The Machine Learning Retail Revolution
Transformative solutions, powered by insights.
15+
2 3 4500+
Introduction
The origins of machine learning can be traced back to foundational theories laid out by the mathematicians Bayes, Legendre and Gauss more than 200 years ago, and the term is thought to have been coined by Arthur Samuel at IBM in the 1950s.1 The practice employs complex models and algorithms that can learn from data and make predictions and decisions based on it. By feeding computers raw information, this component of artificial intelligence (AI) attempts to achieve constant improvement in learning and performance over time. The practice allows computers to take over tasks that would normally require human intelligence, powering remarkable everyday tools like Amazon's recommendation function and always-on problem-solving, smart shopping virtual assistants like Amazon's Alexa. In today's digital-first world, machine learning -- and AI in general -- have gained tremendous momentum, reaching the Peak of Inflated Expectation in research and advisory company Gartner's hype cycle.2 According to Gartner, from 2010 to 2015, funding in the AI sector multiplied almost sevenfold,3 and over the next 10 years, AI will be the most disruptive class of technologies. Machine learning is poised to transform several industries, and savvy retailers have already begun to unleash the potential of computers and the cloud to create a better, smarter, more attuned and effective experience for customers. Precima has used machine learning since the company was founded 10 years ago, helping retailers in the areas of pricing, product assortment, promotions, personalized marketing and recommendation systems. The Machine Learning Retail Revolution provides a broad introduction to machine learning and explores how Precima applies machine learning to improve sales, provide a hyper-personal consumer experience and find efficiencies that lower costs and accelerate delivery of services.
1 Gutierrez, Daniel D., Machine Learning and Data Science: An Introduction to Statistical Learning Methods With R, 2015 2 Gartner, Top Trends in the Gartner Hype Cycle for Emerging Technologies, 2017 3 Gartner, Predicts 2017: Artificial Intelligence, 2016Precima | The Machine Learning Retail Revolution
Machine Learning at a Glance
Currently, machine learning is divided into two distinct types.
Unsupervised Machine Learning
This utilizes unlabeled data, such as user searches or interactions with ads on the web. Enabled by an always-on platform powered by the cloud, it creates a level of personalization, segmentation and customization that simply isn't possible for humans. Google's search engine targeting of models based on individual customers' searching patterns is the most visible example. Thanks to unsupervised machine learning, Google's search time (for 1.2+ billion websites) is less than half a second. The process operationalizes the model in real time to be relevant for each user -- contextually, temporally and geographically.
Supervised Machine Learning
This utilizes labeled data. The machine learning observes regular outcomes and a set of inputs to derive a response function that will predict an outcome based on that data. The most common example of this is an email spam filter. The user inputs certain parameters for what should be considered spam, which is coupled with a real-time response to the filter at work (such as times you mark an email filtered into the spam folder as not spam). With each email that arrives in your inbox, the machine learning system uses the new data to learn, adapt and improve its classification accuracy.
Many of the largest retailers are already taking advantage of one or both methods of machine learning. At Precima, machine learning has several applications, including price optimization. The company uses unsupervised machine learning to classify the customers into homogenous segments based on their interactions with the retailer and other attributes. Once customer segments are defined, supervised machine learning is used to uncover the relationship between sales and price by considering the customer's historical data while controlling for other causal factors.Precima | The Machine Learning Retail Revolution
As the entire retail industry moves to a real-time, instant model, machine learning can provide a significant advantage by leveraging the right underlying analytical models and employing robust, prescriptive analytics.
Using the response functions from machine learning, Precima is then able to recommend the right prices at the right time and the right location to maximize revenue and profit. Machine learning enables Precima to sift through millions of data points and to anticipate and examine several actionable pricing scenarios. For a typical retailer with 50,000 SKUs, three possible pricing actions (increase, decrease or no change) and product prices that affect other products' sales volume, there can be up to 350,000 pricing scenarios to evaluate. Precima's machine learning algorithms efficiently and effectively sift through all of these to not only recommend a pricing action for each product, but actual prices for each one, helping to maximize revenue and profit for the retailer. As the entire retail industry moves to a real-time, instant model, machine learning can provide a significant advantage by leveraging the right underlying analytical models and employing robust, prescriptive analytics.Precima | The Machine Learning Retail Revolution
Machine Learning in Action
We are drowning in information but starved for knowledge.
