Global e-commerce is among the fastest growing industries globally, experiencing 18% growth in 2018. Worldwide, consumers purchased $2.86 trillion worth of e-goods in 2018, compared to $2.43 trillion in 2017.
Because digital commerce is data-driven, the industry is ripe territory for AI. However, lack of knowledge and uncertainty remain the most prominent obstacles to this technology gaining a stronger foothold. To address these obstacles, deepsense.ai and Google Cloud co-organized a business breakfast to discuss the challenges and opportunities and share their remarks on artificial intelligence (AI) in e-commerce. Joining Google and deepsense.ai were experts from BeeCommerce.pl, Sotrender, and iProspect, all companies deliver sophisticated tools for digital business.
Plato’s data cave
“When it comes to building AI applications, it’s all about the data,” said Paweł Osterreicher, Director of Strategy & Business Development at deepsense.ai, during his presentation. He pointed out that the simplest analytics in smaller businesses can be done within an Excel spreadsheet or pen and paper. Preparing a simple segmentation within a client group or spotting best-performing products don’t pose a huge challenge. But those are only the tip of the iceberg. “The more sophisticated insights we gain, the more complicated the task becomes. And that’s where specialized software comes in,” he said.
“The greatest challenge is a lack of flexibility. There is no jack-of-all-trades among the popular tools, and each has its limitations. The problem is when a tool doesn’t fit a company’s needs. And, to be honest, that’s a common situation,” Osterreicher continued. Companies thus often need to tweak the tools at their disposal to make them fit or get used to missing insights from their data.
“Most companies process only a fraction of their data and operate with only half the picture. They are like the prisoners in Plato’s cave, watching only the shadows customers cast on the wall, with no access to or true grasp of their real form.”
The only way to analyze data in a convenient and cost-effective way is to leverage machine learning models. Machines are able to effectively spot patterns even in seemingly insignificant details.
“Sometimes information about how long customers hover over a button or how they go about filling in an online form is a first step to obtaining meaningful information. The model is only as good as the data it was built on,” concludes Osterreicher.
In another presentation, Jakub Skuratowicz focused on the technical aspects of how companies use AI. There are numerous ways for companies to benefit from AI, be it building engagement, personalizing the user experience or detecting fraud.
Google’s expert showed a new application of image search for omnichannel commerce. First applied by the Nordstrom clothing company, the app-enabled users to take a photograph of an item and then search for it in the shop’s database. Thus, the customer could quickly buy the product online or check its availability.
“By using Google Cloud Platform-delivered machine learning tools, the company reached 95% accuracy in recognizing an item shown in a photograph”
AI also thrives in recommendation engines. “It was common to recommend the user another version of the product – a different size of a dress, for example. That’s pointless. Why would one need another of the same dress, only slightly bigger?” Skuratowicz asked. Instead, the AI-powered model recommended products that complemented the one that had been searched for, like sunglasses or a scarf to go with the dress.
Skuratowicz also showed how AI spots fraudulent transactions in international e-commerce. “Manual or semi-automatic checking can be effective, but machine learning makes it more scalable,” he said. By applying AI-based solutions, the international logistics provider Pitney Bowes boosted the accuracy of its fraudulent transaction detection by 80% while reducing false-positives by 50%.
The mind barrier
The presentations were followed by a panel discussion on machine learning in e-commerce. As the panelists remarked, the AI-powered future of e-commerce is a challenge that not all companies are ready for.
In response to a question about the state of data-proficiency in e-commerce companies, Arkadiusz Wiśniewski, Director of Data and Technology at iProspect, had this to say:
“some data are easy to collect, while others provide a challenge, so we have an incomplete view. The legal situation in Europe poses an additional challenge, so it is better to focus on the data owned and make the best use of it.”
“Data-readiness depends to a great extent on company size. But most businesses lack the skills and data to effectively apply machine learning techniques,” agreed Jarosław Trybuchowicz, owner of beecommerce.pl.
The panelists agreed that the situation is hard even though data is becoming a commodity. “Sometimes the problem is the opposite. Despite having huge amounts of data, companies don’t get insights from it. They simply don’t know what questions to ask and what insights to look for,” added deepsense.ai’s Borys Sobiegraj.
The panelists likewise agreed that the key to success for enterprises employing machine learning is to know and properly organize their own data. Getting the data is the first challenge; “deciding what to do with it is a different story altogether,” said Jakub Nowacki, Lead Machine Learning Engineer at Sotrender. “Another challenge is extracting value that often lies in matching the data from different sources. If a company is unable to determine the impact of a sales campaign, then what is the purpose of analytics?” he added.
A question-answer session and networking time followed the discussion panel. The next business breakfast is planned for Q3.