Process of building a virtual assistant

Process of building a virtual assistant (Virtual Agent - VA) with the EM&AI Virtual Agent platform.

The process to build a virtual assistant with the EM&AI Virtual Agent platform is as follows:

1. Identify functions

Each VA should be built to solve a unique use case. Users need to build functions and tasks corresponding to the business process of a consultant.

For example:

• Bank virtual assistant: Offer Loans, open an account, savings, or open credit cards...

• Insurance virtual assistant: Product consultation, contract information lookup...

• Restaurant virtual assistant: Reservations, delivery, information lookup...

2. Develop a scenario

Determine the context, and break down service cases to create training scenarios for virtual assistants.

Build a response scenario that includes a list of intents and responses to the intents. Set up a flow diagram to train the virtual assistant to respond sequentially according to the context, ensuring smooth, natural conversations.

Flow diagram

3. VA Setting

From the scenario, the AI trainer gets the context, list of intentions, responses, and conversation flow. Use the Virtual Agent platform to set intents and responses. The trainer can add a small number of sample training sentences at this step to test the VA's response.

4. VA training

VA training is an NLP training process for VA to understand and process any customer statement.

From the scenario, the trainer builds NLP training data. Training data is a list of a large number of sample sentences, intents, entities, and keywords. Corresponding to each intention, you should prepare at least 20-30 sample sentences showing different ways of expressing the customer.

Upload the sample data set to the EM&AI Virtual Agent system, conduct NLP training, and calibrate until the NLP recognition model meets standards.

5. Check & Adjust

• Check the NLP

• Check the context.

• Check the bot's response results on messaging platforms integrated with the system (Web, Messenger, Zalo).

• Adjust the results until the bot responds according to the correct scenario.

6. Deployment

After completing the setting and testing steps, deploy to interactive platforms (Messenger, Zalo, Website, Application...). Set up necessary integrations with business systems (CRM, Service desk, contact center software, etc.) to ensure a smooth experience.