BEST PRACTICES FOR TESTING YOUR BOT
Testing is a crucial point in the development of your virtual assistant. The quality of the product that will be released is directly related to the amount of time and effort you spend on this final stage of the project.
Purpose: Submit AgentBot to a scenario as close as possible to the real usage it will have when released online, so the content team can adjust the conversation flow or load more questions.
Input: Initial content shared with the PL—the questions and answers developed and loaded onto the platform. We recommend testing only the questions in this document. If you would like to include more information, consult with your PL for the next steps
WHO SHOULD TEST?
It is important that the final tests are performed by a small group of end users (your customers) or by the team that works directly with this audience, who knows how your customers express themselves, are familiar with their vocabulary and, most importantly, with their typical questions.
• It is common that the FAQs the company provides do not necessary correspond to customers' real needs.
During the test, it is important to try many alternatives and ways to ask questions, typos, spelling mistakes, abbreviations etc. Test the content that has been uploaded to your virtual assistant.
To ensure a successful testing process, it is important for all parties involved in this stage of the project to be aware that:
•This is a beta version whose purpose is to find improvement opportunities. Every mistake that happens during the testing stage prevents the same mistake from happening to a real customer.
•The test is about the uploaded content. Your testers should be aware of the scope and topics that your virtual assistant understands and is ready to respond to when launched.
• The navigation assistance tool (Change URL) will not operate correctly during the testing phase. In answers that include this tool, the virtual assistant will disappear, and you must return to the original demo to continue testing. This is because the assistant has not been incorporated onto your website yet.
WHAT IS THE METHODOLOGY?
The content team, besides having entire conversations to analyze, pays special attention to the following points.
•Unresolved questions: Questions that do not have an answer. AgentBot registers these questions in a separate section called “Improvement,” where these questions are either uploaded as new ways to ask or are used to formulate new content.
• Wrong resolution: Questions that received an answer, but not the correct one. Each answer can receive positive or negative feedback. Testers should use the negative feedback feature to highlight what interactions should be reviewed. If the virtual assistant gives the wrong answer, you should give negative feedback.
WHAT SHOULD YOU EXPECT?
The AgentBot content team has already carried out an internal test on all the questions and answers loaded onto your virtual assistant. Based on this, corrections and improvements have been made so that the demo is now ready for external testing.
The information that comes from these external tests includes new combinations and ways to ask questions, technical vocabulary, new words and abbreviations that are specific to your assistant.
All these new ways to ask may not be properly answered by the virtual assistant on the first attempt. The content team will be working with this new data to adjust and improve the content prior to launching.
During this stage, we expect that AgentBot will not be able to solve all questions. It will take a few months to develop a solid automatized customer service solution.
With time, as the traffic increases, so does Agentbot's answering capacities.