Article | July 25, 2017

Utilizing Feature Recognition As A Sales And Marketing Tool For Asset Protection Teams

Source: Grupo ALTO

By Maurizio Scrofani, vice president of supply chain security and intelligence, Grupo ALTO

Retail places emphasis on loss prevention techniques which help companies recover losses. With criminals continually developing more sophisticated ways of stealing company assets — internally and externally — business has had to respond. Technology has been used to make detection techniques better and recovery easier.

But as time goes on and companies look to cut costs, which seems to be a long-term trend, the number of customer-facing staff on the shop floor drops. This is a tipping point in the way any retail operation goes about their business and the payoff is real and obvious: Fewer team members on the selling floor can reduce costs and increase profit.

What About The Impact On The Remaining Team Members?

Brick-and-mortar asset protection leaders are wrestling with this question, making decisions that will affect the future of their business. So, what is the optimal number of team members to effectively promote sales while at the same time protect the company’s valuable assets?

Ideally, you’d want as many associates as possible to generate maximum sales. Dealing with shrink is disruptive and unproductive to this end but, as usual, technology has some of answers.

For instance, facial recognition software applied to modern CCTV units is a growing asset-protection trend. Testing has been performed with this technology and the results have helped asset protection teams known when a previous shoplifter passes by the camera system so they can either escort them from the premises or keep an eye on their activity. The idea behind is similar to license plate recognition systems the police use. The systems alert the police to a car that has unpaid parking tickets or an owner wanted for jumping bail, for example. Of course, it’s easier for a camera to pick up a few digits from a license plate than it is to recognize the face of a potential shoplifter — especially if they try to disguise themselves — but the technology is at an advanced level and hits far more often than it misses.

But the question remains: Is this the best use of available technology available, or is there a better way? Leveraging facial recognition technology to highlight potential issues before they happen is fine, but a better use would be to look at feature recognition software available to solve the same problem but in a much more efficient way.

In retail, trends ebb and flow with asset protection teams in their ascendancy before the sales teams take over. If the latest trading figures shown an unacceptable level of shrink, the focus moves away from sales and there is often a tightening of procedures. The scarce resources of a retailer are at times redeployed into asset protection spend to combat this risk. That is, until the next inventory review cycle show an asset protection positive impact of the business that appears to be under control and sales flagging. Then sales become the number one focus for the company and resources move away from asset protection and back to selling.

So the Holy Grail for any retailer would be to resolve both of these issues at the same time. If resources could be used to keep both sides of the equation in balance, the company could drive profits through greater sales and better shrink numbers. The current trend with retail shows a decline of traditional brick and mortar traffic. The fact companies such as Amazon have been so successful has seen traditional brick and mortar retailers attempt to replicate their success in one way or another. Some believe contracting a store network is the future, but this does take away from the tangible consumer or customer experiences.

Once customers are no longer walking through your stores, you lose the ability to interact with them, offer a great service, or experience and promote add-on sales. The profit margins in the store are not driven by cutting staffing levels to the bare minimum. They are driven by getting new customers through the door and giving them the type of service that both gets them to come back and to buy more when in store.

So How Can Feature Recognition Software Help In This Area?

Feature recognition software can be used more ways than one to help a company deal with both issues outlined above. Once feature recognition software is bought it can be put to use within the stores of a company by the two teams that usually see their roles as a clash in some way — the sales department and the asset protection department.

Firstly, feature recognition is a bit more granular at targeting features within traditional facial recognition software when it comes to spotting potential suspects than its facial recognition counterpart. The storage of the data is key with feature as you begin to take “apart” the facial elements that are captured, left eye in one server, right eye in second server, etc. as you dismantle the data (face) but hold the aggregate elements for logic clustering when running your analytics platforms.

The “what if” here centers on re-deploying some of your asset protection expense (painful, but could be a big long-term win) and funding sales and marketing folks along with smart hardware. So, let's get into the “what if.”

What if asset protection was able to gather more information with less human capital on the sales floor, but a greater amount of physical hardware to leverage their program? CCTV systems used by store asset protection teams and cameras in the tablets used by the sales/marketing folks need to work together under one feature recognition software platform.

Asset protection teams would be able to move quickly and make better decisions on selective surveillance or detentions and therefore should have a positive correlation against those costly litigation incidents. Most CCTV equipment is usually placed not only in high traffic areas or strategic entry and egress areas, but also in key target departments. The flexibility of feature recognition software would allow one to place it in specific cameras and during peak time of the year to offset licensing expenses.

So How Would This Look Then?

