"In God We Trust, All Others Bring Data" had adorned the walls of NASA headquarters for decades. Then, one day on a cold morning in Florida in January 1986, the Challenger space-flight disaster changed everything. Seventy-three seconds into the flight, the spacecraft had disintegrated, killing all seven astronauts. What went wrong? Richard Feynman, physicist, and Nobel laureate, who was commissioned to investigate the disaster, concluded that in the absence of data, the scientists had failed to argue by reason.
This story teaches us an important lesson in applying AI and data science to the world of retail that is led by consumer decisions and choices, which are not always rational. In such a scenario, pairing data-driven decisions with reason in the form of context and experience could make all the difference. Traditional brick-and-mortar retail has always relied heavily on merchant (or "buyer") instincts and experiences to run the business. On the contrary, new-age digital commerce have turned to data and experimentation to make business decisions. Omnichannel commerce, a mixture of brick-and-mortar and digital, has been the messy middle. This is where Lowe's, a Fortune® 50 company, has positioned to reinvent itself and stay relevant for the future.
What Does it Take to Be an Omnichannel Retailer of the Future?
Retail constitutes three foundational pillars – assortment, experience, and value; i.e., what products you need to sell, how would you sell them, and at what price.
At Lowe's, we classify "experience" as any customer touchpoint wherein we try to improve the overall customer journey using the applications of AI and data science. In the current pandemic, more and more customers are starting their shopping journey online, seeking touchless yet convenient fulfillment options.
This is where we have been leveraging initiatives such as Collection products to market and sell an entire experience rather than a single product to customers to complete the look in their outdoor living area. Also, we analyze buying patterns to understand the home-improvement projects customers are undertaking while at home, such as kitchen remodeling, and recommending products for the next stage of the project.
Offerings such as Store Curbside pickup or BOPIS have helped serve such customers while they seamlessly transition from one shopping channel to another, still getting the same experience across Lowe's. Further, just as online has been a virtual extension to stores with endless aisles in the past decade, stores have become an extension to online, offering the much needed experiential buying. Lowe's has tested robots for helping in-store customers find products just as the recommendation engines have served for online.
Stocking seasonally relevant products that customers are likely to buy in their geographic area is the Holy Grail for retail merchandising. With 1700+ Lowe's stores in the U.S. and 40,000+ SKUs in every store (with another few hundred thousand available through our in-store Special Order Sales or on Lowes.com), this problem gets exacerbated and needs scale to advance. This is where we pair a state-of-the-art merchandise planning strategy and processes with machine-learning models to get the right location-assortment mix. We build these intelligent models to create the "customer choice matrix," taking into account hundreds of parameters that range from consumer decision traits such as price affinity, brand propensity and online penetration index to geo-specific factors such as local competition, supplier ecosystem, and seasonal patterns, among others, to increase the likelihood that customers will find what they need. We further leverage optimization algorithms to ensure we play to category strategies to drive the right sales, margin, and traffic.
To drive customer value and positively influence NPS (Net Promoter Score), we price products competitively for "trip driver" categories or seek price-leadership position for "basket builder" categories, using machine-learning models. These models allow us to extract publicly available data on the web, process automated workflows to execute on the category strategy, evaluate price-change performance, and self-correct.
In the same vein, when promoting products, we attempt to understand the true intent behind promotions and forecast key performance indicators using machine-learning models. This allows us to make conscious choices on which promotions to run that would positively influence customer buying experiences and improve our bottom line — as well as how often to run the promotions, at what depth of discount and across which SKUs or categories.
AI in retail, thus far, has mostly solved for boxed problems that are within constraints. The next phase of AI that we are attempting to work toward is in the area of artificial neural networks.
The retail process can simply be summed up as: Define category strategy -> assort products -> forecast sales -> position inventory -> choose the right price points -> place in the right store locations or fulfillment centers -> promote as needed. Building an interconnected deep-learning model for such a process would be ideal, wherein we would have clear visibility and command over controllable factors that impact business performance.
And, it can simulate data for unpredictable events such as Covid-19 or trade tariffs to train and improve such deep-learning models with the eventual goal that the models "self-learn" and react in the most optimal fashion — e.g., shortage in product availability -> increase in cost -> assorting complementary items -> reordering and replenishment, etc., with the intent to optimize the business-objective function without hurting the customer experience. Eventually, an interconnected deep-learning model would be able to run such an end-to-end process for any retail organization most optimally.