The goal of every chatbot is to analyze the customer’s question, take the shortest road to the query, report it, and then escalate it to the right department. When needed, it can ask additional questions if any details are missing.
Before building the solution for our client, an analysis of the project were needed to be made. Focus was made on collecting information on internal processes and regulations. To be able to synchronize to the chatbot technology, it was required to select the business processes that were repetitive and easy to optimize.
During the analysis, we chose the area to improve and made sure it’s the right field to optimize – first of all, if the process is repetitive. Nevertheless, the bot had to learn a number of queries and additional related questions. The main goal of the bot implementation was to shorten the way of the query – to report it and escalate it to the right department keeping the deadlines.
Another important aspect was to gather all of the details of the query, and ask supplementary questions if necessary. We set up a range of questions that need to be answered in order to forward the query to another department. We also created numerous scenarios for a bot to be able to answer as many questions as possible. The next step was to create the dialog trees for the personalized conversations. The bot was expected to be able to analyze the details provided by the client and ask additional questions if any of the details were missing.
Once these steps were completed, the MVP of the bot was released for further testing. The bot’s performance was measured with an accuracy indicator, which can reach even up to 93%, so it is crucial it can respond to questions correctly. The bot is programmed in such a way it can be adjusted after the implementation, for instance, to support the sales process or present special offers. Additional features, such as small talk ability can be developed. The MVP phase was very helpful and allowed us to eliminate the inaccuracies, where some of them couldn’t be predicted.
The team consisted of 2 Developers, a Business Analyst, and a Knowledge Engineer. The Analyst handled the mapping of internal processes in the organization. The Knowledge engineer built conversations. Together they were responsible for bot improvements. They performed testing, conversation analysis, fixed errors, and eliminated irrelevant replies. The UX designer accompanied the team to design a conversation window.
The technology stack we used in the process included Java, Apache, Cassandra, and Spring Boot. At the time we have written the so-called entity extractor based on NLP, and right now there are available libraries on the market.