AI models hold significant benefits, especially when GPT collaborates with a search engine on an online shopping platform, it is likely to improve the shopping experience for customers. The Harvard Business Review states, “For companies that manage to leverage GenAI to achieve their business objectives effectively, the benefits are likely to be substantial and rewarding.”
Supporting this view, McKinsey predicts the possible economic effect of generative AI could be between $2.6 to $4.4 trillion worldwide. This growth is anticipated to be fueled by applications like efficient content creation, optimised use of data, improved product finding, AI-enhanced search features, and more efficient personalised shopping experiences.
Large Language Models (and a specific use case, generative AI chatbots) are currently the hottest trend in the tech industry, with numerous ecommerce leaders hopping on the generative AI bandwagon aiming for website enhancement. By leveraging vectors, semantic comprehension, and customization capabilities, language models spearhead a conversational AI transformation happening globally. Although there are some bugs to be fixed for practical success in the real world, businesses of every scale are diving in to find out if they can achieve significant success with this technology.
So what benefits does AI offer ecommerce?
The main hurdle for every online shopping business is consistency: boosting sales by expanding the range of products and reducing expenses. Achieving success with artificial intelligence in online shopping demands a smooth blend with the shopping experience (UX). Therefore, are big language models suitable and potentially excellent, unsuitable and a risk to profits, or a bit of both?
The brief response: LLMs excel at some tasks but fail at others, and their capabilities could make them valuable contributors to online sales. Just like with other technological advancements, numerous business leaders are exploring how to use generative AI. However, if you’re among some entrepreneurs, discussing generative AI and online shopping might seem confusing. After all, online shoppers have certain must-haves, such as the ability to view various buying options with 100% accurate, contextually relevant details.
In the realm of search, LLMs are not yet at that level. To date, they have a reputation for doing some unexpected things: there’s that hard-to-ignore risk that they might start hallucinating. They’re also likely to share any data they’ve been trained on, which could lead to the leakage of confidential information if the company has provided access. For a website focused on providing accurate information about shopping choices, even a minor error could cause significant issues.
Large Language Models that work well in ecommerce
As usual, we can turn to Amazon for the newest trials with artificial intelligence tools. The company has introduced a chatbot, Rufus, equipped with an LLM trained on the product catalogue and customer feedback, to respond to user questions at different points in the shopping process. So far, feedback on Rufus has been quite negative.
However, this doesn’t imply that LLMs are completely ineffective for online shopping. The issue with many existing generative AI tools is that businesses are using them to substitute current, functional technology rather than just enhancing the online shopping experience. In the realm of ecommerce, in addition to established solutions like vector database technology, they can be quite beneficial.
They are gaining traction in the following areas, where consumers are finding them more user-friendly:
Customer support
“When can we expect this item to be back in stock? If not, what would you suggest as an alternative?” Customers frequently face questions about purchasing before they hit the Buy button. As a seller on the internet, it’s beneficial to provide top-notch, tailored responses instantly. However, what if your customer service team is short-staffed or not available?
If a language model trained with natural language processing (NLP) methods is equipped with the necessary details for answering questions, it can fulfil the needs of customers efficiently. Take, for instance, a customer who has previously reached out to Support. When they return to the website and pose some additional questions to the chatbot, it can utilise the previous conversation’s context to understand the situation better and respond logically.
Content creation
Similar to addressing various extensive marketing requirements, LLMs can efficiently generate text for product descriptions, blog articles, email marketing, and social media posts. They can even tackle all these tasks simultaneously, tailored to the platform and the desired tone. An area particularly suited for LLMs to automate is the creation of compelling purchase confirmation emails that also feature upsell suggestions.
GenAI can manage this task effortlessly when combined with a specialized recommendations model for selecting complementary products. Generative AI can also be leveraged to produce product descriptions in advance, which can then be shown alongside the products. An ecommerce platform can expand on this concept by utilising customer data to generate product descriptions and personalised marketing content in real-time. This approach would significantly enhance the shopping experience for potential customers, making the website appear more like a personal shopping assistant rather than a mere collection of products.
