AI in eCommerce: Challenges & Solutions. How to Use Machine Learning in eCommerce?
Jun 9, 2020 • 5 min read
AI: challenges & solutions. Use cases in e-commerce.
Fifty years ago, people used to fear artificial intelligence (AI), but now it is one of the primary driving forces behind technological progress in nearly all walks of life. There are endless possibilities for machine intelligence and ways to program it to solve a myriad of infinite tasks, be it self-driving cars, machine translation, face recognition, or fraud prevention. However, AI comes with a set of challenges. First, the cognitive capabilities of today’s architectures are very limited. Second, bringing AI into existing businesses is a difficult (but not impossible) task that requires a deeper understanding of the technology. In this article, we will discuss the challenges and opportunities that AI brings to enterprises and then will cover AI applications in e-commerce.
What is AI?
Artificial intelligence is intelligence demonstrated by machines, as opposed to natural intelligence demonstrated by humans and animals. Modern machine capabilities include understanding human speech, competing in a game system, autonomous driving, intelligent routing in content delivery networks, military simulations and so on. The last five to six years have been pivotal for the development of artificial intelligence, partly connected with the emergence of affordable neural networks, the emergence of cloud computing infrastructure and the increase in the number of research tools and datasets. During these years, we have seen that Facebook has developed a system to describe images to blind people, China has accelerated government funding of AI research, that Microsoft is succeeding in automatically translating Skype conversations and that Amazon is thriving on its product recommendations, among other exciting things.
How big is the AI industry: statistics
According to Tractica research, the AI software market shows rapid growth. In 2018 its volume was $10.1 billion, in 2020 it will reach 22.6 billion U.S. dollars. By 2025 it will reach $126.0 billion, showing at least five fold growth. AI Software Market Size
- 2018 — $10.1 billion
- 2020 — $22.6 billion
- 2025 — $126.0 billion
- China — 26.1%
- USA — 14.5%
- UAE — 13.6%
- Financial sector - 41%
- Production - 16%
- Wholesale and retail - 14%
- State Sector - 6%
- Other markets - 23%
Challenges and solutions
According to O’Reilly’s free research ebook on AI adoption in the enterprise, there are a few bottlenecks that hinder AI integration, such as the corporate culture that does not recognize the need for AI, a lack of data or skilled people and difficulties in identifying appropriate use cases. From the above, it is obvious that the major hurdles to integration revolve around people, data, or business alignment. Let’s dive deeper and address some of those common concerns.
Business alignment: digital transformation
There is a common misconception that associates artificial intelligence with software engineering. The association is inherently incorrect. And while AI certainly draws upon computer science, information engineering and mathematics, it is also based on statistics, economics, psychology, linguistics, philosophy, among others. The AI subfields, which are based on technical considerations and tools, include machine learning and artificial neural networks. In the core of the field lies data science rather than software engineering. The lack of being able to differentiate between the two can be off-putting or even lead to choice that does not yield the maximum benefit to the business. All challenges, however, come with new opportunities and solutions. To help your team change the way they think about AI and its adoption, you need to educate them: start with yourself and once you have some knowledge, it will be easier for you to manage the expectations of others and help people learn more about AI. Because AI is inherently different from software engineering, it means that AI adoption requires a set of different competencies, management skills and qualified teams. This can be solved with a clear strategic approach with measurable objectives, KPIs and ROI.
Risks of implementation
After determining your business needs and goals, you need to consider other variables such as data storage, data infrastructure, labeling, ways of feeding the data into the systems. Then, there is research, model training, testing, data sampling, creating a feedback loop and assessing the performance. Then even after successful integration, you still have to train people to use the model and interpret the results. Not surprisingly, all the above requirements make business owners wary of investing in AI in order not to fail. Those issues can be addressed by finding a reliable AI vendor or hiring an experienced data science team. By working closely with the experts and having a clear-cut strategic approach to AI integration and implementation, as well as having reasonable expectations, the risk of failure is mitigated.
Lack of data vs lack of skilled people
While large enterprises struggle to find field specialists, small businesses struggle with the inadequacy or scarcity of available data. Even huge corporations like Facebook, Apple, or Microsoft compete for top talent, not to mention other businesses, where management usually lacks the technical know-how necessary to assess the expertise of the hired personnel. This can be addressed either by outsourcing a data team with a solid portfolio and relevant experience, by outstaffing, or by hiring professional recruiters specializing in top tech talent.
Lack of data
It is a common knowledge that the built system is only as good as the data provided. AI systems require massive training datasets. The biggest question for small businesses is where to get that data. When faced with that problem, you need to understand what data you already have and what else the model requires. The missing part might be publicly available or you may have to buy it from a third party. However, even then, some data might still be hard to obtain or unavailable. If that is the case, there is still something you can do – use synthetic data. Synthetic data is created artificially based on real data. It is especially useful when there is not enough real data to train the model. Other solutions include using RPA robots to scrape publicly available data or a Google dataset search. There is a seemingly growing trend among larger companies that offer exchange opportunities. For example, Sberbank, a Russian state-owned banking and financial services company, organized a playground (the so-called Sandbox) with its own available anonymous data for startups and data scientists to use free of charge, in exchange – Sberbank takes care of projects and keeps track of top talent.
