The technology behind conversational AI

A lot of different components are required in order to create an AI-powered digital advisor that can interact with your customers and provide measurable results for your company.

Our conversational AI contains some of the most robust natural language processing (NLP) and natural language understanding (NLU) in the world, in addition to the machine learning necessary to continually improve.

Combined with’s unique ASU (automatic semantic understanding), your new digital advisor will understand the intent of your customers, no matter how they communicate.

Essential elements that are required in order to create an AI virtual assistant

Automatic Semantic Understanding - Keep the conversation going

Conversational AI takes the hard work out of getting to the bottom of what your customers really need.

In addition to text-processing, spelling correction and our machine learning models, we have greater insight into how the machine learning models both think and extract their semantic understanding.

Understanding what words are considered important and, crucially, when they are important, is an essential part of what makes our algorithm unique.

We call it Automatic Semantic Understanding (ASU) and it’s changing the way our clients interact with their customers in a big way.

Conversational AI technology

So how does a digital advisor process an enquiry?

A digital advisor built on artificial intelligence isn't something that just magically understands what your users are talking about; it's all about cutting-edge natural language technology and state of the art machine learning models working together in harmony.


AI chatbot technology: User input

1. User input

The user's message is typically received through your website, but our conversational AI is compatible with the most popular communication platforms, such as Messenger, Slack and Skype.

The message itself can be a simple question, such as asking for the opening hours of your business, or a more complex question that requires secure APIs to process the request, such as starting an insurance claim or paying bills. 

Whatever the user needs and no matter how the message is received, your new digital advisor has the brainpower and flexibility to handle it. 

AI chatbot technology: pre-processing

2. Text-processing

When the message is received, the digital advisor starts text-processing the content. This consists of several ways to "prepare" the message it receives for interpretation and ensures that the conversational AI will understand it.

To achieve this unrivalled language understanding, numerous advanced processes are required. They each play a crucial role in ensuring that your digital advisor will understand what your users are asking for. Examples of these processes are language detection, spelling correction (tailor made for each domain), stemming and splitting combined words.

When text-processing is complete, the message is in the optimal state to be correctly predicted to the right customer intent.

AI chatbot technology: intent prediction

3. Intent prediction

The digital advisor is now ready to determine the user's intention. Our unique multi-level intent hierarchy has been developed to enhance both scalability and accuracy, a design that allows it to handle thousands of intents.

We achieve this through an ensemble of prediction models in our intent mapping process, including convolutional neural networks (CNN) and recurring neural networks (RNN). The digital advisor is able to understand the context of a conversation without forcing the customer to keep repeating the same keyword.

Together, these methods gives us an unrivalled level of accuracy. It’s what makes us able to make the unparalleled claim of correctly predicting an intent with as little as 10 training messages. 

AI chatbot technology: dynamic response

4. Dynamic response

When the intent has been correctly predicted, the digital advisor needs to provide a satisfactory response.

With’s AI Admin Panel, you get the ability to customize and provide a range of different, dynamic responses. They can be pre-defined text replies, which answers simple questions, but they can also be complex answers and actions using entity extraction and external APIs to perform advanced tasks on the customer’s behalf.

How the digital advisor responds is up to you. When you add support for advanced responses, such as letting the digital advisor interact with authenticated users, you can provide the personalized interactions your customers of the future expect - today.

AI chatbot technology: anonymization

5. Anonymization and privacy takes privacy seriously. The final step of the message's journey is the removal of personal data. Privacy, and the protection of a user's personal data, has never been more topical or more important than it is right now.

When using our conversational AI to engage with customers, you can rest assured that you have the tools and features to handle their data in accordance with local laws and regulations. We’ve implemented technical and organizational features you can use to secure the data processing, including features such as full anonymization.’s conversational AI is GDPR compliant and we are constantly developing and implementing better ways to secure both you and your customers’ privacy.

AI chatbot technology: analysis and optimization

6. Analysis and optimization

When the dialogue with the customer has been concluded, it is time to analyze and optimize the data.

Some of your content may need to be expanded through customer intents covering subjects you did not anticipate. Some replies may be sub-optimal, and some could be even more valuable if you connect APIs to provide your customers with a richer and more personalized response. 

These are just some of the issues you might consider. Your customers will notice the difference when your digital advisor steadily improves. We’ll guide you through this vital process and ensure that your digital advisor continues to absorb knowledge and evolve.