Transforming the Shopping Experience Through Machine Learning

Machine Learning


Whether Father’s Day, Mother’s Day, a birthday or simply “just because,” buying gifts can feel like counting grains of sand - i.e., it’s not easy. Meanwhile, there are 250,ooo—300,000 e-commerce companies in the U.S. all vying for the attention of shoppers. Nevermind all of the brick and mortar shops across America. How are consumers possibly expected to decide where to spend their hard-earned money? And how can retailers create a more personalized shopping experience, rather than facing the same demise as the 6,700 retail locations that closed their doors in 2017?
In short, the retail sector is in a sticky wicket.

A case for machine learning:

Given the high-stakes nature of retail today and the challenges converting online visitors to paying customers, one solution that holds tremendous potential to upend the shopping experience is machine learning. Imagine, for example, that you want to purchase flowers for someone. You want to send something thoughtful, but perhaps you’re also worried about sending the wrong message. You could go to your local florist and explain this—at which point they’ll walk you through a wide array of their offerings. But do you really have time for that? Wouldn’t it be great if there was a way to order the perfect gift for the occasion—but online? thought so too—and in 2016 launched Gifts When You Need, affectionately known as GWYN. Think of GWYN as an AI-powered concierge that can ask you a series of questions about the person for whom you’re purchasing a gift, and then offer more personalized recommendations based on the purchases of other customers with similar profiles. The best part is, with every new purchase, GWYN is able to learn more about customers wants and needs, and then improve its recommendations for consumers. According to CEO Chris McCann, within two months of launching, approximately 70% of online purchases were made through GWYN.
This is just one of the infinite ways in which machine-learning algorithms can parse through a sea of data. Each time a visitor comes to your site, another data point is created. Now imagine the millions of data points that can inform your model. Not only does this help you better understand your customers, but it helps you deliver the right product at the right time—at scale.

Getting started: First-party data vs pre-trained models

The concept of machine learning is simple: by using statistical techniques, programs have the ability to "learn" with data, without being explicitly programmed. When we say “learn,” this means that as programs receive more inputs (i.e., data), they provide more accurate outputs—such as gift ideas.
Given the power of machine learning to completely transform the shopping experience as we know it, one might think that it’s only for large companies with big budgets. Not so. No matter your company’s size or budget, machine learning is something that every retailer can begin applying to their business.
Let’s assume for a moment that your business is well established, and you’re sitting on a veritable treasure trove of historical data. Because machine-learning results improve the more data you have, you’re perfectly situated to begin feeding this proprietary—or first-party—data into machine-learning training models. Whether financial transactions or customer call transcripts, these data can be used to train and optimize your models so you’re armed with unique insights about how to better deliver value to your customers.

Conversely, if your company doesn’t have access to vast amounts of data, you can still become a machine-learning pro using pre-trained machine-learning models that are instantly available in the cloud. For example, Google has Cloud AutoML, which is a collection of machine-learning products that make it easier for developers with limited experience in machine learning to get started.
You might begin with image analysis model, which helps you understand the content within images. Supply images and get text categories (e.g., “automobile,” “Eiffel Tower”). This model can also identify objects, faces, or printed words within images.
For example, CI&T created an automated invoice reader for one client, which helped expedite invoice processing—something we can all agree is annoying but necessary. Since some invoices contain logos rather than the company’s name spelled out, we were able to use the Cloud Vision API logo detection tool to train the invoice reader’s model to identify company names and increase the accuracy of processing invoices. It’s always amazing to see how spending less time on these types of repetitive tasks liberates companies to focus on higher-value activities.
Depending on your needs, you may be able to leverage the text analysis model to better understand the structure and semantic meaning of text. Further, it helps you extrapolate information about people, places, and events from any text such as news articles, blogs, or chat logs. The speech recognition model converts audio to text.
And a powerful video analysis model can identify objects within videos, and tell you what’s happening within an individual shot or frame.

The tip of the iceberg

Retailers looking to leverage machine-learning solutions may focus their energy on supply-chain planning to improve forecasting and orders. We’ve already talked about personalizing the shopping experience—but what about optimizing your pricing strategy based on seasons, supplies available, or any number of relevant factors?
All of these examples are just the tip of the iceberg, and don’t begin to fully capture the potential machine learning has to revolutionize the retail sector.