Large Language Models (LLMs) in Retail: Here's How to Get Started
Large language models such as Chat GPT offer numerous relevant applications in the retail sector. This technology, which has dominated the news over the past year, can be utilized by retailers in various ways.
Let me briefly discuss some applications and explain what infrastructure you need to get started.
What are Large Language Models?
First, let's clarify what large language models (”LLMs”) like Chat GPT are. GPT stands for Generative Pre-trained Transformer. This means that the model generates text (Generative) based on a vast amount of pre-processed information (Pre-trained). The “T” stands for transformer. This is a complex technical concept representing a specific type of language model. Functionally, this form enables a model to better understand and remember context. This allows for the formation of longer coherent texts, rather than just isolated sentences.
Context is crucial for retailers. If you use LLMs correctly, by providing them with sufficient context about your own setting, you can achieve many interesting outcomes.
How to get started?
Getting started is easiest in three steps. Begin by experimenting with Chat GPT itself, move to applications that you can buy “off the shelf”, and finally, see if you can integrate language models into your own systems or applications. This last step essentially involves making your own data available to such a model. This makes what a model can do much more relevant.
Right off the bat!
The most accessible way to start using LLMs is through the chat interface of existing tools like Chat GPT or alternatives like Gemini or Claude. Here, you can summarize texts, write new texts, conduct online research, explain complex concepts (with examples!), brainstorm new ideas, speak with a fictional expert on a particular topic, and much more.
Additionally, the latest versions of these models can also work directly with images, extracting information from them and generating new images or icons.
A very handy application is, for example, help with creating Excel formulas. This is often a cumbersome and error-prone task. There are useful plugins for this, but you can also ask questions directly and specify in which cells certain inputs can be found.
Retail-relevant off-the-shelf tools
Besides direct use in your personal workflow, you can find a lot of tools where this new technology is integrated. Specifically for retail, we mainly think of tools that facilitate the following applications:
- Customer Service: Automating customer service through chatbots that can conduct natural and contextual conversations. These chatbots can help customers with questions about products, orders, and returns.
- Content Creation: Generating product descriptions, marketing materials, and even blog posts. This can save time and costs and increase the quality and consistency of the content.
- Data Analysis: Performing simple data analyses, such as identifying trends in customer behavior or sales figures, without requiring deep knowledge.
- Personal Assistants: Developing personalized assistants that perform specific tasks, such as writing and proofreading emails, brainstorming marketing campaigns, or helping create complex Excel formulas.
- Recommendation Systems: Enhancing recommendation systems by using more contextual information, such as customers' personal preferences and purchase history, to make more accurate and relevant product recommendations.
Some favorites that are definitely worth checking out are:
- Kamichat: There are many AI chatbots, but this one can do just a bit more and integrates with above-average many data sources.
- Fathom: An AI note-taker that integrates with the most common online meeting tools. Works both in Dutch and English. It sends you a summary afterwards that is often quite accurate.
- Tavily: When a task is just too complex for Chat GPT, this layer above Chat GPT can often help. For example, research involving consulting multiple online sources becomes much more powerful with this tool.
- Zapier: A veteran in automating tasks, but one that immediately jumped on the Chat GPT train and can therefore perform significantly more tasks.
Integrating LLMs into your organization
The right context makes these systems extraordinarily performant. Providing the context of your organization means that these systems gain access to product information, customer information, transactions, returns, complaints, emails, etc. By tapping into these sources, these systems can suddenly provide extraordinarily relevant input.
Integrating such software may seem like a big step from your current infrastructure, but it isn't. As a retailer today, you often already have a classic data warehouse and/or a SharePoint available. Essentially, it comes down to making the information from these systems accessible to a model like Chat GPT.
An LLM performs much better when it has the right context. This means enough information but not too much information. Consequently, there is a need for an easy way to search systems and then provide the most relevant information to the language model. Technologies like Pinecone can easily support this process. Such a system indexes all available information, such as product data, images, and customer information.
Experimental use-cases
It is clear that anyone can start summarizing meetings via ChatGPT or requesting a formula for Excel. However, the most impactful use cases are those where you try to combine the power of your current systems with an LLM.
One initial application we help implement for clients involves product information. Having the most relevant product information for each item can be crucial for various processes.
This can allow category managers to better evaluate their assortment. For example, do I have waterproof shoes? Do I have enough shoes around a certain price point?
For a second application, we look at the substitutability of certain items. What is the likelihood that if one product is out of stock, a customer will still buy another product or leave with a different item? This has a significant impact on decisions related to assortment composition or stock management. With ChatGPT and a number of product characteristics, you can easily evaluate substitution scores for certain goods.
What should retailers focus on in the coming months?
To effectively implement this technology, retailers need to think about their technical infrastructure. This can range from integrating existing systems with new AI modules to setting up more advanced data warehouses that can provide the necessary information to the language models.
It is important to start with simple applications and gradually move to more complex and integrated solutions, allowing your teams to learn. At Crunch, we help teams get started by setting up the correct infrastructure.
Additionally, in the coming years, it will be crucial for retailers to connect and strategically deploy the right set of tools. When multiple AI models with different strengths collaborate, they can compensate for each other's weaknesses and provide a more comprehensive solution.
By bundling tools, you can reduce the complexity of one large task. This essentially involves splitting a desired outcome, such as answering the question 'which product should I offer customer X?' into targeted tasks. In this process, advice from consulting agencies with a specific focus on data and AI can also set you on the right path.
Follow us
Would you like to learn more about the latest data & AI trends in retail? Take a look at our website, follow our LinkedIn page, or sign up for one of our webinars.
