Quite often organizations start with FAQ virtual assistants in trial projects but don’t go forward to tackle more complex transactions and tasks. Here’s how you can take it to the next level and realize its full potential to transform your business
Virtual assistant implementation projects often start with a proof of concept (POC). Generally, these involve FAQ virtual assistants that can automate responses to commonly asked questions via chat interfaces. But so often, such projects don’t go much further and are abandoned due to various reasons including a lack of investment, resourcing and infrastructure required to support, maintain and improve these technologies. To to get the true benefits of virtual assistant 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 virtual assistant development, on the other hand, is more about language model training. That means that building virtual assistants that can carry out transactions are 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 virtual assistant take the next step toward automating transactions? Here are some key steps to follow:
Start with the Business Priorities and Operational Metrics
It is important not to get carried away with the results of “Phase 1” of a virtual assistant 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 virtual assistant 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 per cent 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 virtual assistant project can get fixated on vanity metrics like the technical accuracy of the virtual assistant responses and the number of conversations automated. While these are useful, it is more important to evaluate how the virtual assistant 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 virtual assistants for the insurance industry, it could be a way to improve customer satisfaction, contact centre productivity or support and empowerment of agents.
Identify the Most Important Transactions That Will Impact the Business
Taking the virtual assistant 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 virtual assistant and move closer to meeting their business objectives.
| PRO TIP: Pick 4 to 5 transactions to trial and test. Examine how effectively the virtual assistant helps automate them and the resulting business impact. |
To automate these high-priority transactions, you will have to enable the virtual assitant to pull the necessary information from other systems. This will determine the processes that need to be put in place to design a virtual assistant that can automate end-to-end transactions.
Conduct an Audit of the Content and Knowledge Repositories
The initial implementations of an FAQ virtual assistant 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 to updating, validating and generating answers. 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 virtual assistant project, if these individuals are not involved in the project.
Conduct an Audit of the Applications Involved in Transactions
It is also critical to conduct a similar audit to identify the specific transactions the virtual assistant 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 failure 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.
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 virtual assistant 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 virtual assistant implementation. One way to make this happen is via middle layers. This is also helpful to keep 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 virtual assistant implementations will be done on conversational AI middleware that supports multiple NLP back-ends, up from less than 5 percent today.
Adapt the Conversation Design for Transactions
The most important part of improving a simple FAQ virtual assistant to one that can tackle more complexity is the conversation design. The quality of the conversation design can make or break virtual assistants hence it is essential to get this right. Improving the NLP is relatively easy when it comes to transactions, it may not even need to change much from what was done for the first phase of the virtual assistant. 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 virtual assistant 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 virtual assistant ask questions to get valid inputs from the user?
- Clarifying prompts: If the virtual assistant is unable to fully understand the user input, it will need to find a way to respond and request clarification to get the necessary information.
| For example, when a customer asks, “Where is my refund?”, that can mean several things. The virtual assistant should ask questions to determine if the customer is concerned about a product refund or a tax refund and so to complete the transaction. |
- Handling errors: The user input itself may have some errors and the virtual assistant should be able to handle these conversations.
- Back-end errors: Even if the user input is understood, the virtual assistant 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 virtual assistant could retrieve wrong or irrelevant information. The conversation design should keep these scenarios in mind to ensure that the transactions are carried out smoothly.
These are just some of the nuances to consider in the conversation design for virtual assistant transactions.
For most organizations today, automating responses to FAQs has become low-hanging fruit. It is probably the easiest application of virtual assistants 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 virtual assistant to the next phase and cover more complex transactions, talk to our experts today.