Hyper-personalisation relationships with each consumer
As more businesses have embraced personalization throughout the customer journey, from designing products to reaching out to customers, and from the shopping experience to adjusting prices on the fly, it has set a certain standard for personalised customer interactions.
The era of mass media, where broad advertisements to all potential customers could engage a broad spectrum of clients, is over. Nowadays, technology enables every interaction to be unique and personal, and customers anticipate a personal connection with the brands they deal with. A study by the University of Texas reveals that the desire for personalisation stems from the need to manage and simplify the decision-making process.
When products and interactions are personalised, it creates an environment where customers are at the heart of all business decisions, offering them more control over their interactions. This, in turn, affects customers’ decision-making as the information provided is customized to their individual needs, making it more relevant to their requirements.
This customised information also simplifies the decision-making process for customers when choosing products and brands. The mix of convenience, understanding of customers, and emotional connection leads to customer loyalty and increased sales for companies. Customers who are emotionally connected and loyal spend twice as much as those who are not engaged, and 80% of them will recommend the brand to their friends and family.
See our study on the distinction between the top 30 hyper-personalisation vendors
Normal search isn’t going anywhere
Rumour has it that when handled carefully and in conjunction with effective methods, LLM technology has the potential to enhance the shopping experience for online customers and increase profits. Yet, if this technology is mismanaged, it could lead to significant harm to the shopping experience and the reputation of a brand. Numerous businesses are attempting to completely replace conventional search methods with LLMs. This approach is misguided because LLMs:
Can’t use structured data consistently
Artificial Intelligence (AI) cannot generate the kind of detailed, precise search outcomes it’s not equipped to deliver from a restricted set of data, like a list of products. A large language model, on the other hand, operates uniquely; it’s a large-scale neural network trained to “predict the next word”. When presented with a prompt, it analyses the surrounding context and makes an educated guess about the most probable next word, which means it struggles to include structured data, like specifics from a product catalogue, in its replies.
For instance, a record for a pair of jeans might include details on what other clothing items complement them and how each style suits them. Given this kind of context, an LLM might arrange this information into a paragraph, but only for that specific item. Should someone on the homepage inquire about jeans that fit a certain style, the chatbot wouldn’t be able to sift through the entire product database to suggest a specific SKU. Conversely, a search engine could allow the user to narrow down their search by criteria such as category, size, and style.
Don’t possess search features
Ecommerce specialists advise that for consumers looking for products, a traditional search experience is generally the most effective. Why? It’s because individuals need to browse and narrow down their options, which isn’t easily achieved through the interactive nature of a chatbot. While LLMs have their advantages, they fall short of meeting the expectations of search functionality.
The one-answer model falls short when compared to the comprehensive search capabilities of a well-optimised business search engine. “The absence of context and the abundance of choices are unavoidable in a purely conversational setting,” comments Dustin Coates, a search expert at Algolia. Through search, users receive a prioritised list of results that includes product information and options for filtering, ensuring there are no omissions or errors. Therefore, while a chatbot powered by LLMs can provide basic information about a product, it falls short in presenting purchasing options from an online store’s product catalogue.
Can produce inaccurate answers
Large language models fabricate content. “AI delusion is a situation where a large language model (LLM), typically a chatbot or a tool for computer vision, sees patterns or objects that don’t exist or are not visible to humans, producing outputs that are nonsensical or completely wrong,” IBM describes. They are attempting to guess the next word in a sentence, but they can’t confirm if that guess is accurate. This situation can lead to trouble if your entire purchasing journey relies on AI.
If a chatbot consistently provides wrong information, how can a business expect its customers to remain loyal? Online shoppers have shown to be cautious when dealing with minor annoyances like slow website loading times, so with something of this scale, they might just leave. Chatbot replies can also differ greatly depending on how questions are phrased. When a question is asked in a specific manner, the LLM might not understand the context or provide the correct response, but with a slight adjustment — essentially training it to think differently — it might come closer to the correct answer.
However, searches are supposed to make the shopping experience easier, and customers shouldn’t have to wait for a chatbot to find the right answer. Depending on an LLM — even a moderately good one — to assist customers with simple tasks is like asking a child to shop at a physical store and tell you what you need. Even if the child manages to get the right item (and remembers to bring back your credit card), the whole process is absurd.
