AI in retail - big investments can deliver new business and happy customers

The nuts-and-bolts of AI technology are perhaps the best current example of systems engineers and computer scientists pushing the state of the art in computer and data science, system design and semiconductor manufacturing. It’s an area I frequently cover because understanding these elements of AI and data science provides an early look at capabilities such advancements enable as they are incorporated in cloud cognitive services, machine and deep learning applications and predictive data analytics.

As edifying as such details can be, it is understandably hard for busy business and IT executives to get very interested in the internals of the latest AI accelerator, deep learning model or AI development framework when they are running an organization and focused on business metrics like revenue, customer retention and new product delivery. Such indifference to the technology makes things particularly dicey when an IT manager asks an executive to sign a six- or seven-figure check for an AI ‘supercomputer’ or approve monthly 4-figure AWS bills for a single cloud-based forecasting application. Instead of fixating on the technology, business leaders are fixated on business results and evaluated by ROI.

When it comes to AI investments, it pays to think creatively about things you would like to do if unconstrained by technology, since, given the pace of AI technology development, what once seemed impossible can quickly become eminently feasible. Often, the best way to describe what’s achievable in AI is to see how the AI trailblazers already use it. To that end, this week offers an excellent opportunity for those in retailing via the NRF conference, where leaders in the retail applications of AI are touting their achievements without giving away too many competitive secrets.

Four areas where AI improves retail operations and results

There are four business functions where AI can yield significant benefits to retailers:

1. Inventory management and stock availability.
2. Online and in-store marketing including personalized recommendations, localized promotions and store design.
3. Better security from behavioral tracking and transaction fraud detection.
4. Streamlined operations, logistics and supply chain.

Several of these are also dependent on other new technologies such as wireless beacons and tags for precise location sensing and tracking, video surveillance systems and intelligent devices, i.e. IoT, like building control or point of sale systems.

A survey of executives at firms developing AI-enhanced marketing products found that retail trade and e-commerce are the most commonly targeted industries, listed by 42% of respondents, with traditional offline (bricks and mortar) retailers, food service and recreation also landing among ten most important markets for these vendors.

The survey found AI used to improve many different business objectives, with the most significant being:

1. Revenue generation
2. Customer acquisition and retention
3. Competitive differentiation
4. Cost reduction

While AI can improve each of these business functions and objectives independently, the best applications cut across categories to hit several targets at once.

Domino's brings Intelligence to pizza marketing, ordering and delivery

Domino’s Pizza isn’t just one of the best performing companies of the past decade, but a frequent target of commentary here at diginomica. Indeed, the two facts have a common cause: Domino’s is one of the most innovative early adopters of technology in any industry, as we detailed more than three years ago that serves as an exemplary model for other companies. While the use of delivery robots and robotic kitchen helpers grab the headlines, Domino’s is also exploiting AI and data analytics to reduce and more accurately predict delivery time, tweak the menu, customize promotions and optimize its supply chain.

Domino’s made a bold bet on deep learning at last year’s Super Bowl with a “Points for Pie” promotion that rewarded loyalty points when customers submitted a picture of any pizza, from Domino’s or not. Although rewarding someone for buying a competitor’s product might seem foolhardy, the project achieved two significant objectives.

• It provided a platform for Domino’s data science team to hone its deep learning skills and computational hardware by developing an image classification model to identify pizzas and label the various toppings.
• It allowed Domino’s to gather an enormous amount of data about the most popular toppings, including the shapes, colors and texture of various ingredients, the popularity of various types of crust and the regional preferences for different types of pizza.

Developing deep learning expertise enables Domino’s to apply more advanced classification and predictive models to other applications such as delivery estimates and menu optimization. The massive data trove from customer photos provides the raw materials for more accurate predictions and business decisions. For example, Domino’s data science expertise has enabled it to improve the accuracy of predicted delivery times from 75 to 95% by incorporating variables like the number of managers and employees on duty, the number and complexity of queued orders and real-time traffic conditions. Domino’s is understandably mum on new AI applications under development, however, a manager in its data science and AI department says that his group’s experience with image classification and computer vision could be useful further streamlining the ordering process.

From self-checkout to no-checkout

Amazon popularized the concept of no-checkout retail via its Amazon Go stores which use a combination of video surveillance, proximity and weight sensors and software to eliminate the checkout stand. While Amazon was first to market with stores where the shopping experience resembles shoplifting, other companies, many of whom will be at NRF, have been perfecting the various sensor, data analysis and AI technologies to enable such grab-and-go convenience to be deployed by other retailers in many different settings. Some notable examples are:

• AiFi, Grabango, Trigo Vision and Zippin are each developing cashierless systems that mimic the Go Store experience. Indeed, AiFI takes it a step further by offering a complete NanoStore, some no larger than an airport kiosk, that can be dropped almost anywhere.
• AnyVision Insight software uses video surveillance to analyze customer in-store behavior and identify heavy shoppers, estimated gaze or dwell time and build heat maps that can improve store layout and optimize product placement for promotions and revenue generation.
• Malong Technologies also uses in-store video data to minimize shoplifting and theft at self-checkout stations. Its product recognition software also streamlines self-checkout by automatically identifying fresh produce and eliminating the frustrating process of entering item codes.

My take

The use of AI and data analytics to streamline or automate retail business processes are in their infancy and the cited examples are but a small sample of some innovative applications that are intended to whet the imaginations of business executives. “Imagination” is the operative word, since the available technology enables new ways of doing things that would have been unthinkable a decade ago. As IBM concludes in a report on AI in retail and consumer products, it pays to think big by evaluating functions and processes that span existing vertical business silos; for example, analyzing the entire supply chain, not just particular parts like retail distribution. Don’t try to optimize compartmentalized systems that only siloed because of organizational boundaries or tradition, internal politics or legacy software systems.

Beyond creativity, exploiting AI for retail requires building the skills, technical platforms and organizational culture necessary to successfully create and implement new applications. As the IBM report details, a survey of 1,900 consumer products and retail executives cited skills and resources as the factor most essential to AI success.

Likewise, new systems create new risks and challenges. As the IBM survey outlines, chief among these are the possibility of automation-induced mistakes or vulnerabilities, difficulty integrating new and old systems and organizational malaise and inertia. Furthermore, since AI and predictive software relies on data, the accuracy and reliability of new AI-based applications and systems are critically dependent on the integrity, fidelity and timeliness of their data sources. Thus, organizations must prepare the way to an AI-enabled future by first instrumenting and validating all data sources and sensor inputs feeding AI software.

The 2010s laid the technical groundwork for AI via new algorithms, acceleration hardware, distributed databases and cloud services. Due to their technical sophistication and specialization, these were necessarily limited to early adopters and visionary AI pioneers. This decade will see AI technology blossom into a garden of new applications, services and process improvements that will cover every industry and where retailing will be notable for its disruptive innovation.