The use of Large Language Models (LLMs) is becoming increasingly popular, because let’s face it, who hasn’t used ChatGPT these days? But not everyone knows exactly what it is and how it works. In this blog, we explain what Large Language Models are and how they are integrated into CreditDevice’s systems.
Let’s start with the term Artificial Intelligence (AI). If you google this you come up with the following definition, “Artificial intelligence is the ability of a computer system to mimic human cognitive functions, such as learning and problem solving. Basically, this means that you create an algorithm (all difficult pieces of code) that can recognize patterns from data, and then use this knowledge to solve a particular problem. This algorithm then functions all by itself, so certain (business) processes can be automated. Very handy!
How exactly do LLMs work?
LLMs, or Large Language Models, are advanced language models that fall within the AI family. They function as powerful algorithms trained on huge amounts of text data. This allows them to recognize, understand and generate human language. An important aspect is that a language model itself has no real language knowledge; it works purely with mathematical representations of words and sentences. This means that all language operations take place through complex calculations and patterns.
The benefits of LLMs in credit management
The strength of LLMs lies in their ability to interpret and generate language in a human way. This opens the door to more efficient processes, such as automatically processing customer requests and handling communications faster. For credit management organizations, such as CreditDevice, this can mean automating time-consuming tasks without sacrificing quality and customer focus.
Application of LLMs in CreditDevice’s software
First of all, you should think carefully about what you are going to use a language model for and whether you want to use a language model at all. Nowadays, every site has a chatbot, but as a customer, do you really need that? Wouldn’t you rather speak to an employee right away if you have an urgent question? In the back of your mind, you should always consider what the added value is for the customer.
Within CreditDevice’s accounts receivable management module, we have integrated a language model that can classify mails into different categories. For example, a request for a copy invoice or a payment commitment. In the case of the classification “request for a copy invoice,” these invoices can be automatically put into a mail, so that all the customer has to do is send the mail. In the case of a payment commitment, this information can be fed back to our accounts receivable management module and, if desired, the action path for these invoices is temporarily stopped, for example.
Privacy is paramount to us
Since we value the privacy of our customers’ data, we decided to keep the whole process in-house. This means that no data is sent to OpenAI, for example, or stored on another server. The whole process occurs in our own servers. After various language models were trained and tested extensively, the most accurate algorithm was selected. This algorithm achieved an accuracy of about 95% within the different classifications!