-- John Naisbitt, Futurist
Having enough data isn't a problem for retailers; the challenge is understanding how to mine that data to create better customer relationships and grow the business. Until very recently, machine learning fell in the nice-to-have category for retailers, but it is quickly advancing to a need-to-have tool for any brand that hopes to keep up with the changing demands of consumers and the evolving landscape. Today the retail industry faces ever-increasing challenges, and the next generation of personalization cannot be reached without effective machine learning. Machine learning is applied in various areas of marketing and merchandising, including product pricing, product assortment, promotions scheduling and marketing planning. Here is a look at how Precima applies machine learning and provides solutions in three key areas.Precima | The Machine Learning Retail Revolution
Retail pricing optimization is a complex undertaking, requiring data analysis at a granular level for each customer, product and transaction. Countless factors are weighed: Over time, what sales were generated from a customer at different price points If the price increases, does the customer still buy the product but in smaller quantities Is she impervious to price changes for highquality products but price-sensitive for everyday or value products All of this must be analyzed in context, factoring in temporal and causal conditions, among other things. A sales response function modeling the effect of price must control for seasonality, weather, promotions, macroeconomic effects, pricing, product relationships (jelly prices might affect peanut butter prices) and more. Precima applies both unsupervised and supervised machine learning for this process, using a complex blend of statistical and econometric approaches, combined with historical data, purchase histories, competitors' histories, product preferences and demands and more to create a nuanced approach to pricing products.Retail pricing optimization
Precima's machine learning systems analyze approximately one billion transactions on 40,000 products across 800 stores, ultimately delivering optimized product-level price-point recommendations on a monthly basis that drive positive price perceptions and keep up with strategic sales and profit goals.
In a real-world example, Precima delivers retail pricing optimization for one of the leading grocers in Europe. The company prides itself on providing the lowest prices possible, and Precima aids in that process by applying machine learning to determine the right pricing strategy for them, as well as optimized price points on products according to localized competitive situations and many other factors. To accomplish this, Precima's machine learning systems analyze approximately one billion transactions on 40,000 products across 800 stores, ultimately delivering optimized product-level price-point recommendations on a monthly basis that drive positive price perceptions and keep up with strategic sales and profit goals. These prescriptive recommendations have consistently generated an incremental lift in profit of roughly 1-2 percent while growing revenues around 1 percent.Precima | The Machine Learning Retail Revolution
Hyperpersonalization is critical for retailers to succeed in the current marketplace -- a task that can only be accomplished through learning dynamically, one customer at a time. The brands that stay ahead of the competition will (and in some cases already do) go far beyond marketing to broad customer segments or creating monthly or weekly campaigns, instead marketing to each customer individually.Personalized marketing optimization
Value Personality Life Stage
A Definitive View of Each Customer Across Three Dimensions
Machine learning is a powerful tool for determining customer targets and the right offers to make.
A robust marketing program powered by machine learning can help a retailer determine which customers to target and which offers to extend. The practice of personalized marketing optimization requires determining which offer will incent behavior, based on the shopper's history, profile and a host of other factors, while also giving the ability to test and learn to get to that perfect offer. This definitive view of each customer across three dimensions (value, personality and life stage) enables a retailer to have smarter customer objectives and measurable performance metrics, as well as to offer relevant products, incentives and customer treatments (cross-selling, upselling, etc.) that improve the overall customer experience.Precima | The Machine Learning Retail Revolution
For one of its clients in the U.S., Precima applied machine learning to drive customer personalization, delivering highly relevant communication that maximized key customer interactions, drove profitable growth and simplified sales activation. This program uncovered deep insights behind why customers buy, as well as helped clarify customer needs, purchase behaviors, life stages and potential value and drove marketing strategies to boost sales and profits. As a result, the client saw strong response rates, including 21 percent in upsell, 11 percent in cross-sell, and a roughly 3 percent lift in sales and profit. The analysis required to achieve these results was similar to the previously described process guiding retail pricing, but in this instance, the decision target was shifted from in-store product prices to individual offers to each customer, such as incentives to purchase a product or product group. This shift led to a steep increase in complexity; there are significantly more combinations of incentives and product/customer combinations than retail pricing options. To handle this challenge, Precima designed a machine learning system that could accurately predict customer response to specific marketing incentives. This system was able to optimize the allocation of marketing investments, leading to personalized incentives that drove customers into action with timely offers on the most relevant products. The machine learning system efficiently sifted through trillions of scenarios and automatically generated a set of offers that maximized the return on marketing investments -- a feat that is not humanly possible.
Machine learning systems make it possible to sift through trillions of scenarios, providing a set of offers that maximize return on investment.Precima | The Machine Learning Retail Revolution。