LLMs for Business: A Game-Changer for Productivity and Customer Engagement
LLMS, or Large Language Models, are a type of artificial intelligence that uses natural human language, the kind we speak every day, to perform tasks. An LLM analyses and generates content by imitating human language, using the data it was trained on. This means that an LLM's answers are based on information it already has and the patterns it has learned. The content it creates isn’t just a result of the data it was “fed,” but also the algorithms added during the tuning and testing phases. That’s why tools like Chatgpt, DeepSeek, or Gemini won’t answer questions about building a bomb or use offensive language. However, since they work with knowledge provided by humans, they can still reflect stereotypes or create information that isn’t true, which we call hallucinations.
Today, there are about 20 major, well-known models, but on Hugging Face (one of the biggest platforms for ready-made models), there are over 1.5 million different types and versions of LLMs. Models differ in size and the number of parameters, which affects the quality of information they provide, how fast they work (the largest ones are usually slower), their specialisation (some are more creative, others better understand long contexts), their license (some are open-source), and their cost. For example, models like Mistral or Llama are smaller, faster, and can be installed on your computer, while Chatgpt or Perplexity are more versatile, pricier, and can’t be downloaded.
To use LLMs, companies don’t have to build their own models from scratch. They can use ready-made solutions tailored to their industry or type of service. Let’s look at a few areas where models can help grow a business:
1. Customer Service. LLMs can quickly and automatically handle customer questions about products, orders, or complaints, and can do so in different languages. They can summarise customer conversations or emails, saving the support team time. They can also analyse technical questions and suggest answers to common problems. LLMS can personalise responses, adjusting the tone and style to the context or even the customer’s mood (for example, prioritising irritated customersJ). Companies can implement chatbots, virtual assistants, automatically generated reports, or AI agents for these tasks.
2. HR Support. Models can automatically review and assess job applications, helping HR departments with the employment process. LLM creates training programs based on employees’ knowledge, skills, and preferences, and, by integrating various sources of employee data, such as reports, ongoing feedback, and periodic reviews, provides an overall assessment and suggests further development or career paths.
3. Business Decision-Making. Integrating LLMs into company operations can help with business decisions by analysing market trends or monitoring risks, using data and recurring patterns,for example, through the analysis of press articles or current news. Banks and financial brokers can use them to predict market behaviour and suggest whether to make a move or wait. Of course, this doesn’t eliminate all risks or the chance of unpredictable “black swan”[1], LLMs can be a valuable source of insight and help make decisions.
4. Content Creation and Document Review. LLMS are incredible at preparing all kinds of summaries of meetings and documents, as well as creating action plans, diagrams, or organising information logically. Marketers can use them to prepare social media content or entire advertising campaign plans. When adopting popular models, it’s best to avoid asking questions in a negative form, as the model might get stuck on the question itself without understanding the negation. For example, instead of saying, “Describe the most important features of a helmet for people aged 20-29 who ride bikes in the city, but don’t include technical details,” it’s better to say, “Describe the most important features of a helmet, focusing mainly on style, color, and lightness.”
5. Supply Chain Management. If you run a company that delivers physical products, LLMs can help manage warehouses and inventory, predict demand while considering delivery times and seasonality, and more, for example, by querying the chatbot about available resources or products that match specific customer inquiries.
6. Fraud Detection. By analysing large datasets and patterns of acceptable behaviour, and constantly monitoring them, LLMs can spot unusual interactions or spikes in activity and send alerts. This is especially important in finance, retail, and e-commerce, helping reduce risks like money laundering, credit card fraud, identity theft, and insider trading.
There are many ways to use LLMs in a company, from implementing a chatbot or speech-to-text program to preparing a model for your own specific needs, which is a bigger and more costly challenge. This last approach means training the model on specialised datasets or changing its knowledge structure so it communicates in the desired way and better understands the specifics of a language (like Polish, Latvian, Spanish, medical, or legal language). This allows it to give more precise answers than general models.
Effective training and customisation require providing diverse and representative training data, selecting sources carefully, and iterating with the model to minimise mistakes. This is especially important if your main goal is accuracy and reducing hallucinations.
Customising models can be useful in areas requiring highly specialised knowledge. For example, in medicine, you can tune an LLM using patient records, test results, medical literature, and treatment guidelines to help choose the best therapy. Automating tasks with LLMs can also help with work, like summarising symptoms or medications. The entertainment industry can benefit by recommending and personalising information based on a user’s browsing history and preferences.
LLMs are also useful in manufacturing, managing resources, and keeping production running smoothly by predicting when equipment needs maintenance. This predictive maintenance extends the life of machines and reduces the risk of breakdowns. Analysing production data helps identify downtime or inefficient points in the process.
To give a complete picture of LLM use in businesses, we should also mention concerns about their use, such as ethical questions, content authorship (should an LLM be listed as the author?), and how much we can trust content generated by models. Noam Chomsky, a well-known American linguist and philosopher, even questions their intelligence, seeing them as machines that create answers without understanding their meaning. “Whereas humans are limited in the kinds of explanations we can rationally conjecture, machine learning systems can learn both that the earth is flat and that the earth is round. They trade merely in probabilities that change over time.”[2] But these issues are the subject of another article.
Author: Patrycja Dąbrowska-Wierzbicka
[1] Events that are unpredictable but have a very big impact on reality, see N.N. Taleb “The Black Swan. …”
[2] https://www.nytimes.com/2023/03/08/opinion/noam-chomsky-chatgpt-ai.html#