How accurate are virtual agents? There's some work to be done
While use of virtual agents is increasing, there is still a good amount of consumer hesitancy surrounding them. We found that
80% of consumers think chatbots (including virtual agents) need to get smarter before they'll use them on a regular basis.
[v] In other words, they need to become more accurate before they're widely adopted.
In addition, our research revealed that the
strongest emotions people feel about using chatbots are frustration (22%) and anger (11%) - not exactly the types of reactions businesses are going for!
[vi]A number of factors contribute to virtual agent inaccuracy, ranging from scope creep to inadequate training to poor design. Let's now look at some things contact centers can do to make their virtual agents smarter and not leave a third of their customers frustrated or angry.
How accurate are virtual agents? - Steps contact centers can take to make them smarter
The goal of an effective
self-service strategy is to make customers as successful as possible. For virtual agents, this means providing quick, accurate responses or facilitating quick, accurate transactions. When virtual agents can provide resolutions without transferring customers to a live agent or trapping them in an endless loop of “please rephrase your question,” that gives customers the frictionless experiences they expect and value.
The following are some ways to improve the accuracy of virtual agents so customers are wildly successful in this self-service channel.
Thoroughly train them
Like new customer service agents, virtual agents need thorough training before interacting with customers. And like human agents, the quality of the experiences they provide is directly linked to the quality of the training. However, the nature of the training is quite different.
Virtual agents are powered by artificial intelligence, and artificial intelligence has a voracious appetite for data. The more information AI tools consume, the more effective and accurate they become.
Actual customer service interactions should be part of the information used to train virtual agents, and the more the better. It's not unusual for new virtual agents to be trained with millions of historical interaction records, such as chat transcripts. Consuming all this data allows virtual agents to identify patterns, such as frequently asked questions and appropriate answers.
Use virtual agents for the right tasks
Virtual agents are best suited for repetitive, well-defined tasks. Let that be your mantra! Even though they're "smart," they don't have the same problem-solving capabilities that humans do. It's important to be clear-eyed about the strengths and weaknesses of virtual agents when deciding what to use them for. Otherwise, you risk putting virtual agents in a situation where they can't be consistently accurate, and this will frustrate your customers.
When determining how and where to use virtual agents, start with top contact drivers. You may need to drill down a level or use
AI analytics if current reporting is too general. Categories like "billing inquiry" don't provide enough detail to be able to determine if virtual agents can handle those types of interactions.
Look for contact types and transactions that follow structured rules with no or very rare exceptions. For example, if opening a new insurance claim always follows the same process, that might be a good task for virtual agents to facilitate accurately.
As an example of one of these structured tasks, one of our clients successfully uses a virtual agent in their IVR to authenticate callers, who typically already have an account with the company. Because authentication is a process that follows set rules, it's a task that the virtual agent can successfully handle and offload from human agents.
Contact centers also need to be practical and calculate the ROI of potential virtual agent tasks. Just because a virtual agent can accurately automate specific processes doesn't mean it makes financial sense. Additionally,
use cases need to be viewed in light of the end-to-end customer journey to ensure inserting a virtual agent doesn't create a disjointed experience.
Give virtual agents access to an accurate, well-populated knowledge base
Virtual agents rely on
knowledge bases to provide accurate answers to customers. They determine the user's intent and then analyze the knowledge base to provide the best answer. In a classic case of "garbage in, garbage out," if the knowledge base isn't accurate, the virtual agent won't be accurate.
The best knowledge management systems can be accessed by all support channels, including both live and virtual agents. This creates a single source of truth and ensures customers will receive consistent answers regardless of where they find help. A consolidated, enterprise knowledge base is also easier to maintain, which increases the likelihood that the content is complete and accurate.
As a side note, a good knowledge management system is a fundamental tool of a strong self-service strategy. Customers can search it online to find answers to their questions, and the contents can be made available to search engines so that customers can see answers on search results pages.
So, you can see why the question "How accurate are virtual agents?" doesn't have a clear-cut answer. No two virtual agents are the same and their accuracy rates can vary greatly. However, accuracy should be a top concern for any business that wants to implement virtual agents because it directly impacts customer adoption and satisfaction.
Contact centers should strive to make their virtual agents as accurate as possible, and the preceding tips are a good place to start. Without accuracy, organizations will never hit their ROI goals.
To learn more about NICE's AI self-service capabilities, please
visit our conversational AI and chatbots product page.
[i] Gartner:
Gartner Predicts Chatbots Will Become a Primary Customer Service Channel Within Five Years (2022)[ii] Gartner:
Gartner Predicts Conversational AI Will Reduce Contact Center Agent Labor Costs by $80 Billion in 2026 (2022)
[iii]NICE:
2020 Customer Experience (CX) Benchmark, Consumer Wave (2020)
[iv] Marketing Artificial Intelligence Institute:
4 Conversational AI Metrics: How to Measure AI Chatbot Performance (2020)
[v] NICE:
2020 Customer Experience (CX) Benchmark, Consumer Wave (2020)
[vi] NICE:
2020 Customer Experience (CX) Benchmark, Consumer Wave (2020)