AI in e-commerce
What particularly interests us about AI is its practical application in e-commerce, where it has long ceased to be a proof-of-concept to be deployed in ways that have already seen immense real-world implications. From product recommendations and chatbots to personalized services – the possibilities are endless. Let’s take a look at some of the most popular applications of AI in e-commerce as of today.
Development of cognitive abilities
One of the primary concerns of the AI application is to provide more interactive solutions to enrich and improve the user experience. The popular examples are virtual assistants such as Alexa, Google Assistant, Alice (Russian Alisa developed by Yandex), Cortana or Siri. The virtual assistant can process and interpret the user’s native language and come up with an intelligent answer. Another popular and prominent AI application are chatbots, which are the only reason some businesses can operate on a 24/7 basis. The bots tend to offer the customers all possible solutions, answer most common queries and sometimes even help consumers make a purchase decision. The bots typically tend to communicate with the user via text and voice messages (or a phone call). In addition to providing customer service, both voice and text-based chatbots enhance the impact of AI on e-commerce through the following capabilities:
- Natural Language Processing (NLP) that processes voice and text-based interactions
- Deeper insights that address consumer needs
- Self-learning capabilities that improve over time
- Personalization that provides a more unique and personalized experience to each individual consumer
Personalized product recommendations for online shoppers are one of the major driving forces for increasing conversion rates. By using big data, AI influences consumer choices through its knowledge of their shopping and search history, as well as online browsing habits. One of the most famous use cases of a successful implementation of product recommendation is most obviously Amazon. Based on specific data gathered for each individual customer, AI derives key insights that create a personalized shopping experience, creating a kind of shopping window tailored specifically for a targeted individual. An interesting use case of such a personalized experience is Discover Weekly from Spotify. In 2015, Spotify introduced a customized playlist of tracks that a given user will most likely want to listen to. It is carried out with the help of an algorithm that determines a user’s taste based on their listening behavior and most popular playlists among the entire Spotify audience. This is how it works: a user listens to tracks and saves songs allowing the engine to develop the user’s taste profile; meanwhile billions of other users create their own playlists, which the engine keeps track of and identifies those that might suit the user’s taste profile but which they have not yet listened to. Moreover, if the user fast-forwards a song within the first 30 seconds, such action is interpreted negatively by the engine, allowing it to learn and adjust in the future. The technology behind that algorithm was developed by Echo Nest, a music intelligence company, which was later acquired by Spotify.
The efficient management of internal and external business processes can now also be facilitated with the use of AI. For example, one of those applications concerns inventory management, in particular, maintaining the right level of inventory to meet the market demand. AI analyzes the sales trends of recent years, projects the anticipated changes in product demand and takes account of potential supply related issues. Enriching data using AI can bring certain benefits: from obtaining clean and consistent data to yielding deeper catalog insights. Imagine a large online retailer like Taobao with hundreds of millions of products, that are impossible to track or tag manually. Automated product tagging tailored by AI increases business operation efficiency and ensures a better shopping experience.
Another AI-based productivity tool is an intelligent image search or even image processing. For example, when a user searches for a particular item of clothing, say, an empire waist dress, which she cannot fully describe but if she sees it, she knows it. By analyzing the images the user looks through, AI can identify exactly what she might be looking for. The technology behind image searches is based on neuroscience. Deep Neural Networks enable AI to see images the way humans do. One of the most famous examples of image processing at works is Pinterest. Social networks utilize image processing AI to prevent abuse, distribution of pornography and so forth. Real estate marketplaces use image processing algorithms to sort millions of images of properties and identify the best pictures that could potentially be sold. For example, instead of placing a bathroom image as the main image for that property, AI places a living room picture.
Machine translation and website localization are yet other instruments in the AI toolbox. For example, eBay applies automatic machine translation depending on a user’s location by translating a user’s request and responding with relevant inventory from other countries.
Another way online retailers can benefit from AI is by utilizing dynamic pricing. Instead of looking at your competitor’s pricing or other external factors that may influence market prices to determine the best price point, you can now apply a combination of machine learning algorithms to predict prices and establish optimal prices for your products. This way, you have the ability to leverage prices based on data from both internal and external sources. For example, when competitor’s stocks are running low, you can increase prices and vice versa. According to Business Insider, Amazon changes its product prices 2.5 million times a day, which makes an average product cost change every ten minutes.
Service chat bots & sales assistants
Last but not least, there are customer service tools, among which are obviously chatbots mentioned previously and some other instruments that improve the customer experience consistently and continually while reducing the workload and overheads associated with managing customer service personnel. One of the use cases of utilizing AI to enhance the consumer shopping experience is that of Zara online store, where in order to minimize customer returns, an algorithm suggests the right clothing size based on a user’s measurements along with style preferences.
According to IDC, industries are expected to continue aggressively investing in AI software and its capabilities with an estimated $79.2 billion in 2022 with a compound annual growth rate of 38% over the period of 2018-2022. It could not be a better time to jump on the AI bandwagon regardless of its challenges, provided you apply a clear-cut strategic approach for its integration into your business – this way, you will capture the technological benefits of AI, which willimprove your customers’ experience and boost your business performance.