Unless you’re a human being, understanding natural language and holding a conversation is considerably difficult. This is one reason why chatbots often fail. The second one is that users want actual help. You need to look into systems to provide the status of an order, a current balance, take a payment, process an order, unlock an account, etc.
Simply speaking, chatbots fail because they don’t understand the user, cannot engage in dialogue and ultimately are not able to lead the conversation to the point of resolution. Below are some examples that point to the struggles of chatbots during live customer interactions:
1.) They do not understand you (paraphrased example):
2.) They do not understand context.
3.) They do not help you (paraphrased example):
(There’s a good amount of discussion and examples of this topic on the Internet, and in various blogs such as this one here.)
With such poor contextual and natural language understanding, as the examples above illustrate, chatbots are fired (discontinued/retired) quite a lot. One of the most high-profile episodes was Microsoft’s Tay. LATAM had Julia online for only one month in 2013. United’s Alex – live from 2009 to 2014 – was described as a search engine with a face. As a result of their technological limitations, they failed to address customer needs in a fast and accurate way.
However, there’s great potential
The good news is that you can provide great customer experience with a best-in-class virtual agent. And you have a great starting advantage of speed, 24×7 availability and instant scalability which virtual agents provide.
According to Forrester’s North American Technographics Customer Experience Survey, over 70% of U.S. online consumers say that “valuing my time” is the most important part of customer service. (See. here and here.)
If you get it right, you can:
- Provide much faster and better service with your virtual agent.
- Offer additional – even completely new – services and superior availability – e.g. 24×7 support in more languages.
- Free up the time of your live agents to work on other value-added tasks.
How to build a great Virtual Agent
What does a great virtual agent need?
- Great use cases, ones that make sense for a virtual agent and the selected channels.
- A superior user experience and/or user interface.
- The best technology you can get – one that is able to hold conversations flexibly with the best Natural Language Processing capabilities.
- Live agents for escalation — both to handle complex cases and even more so to observe humans and to enable your virtual agent to “learn” from your best agents.
Great use cases
Start with identifying your existing business challenges:
- Which can a virtual agent solve? Start with the most frequent ones. Keep away from the long-tail ones.
- If that includes FAQ answers, great! Don’t be surprised if it doesn’t.
- Expect to integrate with/use your IT systems to help users. That’s what your best human agents do, don’t they?
- And the best part: Your virtual agent can query multiple systems at once at superhuman speed!
- Define an initial scope and roadmap. Don’t boil the ocean.
Superior user experience and/or user interface
A key part of delivering superior customer service is how this service being delivered to a customer. The interface for a virtual agent-customer interaction needs to feel intuitive, use natural language, and be easy to navigate.
Channels also matter. There are certain ones such as online chat, phone and mobile applications that are more popular with certain audiences. Adding functional capabilities, such as editing a previous answer within a chat, makes all the difference from a CX standpoint. You want a platform where you train your agent once and expose it on any channel you need.
The foundation of a great virtual agent is superior technology. You need the best Natural Language Processing you can get. The virtual agent needs to be emotionally aware and able to extract the context from customer statements. Also, during a live interaction, a customer may ask a question while a virtual agent is resolving a separate issue, and a virtual agent can switch context at a moment’s notice and resolve that additional issue.
It also is absolutely vital that a virtual agent be able to tap into an organization’s IT or back office systems to complete business processes. To gain a full understanding of a customer’s account, for example, a virtual agent has to have access to the appropriate systems in order to locate and extract that information.
Nobody knows everything, not even the best possible virtual agent. You need live agents to handle complex and rare cases. And at the same time this means your virtual agent can “listen in” during escalations and get smarter that way.
There is a remapping of the virtual agent’s BPN (business process network) that occurs after this shadowing experience, but the virtual agent cannot employ these new learnings in live conversations until they’ve been approved internally—hence the term supervised machine learning.
Let me show you how this can look like:
What you see outlines:
- A great use case: This is Amelia, our virtual agent, acting as a trusted advisor selling car insurance.
- An enriched user interface/user experience – more than a chat.
- Best-in-class Natural Language Processing – with context switches as needed.
- API integration to serve the need of the user.
And it gets better. This screen shot is not just representing a demo. Amelia is live-selling car insurance on the Internet and does a better job than the competing web form.
With customer expectations at an all-time high, chatbots and their limited capabilities can no longer meet the demands of users who are trying to be helped as quickly as possible, without encountering errors or receiving incorrect information. To be successful, virtual assistants need to sound and interact like humans. Factor in high volumes of customer requests, and it’s clear that virtual agents, not chatbots, can deliver better customer service.