As chatbots grow, it can be challenging to get an accurate view of your chatbot users' experience and prioritise which content to update. The inGenious AI platform provides a rich set of reports and features to help you monitor and improve your chatbot experience.
Review and prioritise content
Keeping your chatbot's question-matching as accurate as possible is crucial for a good chatbot experience. Reviews help you identify where your chatbot most needs additional content and where your question training phrases need more work, so you can prioritise your efforts. You can monitor trends in your chatbot's content coverage and question-matching accuracy over time to keep an eye on the broader picture.
When you've updated your content, regression tests make it easy to test your chatbot thoroughly before you publish changes. Each regression test matches a set of example questions and marks the chatbot's response for you, so you can quickly see how your changes will impact your chatbot.
You can also use the issues report to check your chatbot for problems that may affect its question-matching accuracy.
Analyse patterns and trends
Use transcript reports to monitor different types of chatbot interactions. Each transcript report displays transcripts of chats that match the criteria you specify, such as chats that started a specific passage, used a certain chatbot channel, or occurred with a certain version of your chatbot. Reports can retrieve transcripts from a fixed period of time or from a dynamic period like the last seven days.
If you want to dive deeper into your data, you can create insight reports to analyse patterns and trends and ask questions about your transcripts using natural language.
Create automated tests
Automated tests help you test and improve the prompts for your LLM tasks. You can generate output from a task for multiple test cases at once, automatically evaluate each one, and then generate insights how the prompt could be improved, and some suggested prompt refinements. Each stage in the testing is stored separately, so you can compare results of different prompts and models for the same task.