If there’s one thing we hope you’ve taken away from all of the fresh insights on our blog, it’s that the key to unlocking truly intelligent AI is proper training. That applies to any type of artificial intelligence. It’s especially true in our domain, where companies are grappling with how best to deploy AI-powered virtual agents to improve their customer support.
Ongoing training is crucial to AI development — just as fundamentally important as education is to human development. Even though much of the training for your AI will happen after your virtual agent launches, you’ll want to account for it at the beginning of your AI project because if your AI is going to be successful, training your AI has to be a part of the project plan. Arguably, it’s the most important part of the plan. To extract real ROI from your project — if you want your customer satisfaction scores and your support metrics to improve — you can’t just write code, launch a virtual agent, and consider it done.
The true value of your AI project is dependent upon ongoing training, and proper planning for that begins with the business case.
The business case for AI-powered virtual agents in customer support
As we wrote in our recent report with Contact Babel (“The Inner Circle Guide to AI, Chatbots & Machine Learning), most analysts project that within five years AI-powered virtual agents (or chatbots) will be able to resolve 50-80 percent of customer service inquiries — without human agent involvement. There are three primary ways companies have started deploying AI into customer service:
First, AI is guiding live agents, suggesting the next best actions and surfacing related information for the agents during the course of live chat interactions.
Second, AI is used to automate the routing of customers to the right content, experts, and solutions.
Finally, AI is powering conversational virtual agents to engage directly with customers, automating more and more of Tier 1 support without human agent involvement.
Unfortunately, many companies are struggling with early AI and automation attempts. Gartner estimates that through 2022, 85 percent of AI projects will “deliver erroneous outcomes due to bias in data, algorithms or the teams responsible for managing them.” We’ve surveyed companies and have identified four key challenges: lack of content, lack of good feedback, maintenance (and lack thereof), and lack of empathy (which we have detailed in “Four intelligence gaps that’ll kill your AI project”):
In many cases, AI failures are caused by a lack of planning for the period after an AI system launches. Because no AI-powered virtual agent is going to be highly successful on Day One. It takes continuous training plays to fill in those intelligence gaps. That includes regularly creating new content, offering feedback, monitoring and maintaining performance — and providing human input that can improve AI’s ability to empathize. Training is what will transform raw code into an effective, intelligent virtual agent. It’s what will dictate the difference between a failed AI project and one that delivers value to a company. In other words, ROI.
An example of a bot without training
For an example of how intelligent (or not) an AI-powered virtual agent is without training, let’s look at Alexa, which is exactly what we did in this recent post, “If humans need 10,000 hours of training to be an expert, how much does AI need?” You can also explore this problem by trying to set up a chatbot with Lex. Lex is Amazon’s web service that uses the same natural language processing engine as Alexa.
In short, Amazon makes years of AI research and development available to you to quickly create your own chatbots. If you were to try the “Pizza Ordering” bot as a template, you’ll quickly see that AI “out of the box” isn’t very smart at all. It will reinforce the point that having a plan to train AI is a crucial part of any project.
Three questions to help unlock value from AI with training:
1. Who is training the AI? Are domain experts directly involved in training the AI?
There are many different approaches to training AI, as we wrote in “It Turns Out, AI Isn’t All That Intelligent Without Us.” Some AI learns from monitoring online human behavior, just as Amazon might suggest purchases based on what you bought in the past. Some companies will train AI using internal team members. Other companies will hire “clickworkers” — people who get paid to train AI on everything from image recognition and text analysis to audio recordings.
Directly takes AI training a step further. Our AI-powered support automation platform enables our clients to tap the expertise of their most passionate and experienced users on the products and services they sell.
2. How do you make expert knowledge available?
By tapping expert users to identify gaps in your self-service content and solutions, they can create relevant, rich, and timely support content to improve your AI performance. In our recent post, “’It Takes a Village’ to Create Truly Intelligent AI at Scale,” we explain how we help companies engage with their users to train AI in a way that scales — which internal teams with finite resources cannot.
The end result is that expert-trained virtual agents benefit from thousands of hours of combined expert experience. And through this scalable expert training, AI becomes more intelligent over time.
3. What happens when AI and self-service can’t help the customer? How do you plan to resolve questions that AI can’t, in the same customer conversation?
We believe that a well-trained, expert-trained virtual agent as the first point of contact has the potential to automate 80 percent or more of customer support issues. But AI alone can’t solve every issue. Your virtual agent should be part of a cohesive multi-channel customer experience strategy. Businesses need to define the escalation path so that there’s a seamless handoff from AI to human support when necessary.
What does training look like in practice?
To fully understand the importance of training AI, you might want to see it in action. We’re happy to provide a demo of an expert-trained virtual agent — and show you what our expert-led AI training process looks like.
Regardless of your path to support automation, we encourage you to have a plan to train your AI before you build it. The success of your AI project depends on it.