Quite often organizations start with FAQ chatbots in trial projects, but don’t go forward to tackle more complex transactions and tasks. Here’s how you can take your chatbot to the next level and realize its full potential to transform your business

Chatbot implementation projects often start with a proof of concept (POC). Generally these involve FAQ chatbots that can automate responses to commonly asked questions via chat interfaces. But so often, such chatbot projects don’t go much further and abandoned due to various reasons including a lack of investment, resourcing and infrastructure required to support, maintain and improve these technologies.   

A 2019 Gartner Analyst report predicted that by 2021, almost one in six customer service interactions globally will be handled by artificial intelligence. But the report also suggested that around 40 percent of virtual customer assistant applications launched in 2018 will have been abandoned by 2020. 

In order to get the true benefits of a chatbot technology, organisations must think long-term. Today’s advanced conversational AI technologies leveraging natural language processing (NLP) are capable of automating more complex transactions and tasks to make life easier for customers and internal teams. 

Building a good transactional virtual assistant also involves tasks that are slightly different from those in making a good FAQ virtual assistant. The former is less about AI, but more about integrations and workflows. FAQ bot development on the other hand, is more about the language model training. That means that building bots that can carry out transactions is not necessarily harder. It just means that careful attention must be paid to ensure that the different systems involved can talk to each other. 

How can an organization that has experimented with a basic FAQ chatbot take the next step towards automating transactions? Here are some key steps to follow: 

1. Start with the Business Priorities and Operational Metrics 

It is important not to get carried away with the results of a “phase 1” of a chatbot project. It might have been well received by users in the initial phase of testing and provided mostly accurate responses. But what matters at the end of the day is how the chatbot helps your business solve problems and achieve objectives.  

Depending on the use cases you are trying to address, FAQs without case specific information may form as little as 7-10 percent of the total questions asked, especially for highly specialized industries like Healthcare and Insurance. In other industries, it could be higher.  

This means thinking carefully about operational metrics. Often, the stakeholders involved in a chatbot project can get fixated with vanity metrics like the technical accuracy of the chatbot responses and number of conversations automated. While these are useful, it is more important to evaluate how the chatbot has moved the needle on key business goals.  

Depending on the business and the use case within the organization, the goals can be different as well. For example, in applications of AI in healthcare the end goal could be to reduce unnecessary visits to the emergency department and man hours for clinicians and nurses through better triaging of patients. Or in the case of chatbots for the insurance industry, it could be a way to improve customer satisfaction, contact centre productivity or support and empowerment of agents.  

2. Identify the Most Important Transactions That Will Impact the Business 

Taking the chatbot beyond FAQ responses to the next level does not have to mean automating all the possible transactions. There will be some transactions that are most sought after by customers, such as doctor appointment booking in a hospital or claims submission for an insurance provider.   

Focusing on these high-volume and high-impact transactions will help organisations extract greater value from the chatbot and move closer to meeting their business objectives.   

To automate these high priority transactions, you will have to enable the chatbot to pull the necessary information from other systems. This will determine the processes that need to be put in place to design a chatbot that can automate the transactions end to end. 

3. Conduct an Audit of the Content and Knowledge Repositories  

The initial implementations of an FAQ chatbot and proof-of-concept projects often have answers that are obvious or readily available to users within an organization. But to take this forward for covering more complex transactions, the system must cover a larger set of answers and information.  

The scope of work here will not be limited to writing the answers, but also in updating, validating and generating answers. In fact, getting a single answer right can often involve holding multiple meetings within the organization, especially if issues arise related to compliance or legal responsibility. Ideally, the audit should reveal the following: 

• Coverage: Percentage of intents for which answers exist in scripts or knowledge sources. 
• Quality: Volume of wrong, outdated or ambiguous information that needs to be updated. 
• Validation: The amount of information that requires validation by subject matter experts and legal experts within the organization. 
• Ownership: The owner of each piece of information. If there is a large volume of information in the hands of a few people within the organization, this can often derail the chatbot project, if these individuals are not involved in the project. 

4. Conduct an Audit of the Applications Involved in Transactions  

It is also critical to conduct a similar audit to identify the specific transactions the chatbot will be required to conduct. The number of applications involved in the transactions can be high depending on the use cases. These applications also must provide a conversational output which can be complex and overwhelming. An audit of the transactional requirements will include the following:  

• Action to be taken: The action to be taken in the back-end systems. 
• Complexity: The input requirements from the user, the number of dialogue paths and fail scenarios. 
• Security: Any form of security or authentication that may be required to conduct the transaction. 
• Integration difficulty: The existence of any appropriate APIs that can be interfaced with when conducting the transaction, the number of applications, and any secondary systems that may be required to validate or transform the user input. 
• Output complexity: Transactional systems are not made for conversation,and often the application’s output has to be converted into dialogue, which can generate additional business logic that can represent a project in itself. 

5. Assess the Need for APIs and Middle Layers to Integrate with Back-end Systems 

Once the audit is done, it is helpful to enlist the services of a technology analyst from the organization to move forward with any back-end integrations. They can help interface with the technology team to check if the data from the various back-end systems can be made available through APIs and connectors. They also ensure that the data is consistent with what the business users expect to see as compared to what they experience through the other platforms. 

If the APIs and connectors are already available, the chatbot can use them and the implementation will be quite straightforward. If not, then your technology team will need to analyze the systems and assess the amount of work that will be needed to build these connectors. Ideally, a business decision maker should intervene at this stage as well to make the call on what needs to be built.  

It is also common to expect some issues with legacy systems at this stage. These need to be able to “talk to” modern systems that facilitate chatbot implementation. One way to make this happen is via middle layers. This is also helpful keeping in mind potentially larger modernization exercises that may and most likely will happen in the future within the organization.  

According to Gartner, by 2022, 20 percent of all new chatbot and virtual assistant implementations will be done on conversational AI middleware that supports multiple NLP back-ends, up from less than 5 percent today. 

6. Adapt the Conversation Design for Transactions 

The most important part of improving a simple FAQ chatbot to one that can tackle more complexity is the conversation design. The quality of the conversation design can make or break chatbots  so it is essential to get this right. Improving the NLP is relatively easy when it comes to transactions. In fact, it may not even need to change much from what was done for the first phase of the chatbot. Updating existing intents to include transactional components might be all that is needed.  

In the conversation design process, the main challenge is to enable the chatbot to ask the right questions of the user and to pull the right information from the back-end correctly. This requires a slight change in mindset to think about the following: 

• Validating user input: How should the chatbot ask questions in order to get valid inputs from the user. 
• Clarifying prompts: If the bot is unable to fully understand the user input, it will need to find a way to respond and request clarification in order to get the necessary information. 

• Handling errors: The user input itself may have some errors and the bot should be able to handle these or clarify in a conversational way. 
• Back-end errors: Even if the user input is understood, the bot could end up producing errors as it pulls the necessary information from the back-end. For example, there could be no valid information to retrieve or the bot could retrieve wrong or irrelevant information. The conversation design should keep these scenarios in mind in order to ensure that the transactions are carried out smoothly. 

These are just some of the nuances to consider in the conversation design for chatbot transactions.  

For most organizations today, automating responses to FAQs have become a low-hanging fruit. It is probably the easiest application of chatbots and can be used as a testing phase before moving on to more important transactions and tasks. Getting there requires a step by step process starting with the business priorities, identifying the most important high-value transactions and assessment of existing knowledge repositories and back-end systems. 

To find out how KeyReply can help you improve or scale a chatbot to the next phase and cover more complex transactions, talk to our experts today.