paint) when designing stores and backrooms to ensure that their antennas are able to cover the retail space.

Traditional EAS technology can also identify measures to thwart detection hardware. The figure below includes a matrix of video images, with the upper right-hand image showing an alert indicating the detection of a metal-foil alarm—i.e., the customer has entered the store with a foil-lined bag designed to defeat EAS systems.

Video matrix of security cameras and alerts
Source: Coresight Research


2. Video Gives Retailers the Who, When and Where of Theft and Fraud

As a picture is worth a thousand words, video provides the essential missing piece to a loss-protection platform, providing an image of the suspected thief, along with an idea of what they are carrying, plus the date and time.

The Power of Tags x Video

The combination of “what, when and where” from item tags with “who, when and where” from video provides a clear picture of what was taken, when and by what exit, with an image of the perpetrator, and multiple incidents of the same individual can be combined to file a felony larceny with law enforcement once the relevant thresholds have been exceeded.

Other Powerful Capabilities of Video

The combination of video and the powerful capabilities of artificial intelligence (AI) enables the detection of many types of human behavior, many of which may not be apparent to the person observed. The capabilities of Sensormatic’s video retail platform include the detection/analysis of:

  • Loitering—Computer video can detect customers spending long periods in stores without making a purchase or returning to the store multiple times, which suggest that they are determining the store’s layout and points of weakness.
  • Gait—The walking speed of legitimate shoppers is more leisurely—they stroll more slowly than potential shoplifters, who tend to move more rapidly through the store.
  • Traffic—Video analytics can measure store traffic and traffic patterns, which can be used to deploy inventory in ways that discourage theft.
  • Shelf Sweep/Inventory—Computer video can detect when a suspected thief sweeps an entire shelf of product into a bag to carry out of the store, as well as monitor inventory.

Protecting Associates and Customers

Protecting customers, associates and the store represents a major component of loss prevention, and the above capabilities of AI and video enable platforms to detect and alert store employees of emerging threats and dangers, including the following:

  • Parking-lot activity—Video platforms collect data for loss-prevention and customer-satisfaction purposes, including reading license plates, identifying cars parked in a dangerous manner or for long periods of time, or recognizing the formation of mobs intent on robbing the store. In addition, video can monitor the number of cars in the parking lot or wait times to ensure optimal BOPIS service.
  • Gunshots—A gunshot produces a unique audio profile, and detection software can alert associates to notify authorities or lock down the store.
  • Threatening behavior—Video analytics can identify dangerous customer behavior such as shouting, slamming items down onto counters or wild gestures and alert associates to monitor the potential threat.

Issues Surrounding Facial Recognition

The use of facial-recognition software in loss prevention typically raises significant concern regarding privacy among US residents, although the technology is being used and accepted in other countries. Walmart had reportedly been using facial recognition to identify known shoplifters but dropped its use following privacy complaints, yet the company is the subject of a class-action lawsuit alleging that it used the facial-recognition database of Clearview AI in Illinois.

There are signs that the objections of US residents to facial recognition could be easing: the technology is finding use in other sectors such as aviation—airlines and the US Transportation Safety Administration (TSA) are now using it for identity verification at the security checkpoint and boarding gate. The technology has also been the standard for identifying consumers in several generations of Apple iPhones.

Consumers are also accepting other forms of biometric identification. For example, Panera Bread is testing the Amazon One hand-scanning application in its restaurants in St. Louis, Missouri. Consumers are also possibly more comfortable with the term “biometric identification” than “facial recognition.”

3. Self-Checkout Terminals Offer Consumers a World of Convenience and Thieves New Ways to Steal

Self-checkout terminals offer a boost in efficiency and self-reliance for many consumers, as it enables them to check out faster and take control of the process. Still, every new technology opens the door to new ways for miscreants to exploit their weaknesses, which creates a need for solutions to address them. Leading SCO vendors include Diebold Nixdorf, Fujitsu NCR and Toshiba.

Figure 4 shows selected theft and fraud risks through the basic self-checkout process. Devious minds will be able to imagine even more ways to circumvent even the best security measures.

