Is ChatGPT e-commerce’s new weapon?

In retail, generative AIs like ChatGPT could boost business revenues by $400 billion to $660 billion a year, according to a McKinsey study¹. In fact, 45% of business leaders revealed in a Gartner survey that ChatGPT has accelerated their AI investments². First presented in November 2022 by Open AI, ChatGPT generative AI quickly became the topic of the moment. But to what extent could this "generative pre-trained transformer" contribute to the development of e-commerce?

Just how good an asset is ChatGPT for e-commerce?

Above all, ChatGPT could prove useful in developing web products, services and platforms in the field of e-commerce. 

A tool for research and analysis

ChatGPT’s analytical capabilities can be used to understand customer data better, in order to improve their experience. Indian market research company Insights Opinion³ noted in a recent newsletter that the generative AI’s natural language processing capabilities brought to light data that could have been lost in a more general analysis, and thus results that were more detailed in terms of specifying customer preferences. In the sphere of e-commerce, ChatGPT can be used to survey customers and analyse the behaviour of users on a platform.  

‍

"We could, for example, ask a user to describe their consumption habits in a few lines," adds César Tonnoir, CEO of Miam. "If the user writes: "There are four of us at home, including two children, we never spend more than €100 a week on shopping, and we try to limit our meat consumption while favoring local produce", ChatGPT will be able to automatically apply the right tags to the user's consumption profile: nb_adults = 2, nb_children = 2, panier_max = 100, score_viande = 0.3, score_local = 0.7... And all this, without asking the user to fill in an off-putting and time-consuming questionnaire.Inaddition to interacting naturally with the customer, ChatGPT's ability to "read between the lines" is proving invaluable in fine-tuning an e-commerce experience".

Writing product sheets and item lists  

Generative AI is a powerful tool for developing e-commerce platforms. Certain companies rely on ChatGPT to perform tasks that are difficult or repetitive for a human being. For example, lingerie brand Undiz, a subsidiary of the Etam group, uses it to draft its product data sheets and catalogues⁴.

Developing e-commerce platforms

ChatGPT can also assist companies with the technical development of their e-commerce platform.⁵. This conversational agent can generate lines of code for developers, recommend technical tools and resources, explain how specific software works, help troubleshoot technical problems and more. 

Creating content

One of ChatGPT’s main features is obviously its capacity to generate textual content. For several months now, some companies have been using chatbots such as this one to create marketing emails and newsletters. One of these is American software company Salesforce, as noted in an article published online by US business news channel CNBC in March 2023. Salesforce has in fact integrated Einstein GPT – based on the same technology as ChatGPT – using it to draft personalised e-shots for sales⁶. Chatbots can also be used to create advertising visuals for businesses like supermarket chain Carrefour, which chose ChatGPT to make its promotional videos⁷. In them, an avatar designed by the retailer explains the advantages of purchasing from the French food retail giant. Finally, ChatGPT can, of course, also create blog articles and pages for companies⁸as well as producing relevant marketing content.  

"AI can’t operate without us making sure of its authenticity: that it’s not hallucinating, that it’s not harmful," Nico Beukes.

What are the limitations of ChatGPT? 

Although it is seen as a considerable asset by some, there are nevertheless several limits to the chatbot’s capabilities.  

‍

One of these is the risk of misinformation, highlighted in an article inLe Monde, early this year by Laurence Devillers, a professor at the Sorbonne specialising in AI and its ethical implications⁹. Generative AI can indeed give genuine responses to users’ inquiries, but can just as easily produce something that is a complete fabrication. These “hallucinations”, sometimes also called “confabulation”, may even be very credible and can easily mislead an internet user. Furthermore, if its error is not pointed out, ChatGPT can build on this new, manufactured data when responding to queries from other users, who in turn may find the responses credible. This then creates a vicious circle of misinformation within ChatGPT’s own system.

‍

"In the case of a food e-commerce experience, the retailer who recommends products or meals to a consumer bears a particular responsibility. We can imagine a case where ChatGPT generates an inedible or dangerous recipe (without taking allergens or food balance into account, for example)", explains César Tonnoir.

‍

An article by US business magazine Forbes notes that the chatbot has not yet been optimised to be able to give the best response to human inquiries. In particular, ChatGPT has trouble with understanding the context of a query, from the point of view of both the way it is framed and also its very nature¹⁰. Furthermore, generative AI struggles to respond to multiple queries, as it does not naturally know how to order them by priority. This leads to a decrease in its efficiency and in the relevance of its responses. ChatGPT therefore proves to be less effective in the case of a compound query on the same topic, that is, a question made up of several smaller ones. Finally, generative AI experiences difficulties when it has to deal with certain, particular subjects, because its knowledge is relatively restricted. So, when faced with questions that are too specific or that are not covered extensively on the net, ChatGPT is not an effective tool for users.

