In the world of online commerce, search is a key element in attracting and converting potential customers. Today, we’re going to tell you about a technology that is revolutionizing the way companies interact with their customers: Search Commerce with GPT (Generative Pretrained Transformer).

What is Search Commerce with GPT?

Search Commerce is a strategy that uses online search to attract and convert customers. GPT, developed by OpenAI, is a language model that uses machine learning to understand and generate text. By combining these two concepts, we obtain a technology that can understand naturally formulated customer queries, generate relevant answers and help companies acquire new leads.

The benefits of Search Commerce with GPT

Our experience with GPT Search Commerce has revealed several key benefits. Firstly, the technology enables a better understanding of customer queries, resulting in more relevant search results and a better user experience.

Indeed, customers can express their needs in natural language, meaning they finally have the opportunity to express themselves as they would with a personal shopper: describing a situation, a need, rather than looking for a solution they’ve already had to think about for themselves.

What’s more, thanks to its ability to generate text, GPT can respond to the customer in natural language, while at the same time being able to describe products in a more attractive and personalized way than what is stored in the database, which can increase conversion rates.

What is prompting?

In the context of large language models (LLMs), “prompting” refers to the method of interacting with these models by providing an initial input (or “prompt”) to guide their text generation.

For example, if you give an LLM the prompt “Who wrote ‘War and Peace’?”, the model would generate a response like, “War and Peace was written by Leo Tolstoy”. The prompt serves as a direction or cue, guiding the model on what output the user is aiming to get.

In essence, the effectiveness of prompting largely depends on how the questions or instructions are phrased. A well-crafted prompt will help the model generate more accurate and relevant responses. For example, a clear and specific prompt like “What are the effects of global warming on marine biodiversity?” will help the model generate a much more targeted and detailed response than if the user simply asked, “Tell me something about global warming”. This is how The NorthFace, in collaboration with IBM Watson, set up a series of preliminary questions to provide a context conducive to quality recommendations for its personal shopper experience.

In some cases, more elaborate prompts can be used to “program” the model’s behavior in a more complex manner, by using specific instructions, providing context, or asking the model to take on a certain role. For example, a prompt like “Imagine you are a tour guide in Paris. Describe the main points of interest to me” will encourage the model to generate text from the requested perspective.

Our experience with Search Commerce with GPT

We recently developed an embryonic Search Commerce solution with GPT in our company, based on the Gally searchandizing solution. Here’s some key feedback:

1. Understanding customers is key Before implementing this technology, it’s important to spend time analyzing your customers’ search behaviors. Fortunately, we were able to draw on the expertise of our teams responsible for developing the #1 Searchandizing solution for Magento: ElasticSuite. This enabled us to save time on prompting GPT to best meet their needs.
2. Integration requires technical expertise Integrating GPT into your e-commerce platform requires the help, for the time being, of a development team to connect the search system with the natural language interface developed with GPT. It is important to note that the implementation of this technology may require AI skills, especially in the field of prompting.
3. Testing is crucial After integration, we spent a lot of time testing how the model perceives the content, the data model (categories, as well as all other structured data such as colors, materials, etc.), but also the API of our searchandizing solution. This was an iterative process, but essential to ensure the quality of the queries sent to the search engine, as well as the responses generated by GPT.
4. The results are promising With the implementation of this technology, we have seen an improvement in the relevance of search results, which we can expect to have a positive impact on conversion rates. Although the initial results are very encouraging, this raises the question of how users search for products. Nowadays, it’s essential to simplify our queries when using a search engine on an e-commerce site. With these new solutions, made possible by large-scale language models, users will be able to describe their needs directly, in natural language. From “red cotton dress” to “I need a dress for an afterwork party at work this Friday”. We see this as a significant transformation in usage, and one that requires a great deal of work to rethink the user experience.

Conclusion

Search Commerce with GPT offers great potential for improving customer lead acquisition. By better understanding customer queries and providing more relevant answers, this technology can help companies attract and convert more customers. Of course, to work at its best, your product data must be up to date (attributes, descriptions, categories). So content remains the sinews of war.

Our experience has been positive and encourages us to explore this technology and discover how it can benefit their business. If you’d like to find out more about how our team set all this up, you can read the dedicated article on GPT experimentation on the Elasticsuite blog.

Follow our adventures on our social networks, and if you’d like to keep abreast of the latest innovations, come and share our technology watch by subscribing to our tech watch.