Launching a Virtual Agent is an exciting and challenging project. There are several aspects to consider making it successful. Steve Heggen, Head of HyperAutomation at Fujitsu Luxembourg, experienced in their conception, shares some tips with you. 

        1. Define the identity of your virtual agent 

To begin with, it is crucial to think about the way the bot will introduce itself. It should always be clear that the user talks to a virtual agent. The bot can be given a name. 

Then, it will explain the topics to which it is dedicated. These should be specific, for example the introduction of a support ticket (similar to a help desk call). And if it is possible, the bot will guide the conversation with a first closed question. If the bot is dedicated to the submission of support tickets, it could be « Hello, for which application do you want to introduce a support ticket?”. The bot can as well show some empathy in its replies, but not too much as it may lead to misinterpretation. You prefer to avoid a situation in which your bot is cheerful if someone is absent for a familial funeral.

         2. Select the right target for the bot

In order to get a valuable chatbot, you need to choose an expertise field that will be relevant for most people or for a representative sample of people about the most requested topics. This topic has to address simple questions.

         3. Select the most frequently asked questions 

Think carefully of your questions series. It is not recommended to add less frequently asked questions, such as those asked every six months or too specific ones. Indeed, it is preferable to follow the concept of 80-20. The bot takes in charge 20% of the common questions that represent 80% of the volume and vice-versa for human support, who will take in charge the more complex questions. 

          4. Provide simple answers for a better impact

Avoid answers that are too elaborate and keep your answers short. If the answer needs more details, suggest a specific documentation. 

           5. The bot should always fall gracefully

Follow the three strikes rule. When the bot does not understand a user’s question, it should not answer systematically negatively which could result in user frustration. It needs to proceed in 3 steps to avoid the situation: 

1. Ask the user to reformulate its question

2. If it still does not understand the request, the bot must remind its scope of activity, the context in which it is skillful and direct the user elsewhere

3. If it is still problematic, it must invite the user to use another reference (via documentation), or by handing over to a human support

         6. Make sure you include all people concerned by the chatbot from the beginning

It is important to make people, who are going to work with the chatbot, part of the project. For instance, if you plan to create an HR bot, you should involve the HR team. Make the team quickly aware of what the bot can and cannot do, explaining how the bot will support their work. An awareness session will help a lot. 

        7. Organize your testing sessions 

Proper testing and training of the chatbot is essential. It should not involve 1 or 2 testers for 2 weeks, but rather 15-20 testers for 30 minutes each. Indeed, the bot needs to adapt itself to many subtleties of language. This means that the more people test it, the more efficient it will be. If the same person tests it for a long time, the result will not be as fruitful.

        8. Think of project evolution

Last but not least, think about the project evolution after its launch. The bot must be maintained and fed. This will avoid the bot becoming ‘insipid’ over time.