What form do these limitations take in the field of e-commerce?

ChatGPT can suffer from several issues in the area of e-commerce when it has to deal directly with customers.

Customer support that leaves room for improvement 

As a chatbot that offers many benefits, ChatGPT can play an active role in the support given to consumers on e-commerce sites. 

‍

In France, the Carrefour chain immediately jumped on the innovation bandwagon, integrating ChatGPT into its e-commerce site. Under the name Hopla, the chatbot advises and accompanies customers on their Drive shopping journey.¹¹. Unfortunately, ChatGPT's current difficulties are also evident in this use case. Indeed, LSA 's retail experts have tested this functionality and detailed "[that] there is still progress to be made on query understanding and recommendation accuracy"¹². However, Hopla has only been present on Carrefour's website since June 2023, and promises to improve its predictive skills as it interacts with the company's customers.

‍

Southeast Asian e-commerce platform Lazada, which offers mass-market consumer products, has also launched its own chatbot, called LazzieChat. Powered by ChatGPT technology, it is hyped as being the ideal assistant for customers purchasing clothing and beauty products, and yet it suffers from a few problems. Lazada itself revealed as much to Dutch business analysis firm The Paypers¹³. Nevertheless, the Asian marketplace is insisting that it is continuously improving the service and hopes to make it more and more effective.  

"Generative AI is by definition dependent on the 'prompt'. Nevertheless, it is and will be a powerful and extremely promising tool for our business, with a wide range of uses", César Tonnoir.

A risk of misinformation in after-sales care 

As mentioned, ChatGPT does not always provide users with reliable answers to queries. This has become an area of concern, often noted by experts and journalists in the field.  

‍

Recently, the client communication platform Intercom integrated ChatGPT into its new chatbot system, Fin. According to Fergal Reid, senior director of machine learning at the firm, this new functionality is indeed an effective tool, but one which needs fine-tuning¹⁴. "We’re … continuing to pay attention to Fin’s ability to communicate in a human-like way and adhere to the guardrails we put in place to avoid misleading information and hallucinations."  

‍

Since May last year, marketing and research firm Yext also uses ChatGPT as a component of its customer service chatbot. However, its managing director, Nico Beukes, insists on the need for a level of human control behind these technologies: "AI can’t operate without us making sure of its authenticity: that it’s not hallucinating, that it’s not harmful". In the context of post-sales customer support for e-businesses, ChatGPT therefore needs to be constantly monitored so that it does not propose non-existent solutions to customers or misinform them about product or service specifications.

The risk of filter bubbles

One of ChatGPT’s drawbacks is the quality of the data it has access to, since it is trained to sift through vast quantities of text available on the internet. As Forbes notes, the chatbot’s data “may contain biases or prejudices”, which in turn means “the AI may sometimes generate responses that are unintentionally biased or discriminatory”¹⁰. These in turn sometimes contribute to the risk of being caught in a filter bubble. This phenomenon, described by Eli Pariser in around 2010 when he coined the term, is a state an internet user finds themselves in where they are shown only content that they like or approve of, or which corresponds to data defined by a specific algorithm (generative AI, recommendation engine, etc.)¹⁵. An article by the freedom of expression watchdog Index on Censorship noted that "this can … limit exposure to diverse perspectives, and impede critical thinking¹⁶".  

‍

This is why on social networks, algorithms now recommend a certain amount of random content, in order to confront internet users with content they either don’t know or are not likely to approve of. For the moment, however, ChatGPT does not have this function. This generative AI therefore controls the information it shows users, limits the type of content they have access to and curtails the exchange of ideas. In the context of e-commerce, the risk of shrinking the scope of possibilities and choices offered customers is therefore higher. Indeed, ChatGPT’s built-in limitations on how it chooses content to show the user, combined with any additional search parameters a company using it could train it to – such as not suggesting a competitor’s products, for example – could limit the products or services it recommends and so, de facto, not genuinely give users freedom of choice over the purchases they are making.  

Conclusion

"Likemachine learning, generative AI is not magic. There are certain use cases where its relevance is still limited, for example to inspire a consumer when he doesn't yet know what he's looking for. In other words, generative AI is by definition dependent on the "prompt". Nevertheless, it is, and will continue to be, a powerful and extremely promising tool for our business, with multiple use cases on which we are already working", concludes César Tonnoir.

‍