Everyone talks about, over and increasingly with chatbots. But what should be considered when developing a chat bot? In addition to mastering the technical challenges, detailed preparation and planning are also included in clearly defined 7 steps to the chatbot.
Like private households, online shops also use language-recognizing and writing chatbots and talking virtual agents when it comes to providing standardized information or pre-filtering inquiries. The number of medium-sized companies that use or develop chatbots is also growing. Gartner expects chatbots to handle 85 percent of customer service interactions by 2020. In 2017, only four percent of companies had installed chatbots, yet 38 percent of all organizations are already planning to use this technology.
The right team as a prerequisite
The recipe for successfully implemented projects is a highly qualified team that meets the challenges. “Before the actual conception, therefore, a suitable mix of abilities, skills and experience in the project team has to be taken into account. Thus, deep business knowledge of the subject area, which the chatbot should cover, is just as essential as the way in which users will interact with it, “says Simone Schubert, Project Manager KI Chatbot at mip.
In addition to proven experience in SCRUM-oriented project management and the integration of cloud-based program interfaces into the corporate infrastructure, strong know-how is required to analyze interactions and train services. Examples need to be developed to promote machine learning and define business rules. In addition, proven user experience skills are needed to take holistic customer experiences into account.
In doing so, skills for interaction with different business and technical stakeholders are outstanding. Finally, it is important to consider the overall architecture and an end-to-end solution approach at the same time. In terms of agile software development and the resulting continuous test runs, not only flexibility in changes, corrections and enhancements is required, but also and a comprehensive project overview is needed. Knowledge Management also promotes a transparent project process. To develop, introduce and operate chatbots successfully, a very thorough conception is recommended.
Step 1: Develop requirements and area of application
Before any consideration of implementation, it is important to create a comprehensive definition of requirements and application areas. The following points should be clarified around what the bot should do:
- Which goal is pursued (entrepreneur’s view)?
- What added value is generated for the user?
- What is the purpose of the chatbot (user view)?
- What does the target group look like?
- Which success criteria must be met?
- Which restrictions apply?
- Which interfaces are there? How are they served?
Step 2: Set the solution design
Before the bot development is to clarify in which IT infrastructure the chatbot should work. This prevents considerable additional expenses, for example through the subsequent integration of databases or other systems. To assess the IT-technical requirements and possible equipment extensions, it is therefore advisable to take stock of the existing IT landscape and to compare the different technologies on the market:
- User Interface: Which interfaces should be supported?
- IT infrastructure: Which systems are in use?
- Back-end systems: Which servers, disk / tape storage and central network components are used?
- Where is relevant data stored?
- Will the application be a cloud, on-premise or hybrid solution?
Step 3: Create a questionnaire
Chatbots work within a defined information space and access predefined data to handle requests for information, searches, and queries. Therefore, it is advisable to collect the highest possible number of question variants, at best through user surveys. A number of 300 to 500 representative questions are suitable as a good starting point.
Step 4: User Intentions Defining Conditions and Context
For a fluid course of the conversation, the user’s intentions, the conditions and the context are precisely defined. Thus, the intentions (intents) of the user, for example, “search for product” or “information request”, determined exactly. Likewise, the entities or the relevant conditions (entities) must be defined, for example, the shop in which the product is in stock. As conversations become more relevant and personal, the more context is known, care must be taken that such “background information” can be taken into account. This includes, for example, information already mentioned in the conversation, personal characteristics, location, type of device used by the customer, product preferences, and history of previous contacts.
Step 5: Determine dialogue between chatbot and human
There are two possibilities for the dialogue between chatbot and human. During the guided dialogue, the bot gives the questions and guides the user through the conversation. In free dialogue, the user can ask questions himself. In order to cover as many customer inquiries as possible in this case, the dialogue structure must be clearly designed. Only then does the chatbot know which answer to give to which question. Possible are simple answers with / without variables as well as complex answers, for example with forwarding via link. The selection of the respective structure depends on the goal that the chatbot is to achieve. The result should be a structured dialogue process that determines how the application responds when the defined intentions and entities are identified.
Step 6: Configure User Interface
Already at the beginning of a project, it must be clearly configured where the chatbot will be used and which integration and communication with external systems will take place. Thus, a Chatbot can be integrated into an already existing application environment or programmed as a stand-alone solution.
Step 7: Create a prototype
After conception and design it is advisable to create a prototype as early as possible, which will be further developed agile and iteratively. The status quo of the development can be captured quickly in this way, the later user experience can be tested directly. On the basis of these experiences, a concrete feedback and a timely intervention for changes is possible. Depending on the requirements profile, Chatbot development repeatedly cycles through the cycles of testing, optimization, feedback, analysis, and adjustments – until the desired result is achieved.
After the live start of the chatbot, the participants should closely observe and investigate the operation. The intensive monitoring helps to identify and integrate previously unanswered questions and use cases. Which intents need to be added? Which answers are not satisfactory? Also during further operation, the monitoring and maintenance must be continued constantly. Here, the focus is primarily on training the chatbot. “Our experience has shown that after initial day-to-day testing, these continue to decline. Do not forget to observe and evaluate the user behavior. Because this information helps Chatbot to become even better, “says Simone Schubert.