Figure 4. Simplified Self-Checkout Process: Risks of Loss, Theft and Fraud

Source: Coresight Research


Current self-checkout terminals contain several sensors and measures to reduce theft and fraud, including scales and cameras. Figure 5 contains an annotated self-checkout terminal, the elements of which we discuss below.

Figure 5. Annotated Self-Checkout Terminal

Source: NCR/Coresight Research


Downward-facing camera. The downward-facing camera enables loss-prevention officers to catch theft and fraud that happens outside the front-facing camera’s field of view, such as transferring items directly into the customer’s bag. It also leverages AI to catch mis-scanning, product switching, leaving items in the basket and leaving items in the cart. The module contains an indicator light to signify the need for customer assistance, or a potential theft/fraud situation.

Item scanner/front-facing camera. This assembly includes two sensors: the item scanner on the bottom and the one facing the customer, which includes a camera that can identify products and labels, as well as monitor the customer’s activity. These images are used for pick-list assistance (for example, specifying which type of apple scanned) and product assurance. Although the video feed is not necessarily monitored by a loss-prevention associate in real time, it can be recorded, and the message “you are on video” serves to deter some would-be thieves.

Scale in bagging area. The scale performs a very important function, comparing the incremental weight of an item to the weight that is averaged over time in a database of products the retailer sells. The divergence of an item’s actual weight compared to a baseline (within a certain tolerance) could represent a case of label-switching. The weakness of this method is that this average can be corrupted through misuse or error to abuse the system.

Cameras in self-checkout systems can also leverage AI for purposes such as age verification for the purchase of alcohol and tobacco.

4. AI-Based Analytics Use Automation and Computing Power To Identify Theft, Fraud and Waste

Not all effective loss-prevention technology requires item tags or computer-video data. There is a great deal of valuable information embedded in everyday inventory and POS (point-of-sale) data that can be extracted using automation and the ability of AI to identify hidden relationships within data.

We summarize the capabilities of AI as the following:

  • Harnessing enormous amounts of computing power
  • Finding (often hidden) relationships among data
  • Determining the best model or algorithm to make predictions
  • Unlocking the benefits of automation

Prescriptive Analytics Continuously Scans Store Data for Theft and Fraud, Waste and Noncompliance

Analytics platforms such as Zebra Prescriptive Analytics monitor data to determine a baseline, normal level for certain data items, which are used to detect anomalies. Many of these forms of theft and fraud can be detected in the checkout process—the difference here is that this is a software-powered solution. Examples of these anomalies include:

  • High shrink—Unusually high shrink levels could indicate issues with shipment, theft or fraud.
  • Sweethearting—Refers to employees giving items to friends and family for free or at a reduced price (a “sweetheart” deal)
  • Sliding—When an associate slides a product over a scanner while obscuring the barcode so that it is not registered
  • Illegal voids, markdowns or discounts—Scanning items and quickly voiding the transaction or price tag switching conducted by a store employee
  • Cash refunds—Issuing improper refunds from the store’s own inventory in cash
  • Training mode—Switching the POS to “training” mode, in which transactions are not linked to the cash in the drawer
  • Return fraud—Customers can attempt to receive refunds for items taken off the store shelf, or buy a product legitimately and try to return a cheaper product, making a profit in the price differential.

In each of the above examples, a baseline is established where data from the attempted fraudulent transaction is outside of normal bounds and generates an alert for further inspection.

Figure 6. Overview of Types of Theft and Fraud That AI Can Detect

Source: Coresight Research


Prescriptive analytics also helps retailers identify hidden revenue opportunities, sources of waste and noncompliance, and inventory imbalances, such as the availability of goods in the stockroom but not on the store shelf, which is a highly sensitive area of customer satisfaction.

Other Uses of AI Today

AI is in practice today in several other applications, including the following:

  • Online fraud detection—Companies such as Signifyd and Riskified use AI to identify fraudulent online orders and, more importantly, verify legitimate customers and orders in a speedy manner so as not to generate unnecessary friction.
  • Cybersecurity—Cybercriminals and ransomware installers relentlessly seek access to complex corporate networks and are employing AI-based tools to invade networks. Many of these hackers gain access with standard network-administration tools, which would not appear as malware, and network operators possess an ever-shrinking window within which to interpret the faint signals these attacks generate to launch a response. Since many attackers are employing automated, AI-based tools, retailers would be well served to “fight fire with fire” and employ similar technologies in their defense, as described in our report offering key points from the 2023 RSA cybersecurity conference.

