Customer service is expensive for companies and often poor experiences for customers. Artificial intelligence can help mitigate some of these problems, but implementing successful automation at your company is no easy task, especially if you lack strong technical capabilities. Below are the key technologies and best practices employed by enterprises leading in customer experience automation.
Key Technologies For Implementing Virtual Assistants (VA)
A 2016 Stratecast analysis pinpointed five key technologies that provide the foundation for VAs:
Machine learning (ML) is a set of technologies that enable a computer to develop new behaviors without explicit programing. Programs can be built to take in data and adapt responses based on evolving inputs.
Artificial intelligence and artificial general intelligence (AI/AGI)
Artificial intelligence builds on machine learning by enabling machines to exhibit intelligent, human-like behavior. Customer service departments thrive on human-to-human connections, but they no longer produce sufficient revenue per interaction. An intelligent program that can interact like a human, but without a salary, is the ultimate cost-effective and scalable means to deliver personal customer service affordably.
Natural language processing (NLP)
NLP is the ability of a computer to interpret or understand human language, both typed and using voiced speech, and take appropriate action. Typically, computers are not programmed to think or speak like a human, resulting in a fundamental disconnect when the two are required to communicate. Furthermore, customers and businesses also communicate differently. NLP attempts to bridge these gaps.
Real-time analysis of customer inputs and responses occurs via stream processing of data in primary data storage, or random access memory (RAM), before the data is sent to secondary data storage. Though this technology is promising in terms of increasing efficiency and customer experience, it is currently in a development stage.
Web services and user profiles
Web services support interoperable machine-to-machine interaction. This enables systems to develop user profiles by accessing customer demographic and usage basics as well as a variety of contextual, social, and behavioral data points. Ultimately, this data could be processed by a machine intelligence to offer the consumer custom recommendations for products and services.
AI Implementation Strategy
The implementation of AI requires some forethought and should be undertaken with the following considerations in mind.
Start with a simple chatbot before upgrading to a VA
The data obtained from the bot allows the business to develop a VA with greater efficiency and that is better attuned to customer needs. It is of the pilot and smart small approach.
Always keep the customer in mind
Knowing where and how to best deploy AI requires an understanding of the customers’ needs. It also requires a knowledge of the customers’ language and how it is distinct from the language of the business. Meet the customer on their own terms. For instance, when surveyed, more than half (62%) of respondents prefer bots to use a casual or friendly tone, while only 21% preferred the bots to use a formal tone.
Have the appropriate back-end infrastructure
Launching self-service tools may not be complex, but they will only be effective if they are faultlessly integrated into existing customer relationship management systems. Customers need to be able to flow seamlessly from VA to chat to live agent to avoid frustration.
Tie-in transactional systems
Eventual integration of self-serve assistance with transactional systems like billing yields greater overall benefit for the customer and gets the most value from the VA.
Know when escalation is required
Rules must be established as to when a VA escalates a customer to a live agent. Overly complex queries, sensitive information, or even the detection of frustration on the part of the customer should all be signals that human input is required. It must also be clear to the user whether they are speaking to a human or a bot.
Allow the customer to self-identify across channels
Find a way to identify a customer, either through an email address or ID number, to which the customer can refer during subsequent contact. This eliminates one of the major customer frustrations and streamlines the customer service process.
Pitfalls To Avoid When Introducing AI
The introduction of AI to the customer service department of a business is proven effective, but it is important to initialize correctly to limit the risk of roll-out problems and resultant customer loss. Avoiding common pitfalls ensures a seamless transition to self-service.
Incomplete knowledge base
Providing VAs and chatbots with an incomplete knowledge base occurs when the business expects the AI to do too much, too quickly. Launching self-serve solutions in business areas for which a strong knowledge foundation is available is the best way to begin implementation. There should also be continuous upgrades as the knowledge repository increases. Avoid using different knowledge bases across different service channels to prevent giving customers conflicting information.
One-size-fits-all escalation strategy
A one-size-fits-all escalation strategy is also a mistake as differing types of businesses and customer demands may necessitate their own set of rules. Take the time to establish appropriate points of transfer between bot and human. Sensitive or embarrassing information may be better served on different platforms. An escalation of an issue on Twitter is very different than a one over email.
Lack of access to customer accounts and history
There is no greater frustration to a customer than being required to speak to multiple bots, agents, or departments to resolve an issue. When rolling out an AI-enhanced customer service department, ensure that the VAs have the same access to customer accounts and histories as the live agents. Furthermore, ensure the proper APIs are connected for the bots to perform the most frequently asked questions (e.g. bill pay or check order status).
Not continuously improving AI
Failing to continue to train and update VAs is another potential pitfall. Much like a human employee may require retraining to keep pace with business developments, AIs must be updated as the business grows, requirements change, or policies shift.
Not knowing customer preferences
Bear in mind when a customer may prefer to speak with an agent instead of a bot. Studies show that not all tasks are created equal given a consumer’s comfort level in trusting a chatbot. A recent Survata Bots 2016 survey offered the following results, some of which are counter-intuitive:
- 52.4% of respondents would trust a text bot to assist in recommending a restaurant
- 12.4% would trust a text bot to assist in finding and applying for a credit card
- 9.6% of respondents would trust a text bot to assist with finding a match for a date
- 4.9% would trust it to diagnose an illness
- 16.5% would trust a bot to count their ballot for president
- 5.1% would trust a bot to give legal advice
- 6.2% would trust a bot recommend a financial plan
- 30.5% wouldn’t trust a text bot to assist with anything
These findings highlight that the most important factor when considering an upgrade to automation is to know one’s customers.