Using (Large) Language Models for Your Business

How can any business use language models to actually drive either higher sales or lower costs? This article is looking at how language models can lower a company’s operating costs in customer support and onboarding in general. The article is based on real life lessons we’ve been gathering while launching our AI concierge platform, Clipport.

The capacity of any business to use, acquire and improve knowledge and knowhow has always been the key ingredient in driving success. From manufacturing to e-commerce and consulting, the ability to manage knowledge both internally and externally has set apart successful enterprises from the rest. This is not simply about storing information to be accessed at a later date. It is about being able to build the ability to use such knowledge continuously for specific situations while growing the business - and this is not cheap. Just check the cash burn of most e-commerce businesses.

Language models are the AI product driven interface that can allow businesses to automate the above processes, for internal or external use, specifically because they are able to deal directly with text, just like a person would do. They come in various sizes and capabilities, just like people do. Recent technological and research developments have led to major breakthroughs in such language models capabilities, with ChatGPT shooting to fame towards the end of 2022 being one such example. Given their capabilities, the natural question becomes how can any business use such technologies for their daily operations, in a low cost manner and without increasing headcount?

The answer to this question is key to any successful deployment of such technology. This is because despite the capabilities of ChatGPT or similar models, they are no better than the smartest person out there joining a company for the first time: they know nothing about such company internal processes, about how the people working there use such processes, or about how to work as a team member with the people there. 

There are 3 main features that can define the use of a language model for a certain business task:

1. the size of language model, 

2. the nature or complexity of the task,

3. and the available data to train.

While there is a linear relationship between the size of the language models and its capabilities, i.e. the bigger the better, the quality of available data can materially change such a relationship. This is because any business has very specific needs. For example a car dealership doesn’t really care if the language model works well for a refinery, just as an investment bank doesn’t care if the language model works well for a coffee shop. Each business has very specific requirements internally and for their customers, and need such language models to work very well for them first.

This is why “available data to train” becomes very important, especially when it comes to costs. Similar to people, no matter how “smart” the language model is, specific business knowledge is always necessary to get better. Many start-ups, as well as many large companies have been investing millions of USD to develop such specific language models that they can use for their own needs, with mixed results in many cases. The cost for building such customized models meant that many small businesses could not afford to build and deploy one. Furthermore, once such models were developed, they needed (and still need) expensive infrastructure to run on, putting the small and medium businesses at a disadvantage.

The advantage of the very large language models, like GPT-3 and above, is that in many cases businesses that do not have initial data to train their models, can start using such models directly out of the box. For many basic functions like basic customer support, product booklets and description, internal knowledge databases, large language models are enough to help provide the correct answers in around 80%+ of the cases. While this is not 100%, it can make one person today as productive as 5~10 people when it comes to customer and business support, with less than 5 minutes set-up time and  zero IT knowledge required (3 clicks on Clipport for example).

The best things happen after one starts using language models though. Once a language model starts “working”, it is able to start collecting all customer inquiries. It’s only human to forget, and I doubt most (any?) sales people or customer support professionals would remember a random client question from 3 weeks ago. Computers don’t forget. In time (think only a few months, not years), this allows businesses to continuously update themselves, further improving customer service and support, and also to see what customers really want, and especially how such customer needs change over time. Current technologies even allow for basic analysis of such customer interactions, like specific interests, errors, identifying lack of resources that were not available, etc, in real time. No need to wait for the end of the quarter to get a report with such developments.

And this is actually a critical point for any company out there, be it a startup or small shop or a large corporation. Being able to keep in touch with clients and customers by answering their questions on the spot is what will make customers feel good and happy - and happy customers always like spending more. Just ask your local bar. More importantly though, any company will understand first hand what they need to offer their customers - they’ll just have to see what the AI concierge could not deliver.

So how can a language model help your business grow faster?

PS: If you are a startup or SME and would like a short discussion about how language models can help your business, feel free to book a meeting. Only a few slots available, and it’s free @ https://finrsch.com/consulting.html