Implementing Large Language Models and Generative AI, CM.com’s first features

Our last article discussed a few use cases around Large Language Models and Generative AI for our products. Today we are launching the first implementations. Let’s see what we implemented and what the advantages are.

Conversational AI Cloud

CM.com’s Conversational AI Cloud has always been a product that enables businesses to automate their conversational flows easily. We’ve primarily targeted business users and tried to make their lives easier. Ensuring they can spend time doing what matters instead of the busy work involved when setting up and managing a conversational AI platform.

With release 4.8, CM.com introduces generative AI, a complete overhaul of its intent classification recognition engine, and how it’s embedded into the product’s multi-engine NLU, automated testing, and several other changes.

• Generative AI

  Instead of manually adding utterances to train the model, generative AI generates new training data. This helps improve the intent model's accuracy, precision, and recall. It can generate an entire intent model of intents, descriptions, and utterances. It can help synthesise new examples, aiding conversation designers in the creative process required to increase the diversity of the training data and help to overcome any biases in the original data set.

• Quality Control & Monitoring

  Conversational AI Cloud now allows its customers to define a test dataset and their training dataset. This dataset calculates precision, recall, and F1-score on each training cycle, instantly showing you the result of your changes and allowing for precise fine-tuning of your model to match your conversational goals.

• Multi-engine NLU

  Further pushing the collaboration and benefit from Conversational AI Cloud’s intent classification engine and entity-driven rule-based recognition engine. Customers can now leverage the strengths of rule-based recognition in their intent classification model to ensure their end-users get the best contextual answer.

Ultimately, there is no replacing experienced conversation designers, customer insights, and fundamental end-user interactions to drive NLU optimisations. We hope our generative AI will help our customers in the creative process and allow them to go live faster and iterate better.

Mobile Service Cloud

For our Mobile Service Cloud, we’re always thinking of ways to improve our customers’ (and their agents) lives. Ways to increase efficiency and quality and reduce cognitive load on our customers is how we offer value to their business and enable our customers to focus on what matters, their customers and their experience.

With this release of Mobile Service Cloud, we’re introducing our first AI-powered feature: conversation summaries. Currently, when an agent receives a conversation in their agent inbox, they have to read through the intake and, even worse, the entire conversation between their customer and their chatbot. As of this release, agents will receive a summary of the conversation between the bot and the customer. The agents can get up to speed faster and provide a better and more timely first response. Any minor improvement in first response time is also correlated to a higher satisfaction rate, expressed in NPS, CSAT, or CES.

What are the most important advantages of these changes?

With these updates to our Mobile Service Cloud and Conversational AI Cloud, we’ve delivered on our promise to:

• Reducing customer time to value in our Conversational AI Cloud: whether you’re a new customer just starting with setting up your intent classification model or an existing client looking to expand your conversational use cases. Our generative AI will help you along and make the creative process as smooth as possible.

• Reduce agents’ average handling time and time to the first response by leveraging Large Language Models to summarise conversations and deliver value for our customers.

• Improve our recognition by combining the strengths of Conversational AI Cloud’s multi-engine NLU to leverage rule-based control flows in our intent classification engine. Providing our customers’ business users with the ability to easily define, manage, and improve conversational flows.

• Ensure recognition quality and consistency over time and across training sessions. Automated testing and calculating machine learning evaluation metrics such as F1-score, precision, and recall enable fine-tuning and optimisation.

Conclusion

We’re excited to present these new AI-powered features embedded into the CM.com platform. CM.com is continuously looking to improve its products and will leverage the power of AI whenever it adds value. As our teams work hard on new use cases, we look forward to sharing more exciting news towards the start of Q2 2023.