5. Generative AI Could Offer New, Innovative Ways To Catch the Bad Guys

While generative AI technology has been years in the making, the technology made a splash in November 2022, with OpenAI’s launch of ChatGPT (based on GPT-3.5), and GPT-4 is already available. The early capabilities of the platform include its ability to answer questions in natural language and summarize and generate text. Since ChatGPT’s launch, several global technology vendors have announced the addition of GPT functionality to their platforms, including Google, Microsoft and Salesforce.

The combination of natural-language processing (NLP) and the ability to find connections among large amounts of data means that generative AI could be a powerful, flexible tool for retailers, with these strengths enabling it to replace many dedicated software platforms. Moreover, this flexibility could enable it to create solutions in short order—a major benefit for retailers—since new forms of theft and fraud can emerge quickly and be spread quickly by social media, and software platforms need time to be written, debugged and tested.

At this early stage, we can speculate on likely applications of generative AI, including the following:

  • Finding patterns in theft and fraud data, such as which items were taken and when
  • Predicting theft and fraud times
  • Predicting new items targeted by thieves
  • Finding new patterns in POS and self-checkout data
  • Determining optimal store layouts to reduce theft
  • Analyzing and finding patterns in ORC behavior

Again, with AI chatbots, loss-prevention employees can ask questions in plain language, and the AI finds patterns, which can be queried and refined, which provides a more interactive experience than fixed reports and dashboards. At the time of this report’s writing, no loss-prevention vendors have announced products leveraging generative AI; however, the advanced ones are likely experimenting with it, and it is only a matter of time before new products and capabilities are announced.

What We Think

The criminal mind is constantly looking for new opportunities to receive ill-gotten gains, and unfortunately, retailers need to remain vigilant and implement new means of thwarting them. Fortunately, there are several technologies that can help retailers reduce the risk of shrink or at least identify theft and fraud where it is not possible to prevent or stop it. The combination of the watchful eye of video cameras plus the ever-increasing capabilities of AI offers new, powerful ways to identify behavior that suggests the intention to shoplift, enabling retailers to spook or shame would-be thieves to lose their nerve. AI’s ability to use computing power to find hidden relationships in data enables retailers to identify theft and fraud not visible to the human eye. Finally, the power of generative AI unlocks a great deal of analytical power and flexibility, which can be ready to identify and prevent new and clever means of theft and fraud.

Implications for Brands/Retailers

  • RFID tags are a mature yet shunned technology that offer retailers a wealth of data that can identify which items were taken, at which time and via which exit, and these data are valuable for building a loss-prevention case as well as maintaining inventory accuracy.
  • Retailers and brands should remain informed as to the latest loss-prevention capabilities of self-checkout due to the avenues for theft and fraud they create.
  • AI-fueled prescriptive analytics analyzes POS and inventory data to identify theft and fraud, in addition to hidden revenue and other opportunities to increase efficiency and compliance.
  • AI-powered computer video enables retailers to analyze consumer behavior to identify potential threats, safety issues and offers other capabilities such as inventory tracking.
  • Generative AI promises to unlock the power of AI to find relationships in data and store information in new, previously unimagined ways.

Implications for Real Estate Firms

  • Retailers and landlords will need to reconsider retail spaces and entry points, reconfiguring them to make them more resistant to shoplifting and robberies such as “smash and grabs.” Landlords also need to fortify walls between stores to prevent break-ins from less well-defended neighboring stores.

Implications for Technology Vendors

  • Vendors can combine loss prevention, inventory visibility and store analytics in one broad-based, store operations platform.
  • AI-based video analytics is a powerful tool for analyzing customer behavior in the store and could potentially identify shoplifters before they steal and discourage (or shame) them into backing off, or alert store employees.
  • The power and capabilities of AI—especially generative AI—continues to increase and offers opportunities for developers of software platforms and applications.


Informing the data in this report is an online survey of 401 US consumers aged 18+, conducted by Coresight Research on April 17, 2023. The results have a margin of error of +/- 5%, with a 95% confidence interval.