Most innovations introduce small and welcome improvements to our lives: zippers are more ergonomic than buttons or ties, microwaves are more convenient than ovens, ball-point pens leak less than fountain pens. But some innovations have a seismic impact on society and the economy: fire, the printing press, the internal combustion engine, the transistor. Although only time will tell, many thinkers believe artificial intelligence will join the second group, radically altering the way we live, work, and interact with each other.
Artificial intelligence is transformative because it can reduce costs and improve efficiency in almost any industry. Few areas cannot be enhanced — or put at risk, depending on your point of view — by machines that do our thinking for us. Machines are faster, more accurate, and they don’t tire.
What’s the role of AI in retail?
Retail, and online retail in particular, has embraced artificial intelligence. Retailers are working towards an eCommerce experience shaped by intelligent algorithms which select the products customers see, decide how they are presented, and determine their price. When a customer buys a product, AI-powered machines will pick it, pack it, and figure out optimal delivery routes. If the customer needs support, their first point of contact will be a chatbot that uses speech recognition and deep learning technologies to answer questions.
Customer service and support
According to Gartner, by 2020, a quarter of customer service and support operations will integrate virtual customer assistants (VCA) — industry jargon for chatbots. The most sophisticated chatbots use AI-technology to recognize what customers say and assess their intent, crafting responses that are, if not human-like, at least useful.
Customer support is expensive. Aspect Software, a multinational call center technology provider, estimates that the average customer phone interaction costs $35 – $50. Text chat is much cheaper, and if AI chatbots can field most support conversations, the cost is even lower. According to Juniper Research, chatbots will save around $8 billion in support costs by 2022.
Related read: What is a Chatbot and How to Use It for Your Business
Writing product descriptions
It’s not just support staff that will be replaced by machine learning algorithms. Earlier this year, OpenAI announced that it had developed a language model called GPT-2 that can “generate conditional synthetic text samples of unprecedented quality.” Feed the model a chunk of text, and it generates similar text. There is nothing new about this in principle, but GPT-2 is scarily good, so good that OpenAI refused to release the fully trained model in case it was used generate false news articles indistinguishable from the real thing. In the near future, similar technology will be used to write product descriptions on large eCommerce websites, displacing copywriters.
GPT-2 analyses millions of lines of content and builds a model so that when it is given a new chunk of text, it can generate the next word, and the word after that, and so on. This is a generalizable capability of AI systems. A machine learning algorithm looks at lots of things and builds a neural network that maps the connections between them. When you give the trained model a similar thing, the neural net spits out what comes next, whether it’s the next word, the next action, or the next rainstorm. Neural networks can predict the future.
Personalizing shopping experiences
Recommendation engines have, until recently, been quite primitive. We’ve all chuckled at the absurd suggestions that eCommerce stores and social media networks sometimes make. It is challenging to predict what a shopper wants based on what they have already bought, but this is another version of the “what next” problem. With enough data, it’s possible to build machine-learning models that, when given a product history, can more-or-less accurately predict and display what the next product will be, what combination of products is most likely to elicit a sale, or even what discount should be applied to maximize both the likelihood of a sale and the retailer’s profit margin.
Nick Rishøj Danmand, a data scientist, wrote last year about how recurrent neural networks can be used to predict a shopper’s next action. He used session data from some of the largest eCommerce retailers in Europe, training a neural network model to predict the “the next customer move”. The results are impressive.
“We found that recurrent neural networks outperform the baseline model by 41.3% (RSC) to 161.9% (AVM), increasing accuracies from 50.65% and 20.18% to 71.55% and 52.85%, respectively.”
As Danmand points out, the ability to predict shopper behavior will allow retailers to craft personalized experiences that increase customer loyalty and revenue.
Warehouse and delivery automation
For many of us, the term artificial intelligence brings to mind robots, and anyone who follows this space has seen videos of warehouses filled with robots whizzing between shelves. Picking and packing are labor intensive, and many of the world’s largest eCommerce retailers are searching for automation solutions that use artificial intelligence technologies like computer vision and machine learning.
JD.com, a massive Chinese eCommerce retailer, has recently invested $4.5 billion to build an AI center in China and a robotics research hub in Silicon Valley. Founder and CEO Richard Liu would like to remove humans from the retail process altogether. His vision of retail has machines doing everything from picking and packing in a people-free warehouse to delivering products to customers’ doors — like Amazon, JD.com is working on delivery drones.
In spite of significant investment, picking and packing today involves people working alongside robots. The most sophisticated systems in widespread use are goods-to-persons systems. Robots bring bins or shelves of products to human packers because only humans have the dexterity to handle and pack fragile products.
Current research is focused on removing this limitation. Ocado, an online-only grocery retailer, has demonstrated a robotic arm that can safely pick and pack delicate items such as fruit. It is working on a computer vision system that can “look at” a collection of products and figure out the best way to handle and pack them. This March, Daniella Rus of MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) demonstrated a robot gripper designed to pick up fragile items without crushing them.
Richard Yiu’s dream of a human-free eCommerce process will take some years to achieve, but with the current pace of machine learning and robotics research, I wouldn’t bet against a machine-dominated retail industry becoming a reality soon.
This is a guest post by Graeme Caldwell. Graeme is a writer and content marketer at Nexcess, a global provider of hosting services, who has a knack for making tech-heavy topics interesting and engaging to all readers. His articles have been featured on top publications across the net, TechCrunch to TemplateMonster. For more content, visit the Nexcess blog and give them a follow at @nexcess.