Well, let’s pick a target department that is high end, trendy, and expensive. Let’s call it “ABC123” and they are soft lineapparel vendor. Within the selling floor area where vendor ABC123 is displayed, we would have the appropriate cameras with the feature recognition software.We would overlay the third party (preferably one that is most knowledgeable about vendor ABC123) sales/marketing team in department ABC123 with their tablets that have feature recognition software. As customers visit the area, there is a collective effort to gather data and our CCTV equipment is capturing and storing feature elements (left eye, right eye, right eyebrow, etc.) in independent servers.

Our sales/marketing team approaches many customers and shares with them new styles, colors, and patterns, including those not covered by this store’s assortment. As they are showing the customer how they look, they capture the body and face combination that has virtual clothing layered over it and software separates all features and houses them as noted in a prior paragraph. Our store investigators are able to maximize their visual resources to work for them and focus on targeted surveillance and quality productive detainments.

How Will It Be Received By My Partners In Legal, Risk Management, And Brand Recognition?

The focus could be centered more around mathematics and anonymity. One of the positions I have heard is we are storing a customer’s face and therefore there matter of PII matter comes up. I would remind the person taking this position we capture video today and this would be a far more granular ability to capture the data, but there are a few major differences.

Most CCTV systems store video for about 30 to 90 days in the cloud or 2 to 4 weeks on DVR. The major difference in this platform would be you are storing smaller video pieces as almost .jpg versus a traditional video. Let me share an example: Traditional video would store the event, customer walking into area, sifting through racks, selecting item, viewing item, transporting item to fitting room, different sales floor area or POS counter to try on, other part of the sales floor to continue shopping for additional items or purchase of said items.

Feature recognition’s software focus and potential use would acknowledge the customer entering the sales floor area and capture the independent features of the customer entering area. The software would decouple facial features and place them in their independent server locations (at its conclusion, creating a math problem).

Consider looking at the face with nine dominant features for discussion purposes, right and left eyes, right and left eyebrows, right and left ears, nose, and upper and lower lip. Now consider placing these nine facial features into their own server as data points: We would now have a “left eye server ONLY”, a “right eye server ONLY”, etc. when the debate in the conference room begins as to why we can’t or shouldn’t use this technology.

By storing pieces of data versus individual video for non-criminal reason (video evidence of theft), one is able to support the logic clustering of data. If asked what the average customer in department ABC123 looks like, it is a unique response. You can take all the individual data points you stored, bring back the average left eye, average right eye, etc., and re-create the face of your customer for a particular department for that window of time (date, hour, if there was a sales event, even weather conditions if you store long enough and layer weather event history). What you draw back from your multiple servers would be in gray scale as the software does not need to store color, height, weight, clothing style, etc. It is an accumulation of all of your left eyes, right eyes, right eyebrows, etc.

The argument of a store detective being selective based on gender, color, height, weight, etc. is now more agnostic and scientific. Your denominator is nine times larger and anonymous, and your numerator is crisper and clearer.

Feature recognition also takes detainment and apprehension data captures in a more anonymous fashion. Yes, you need to retain the video for the entire shoplifting event, but the support given to why you monitored this customer versus another is provided.

Well, How Would We Fund This?

What If we put 20 percent of the asset protection floor team budget in a big box retailer and funded the third party sales and marketing company who represented one of your vendors (in this case ABC123)? Let’s layer a piece of hardware such as a tablet that had virtual closet software as an application along with feature recognition that is part of the licensing agreement with your solution provider. The additional sales, impact to margin erosion, inventory accuracy and availability, and decrease in shortage will support your expense effort.

You are now delighting the customer, have more floor presence, are promoting the brand in the correct department, capturing data for the buying team based on what the customer prefers while at the same time capturing faces, features, and type of customer in the problem department at specific times.

In some places, some may call it the trifecta. We have more sales people on the floor, we are pooling customer wants and needs, and we are preventing theft. Beyond that we are capturing features and are able to understand what our virtual customer actually looks like.

Anything a retailer can do to become more of a destination will change the way potential customers use them. If a customer has gone to your location specifically because they like the service you offer and the way your team interacts with them, it will change the way you are viewed by customers.

In retail, there are transaction and experience customers, the later are looking for “something” that can’t be offered anywhere else. As technology becomes more accessible, your competitors will latch on to whatever tools they can use to help push their business forward.

The difference will be if you are able to understand the use of the technology available, make it look and feel different to what other retailers offer, and train your team to be the best in it. The old-fashioned retail values of excellent customer service are more important now than ever before. The drive of the last 20 years to open stores for more hours every day, cut prices, and push people online is being reversed.

The best retailers know how to treat their customers and how to get the most from them. By providing welcoming, secure stores that give a delightful customer experience, the basis is there for a retailer of the future. Feature recognition technology is one more tool of our asset protection toolbox that allows for more touching of the flesh, engaging with the customer while helping to keep costs down through lower shrink numbers, and providing a shopping experience customers will love, return to, and talk about.