Chatbots and Virtual Agents
The first and most familiar use case is chatbots and virtual agents. They hold a conversation with a customer, answer questions and provide support around the clock. With an LLM behind them the answers are quick and relevant, which is what most customers want from support.
There are already real successes here. Klarna, for example, reported that its AI assistant handled two-thirds of all customer service chats.
- The AI assistant has had 2.3 million conversations, two-thirds of Klarna’s customer service chats
- It is doing the equivalent work of 700 full-time agents
- It is on par with human agents regarding customer satisfaction score
- It is more accurate in errand resolution, leading to a 25% drop in repeat inquiries
- Customers now resolve their errands in less than 2 mins compared to 11 mins previously
Klarna expects it to add USD 40 million to profits in 2024, on a technology investment in the single-digit millions.
Klarna's support is relatively low-touch, but even high-touch support departments can use such a system to help staff answer inquiries faster.
Co-pilots
Co-pilots work alongside a person. They help with writing, coding and design by suggesting, correcting and sometimes generating content, and they make complex tasks more manageable.
Co-pilots are already part of many developers' daily work, so there is good evidence on how well they perform. GitHub, which released its Copilot two years ago, ran an a/b test with the following results:
- Higher rate of completing the task (78%, compared to 70% in the group without Copilot).
- A striking difference was that developers who used Copilot completed the task significantly faster (55%) than the developers who didn’t use Copilot.
Just as interesting, they also reported that:
- Between 60-75% of users reported they feel more fulfilled with their job, feel less frustrated when coding, and can focus on more satisfying work.
- Copilot helped them stay in the flow (73%) and preserve mental effort during repetitive tasks (87%).
Scott Galloway, an NYU professor, wrote an extensive piece on why AI is already responsible for the recent round of layoffs and record profits posted by tech companies across the board.
Chatting with Data (RAGs)
RAG, or Retrieval-Augmented Generation, lets an LLM pull information from your own sources at the moment a question is asked, and use it to answer, recommend or explain. In practice it feels like chatting with your data, and it makes that data far easier to get at for the people who need it.
This use case mostly involves internal company data, so public success stories with measurable results are harder to find. OpenAI did publish a case study of such a system at a financial institution, and the numbers are impressive. They started with an accuracy of 45% and over time raised that to over 95%, which made it a handy tool for employees looking for internal information.
Classification and Typical NLP Tasks
LLMs are very good at classification and other Natural Language Processing (NLP) tasks, such as sentiment analysis. That lets a business automate the reading and sorting of text, from customer feedback to market research, and get more out of data it already has.
An LLM can sort and categorize emails, support tickets and social media posts, so a company can prioritize and respond to what matters. For large businesses with a diverse customer base this makes a real difference. The same goes for content moderation, where LLMs identify and filter inappropriate or irrelevant material automatically, keeping platforms safe for users and taking work off human moderators.
Building a good NLP system used to be hard work, with a lot of manual effort to tailor algorithms to a specific language or task. With LLMs, use cases that were too complex before have become viable, and the results often beat traditional systems. An LLM adapts to a wide range of tasks with little customization, and usually with better accuracy.
Autonomous Agents
Autonomous agents are the newest of the five. They perform tasks, make decisions and learn from their interactions with little human oversight, from automating routine work to managing whole systems. They give a sense of where AI is heading, with machines taking on more responsibility and initiative.
It is probably the most complex use case, but there is a lot happening. Sales automation is a clear example. You can give a virtual SDR the assignment to grow your pipeline, along with access to your LinkedIn account, and it will:
- Discover the best-fit customers and establish a lead list
- Research all the leads in-depth
- Personalize outreach to drive conversions
- Engage with prospects on multiple channels to get you meetings
The results for systems like this are early and mostly self-reported by the companies that sell them. Still, it does not take much imagination to see that if these agents do their work well, sales people can spend their time at the end of the funnel, where it counts.
And your business?
Large Language Models are changing how industries work, and we have only just started exploring what they can do. We hope these five use cases give leaders enough of a picture to start tinkering with where LLMs could fit in their own organization.
