Artificial Intelligence 101 for Customer Support Managers
The best customer support managers are always looking for the latest tools and best practices to superpower their support teams. If you’ve been online in the last year, you’ve probably heard about Artificial Intelligence and wondered how it will impact your work.
Artificial Intelligence is the ability of a computer to simulate human qualities such as the ability to reason, discover meaning, generalize, or learn from past experience. Harnessing this power is crucial to customer support teams who want to provide excellent, personalized support to customers at scale.
In this article, we’ll give you a great starting point for researching how AI can benefit your organization and walk through the various types of AI tools available today.
Defining Artificial Intelligence
Before we jump into how customer support teams can utilize AI, there’s a few basic definitions that will help you understand the tools available. Here’s three methods that our AI at Aida uses to help understand customer conversations: machine learning, unsupervised learning and natural language processing.
Machine Learning is exactly what it sounds like — a computer learning. But machine learning doesn’t happen by sitting a computer down in a classroom and showing them flashcards until their memory is all filled up. Machine learning is a unique method of giving computers the ability to learn on their own — much like how a toddler learns by looking at the world around them. Although while toddlers can pick up things quickly, computers need lots and lots of data to learn from. The process of machine learning is often referred to as “training” the AI.
Deep learning is a subset of machine learning that uses neural networks to learn. Synthetic neural networks are based on the efficient way our brain makes connections as it learns. Machine deep learning requires a lot of data, especially labelled data. Labelled data might be a picture labelled as a cat, a customer conversation labelled with the product type or a sentence labelled with it’s meaning.
Unsupervised Learning is a specific type of machine learning that uses unlabeled, unsorted data. AI programmed to use unsupervised learning can look at a whole mess of data and pick out patterns that they weren’t explicitly told to find. It’s an incredibly powerful technique because it doesn’t depend on the human brain to point the AI at something. AI makes it’s own unbiased conclusions, strictly on the data. Plus, most of the data available in the world is unlabelled. Unsupervised deep learning methods can take advantage of this data to make connections.
Natural Language Processing (NLP) is the science of making computers understand how humans talk. Humans are such messy creatures, with very nuanced languages but years of talking to each other has made us very good at understanding speech and text. Traditional conversational interfaces, like chatbots, tend to fail because they have a narrow understanding of human language. Many chatbots only have basic NLP capabilities that understand language granularly, looking for keywords that are programmed into them. Advanced NLP provides AI a way to interact more broadly with the world around them. For example, an AI with advanced NLP ability will understand you would like a coffee whether you say “Can I please have a cup of coffee?” or “A grande dark roast.”
Having a basic understanding of these AI methodologies will help you better evaluate which AI software your team can benefit from.
AI, not chatbots
When you start researching AI, you’ll start hearing a lot about chatbots. But the two technologies are not the same.
Chatbots are rule-based programs. They are great at handling very narrow, specific interactions where customers are prompted with potential questions or next steps. For example, China Merchant Bank handles almost 2 million customer conversations each day through a WeChat bot. The bot can deliver bank account balances, transfer money and pay bills.
But if you push chatbots outside of pre-programmed interactions, introducing unanticipated questions or edge cases, they create a frustrating user experience. Chatbots only recognize phrases that have been programmed in. For example, a customer service team might set up their chatbot to reply to “waybill number” with a link for the customer to track their package. But what if the customer asks about “estimated delivery date”? Or “tracking ID”? All of these will need to be programmed in separately, because the chatbot doesn’t actually understand the question — it just recognizes the phrase.
As you can see, these programmed responses can be very brittle, so teams will need to spend time and money maintaining them — just like old school IVR systems.
AI offers a huge advantage over chatbots: the ability to learn. If you’re investing in AI, you want to make sure you’re getting true AI and not just a clever call-and-response bot. The best AI can even tell when a question is too complex for itself to answer and smoothly pass it off to a human for processing.
When sourcing an AI vendor, ask about the type of machine learning their technology utilizes. Their team should be able to explain how the AI breaks down and analyzes conversational data, and they should talk about NLP. Do they require sorted or unsorted data? If you need to provide structured examples of questions and answers, it’s likely a chatbot. If new information is added, do you need to update the program? AI learns as it goes, but chatbots will require the new information to be input. How much configuration is necessary? Chatbots require consulting and programming, AI just needs the data to program itself.
With AI being such a hot field right now, there’s a lot of companies claiming to offer AI-like services. In reality, many of these services are glorified chatbots and may not help improve your customer’s experience the way you expect.
How Support teams use AI
There’s three types of AI on the market for customer support teams today. Each one targets a different area of the customer experience to help customer facing teams be more productive and effective.
AI in Self Service
Self service is the ability for customers to help themselves. This could be in the form of a knowledge base article or a tool that helps them update information. Customers love being able to self serve. Finding an answer yourself takes much less effort than submitting a ticket, waiting for an agent to respond, clarifying your questions and waiting for another answer. It’s even worse if you actually have to pick up the phone to call! Making it difficult to get help is the leading cause of disloyalty. The CEB found that customers were far more likely to churn when they experienced a high effort experience. Companies are taking notice. IBM predicts that by 2020, 85% of all customer interactions will be handled without a human agent (IBM).
AI for self service helps unearth the right information for customers at the right time. A customer opens up a chat window or a contact us form, enters their question and clicks submit. The AI program checks to see if they have a good answer and suggests it to the customer before creating a ticket. If the customer finds what they need, there’s no need to get a human involved. If not, no worries, a ticket is opened and a human is available to help.
Self service AI can look a lot like a chatbot, but it goes deeper than that. Remember NLP? Being able to understand your customers, no matter how they ask, is a big part of AI for self service. Aida uses machine learning to only suggest answers to simple questions, where self service is appropriate. If it determines that the customer will need to talk to a human, like when the question is very complicated, Aida skips straight to opening a ticket. Using a self service AI further reduces the effort customers put into getting help — meaning your customers are happier and more loyal.
AI Assisted Agents
For companies that don’t see a lot of frequently asked questions or that require more technical in-depth conversations, a front line AI might not be the best option for them. The questions wouldn’t be simple enough for an AI to answer, or perhaps each question is too unique to be answered through historical data. Trying to implement self serve answers that were never helpful would just annoy customers. Instead, these companies should look to implement AIs on the agent side to help increase response efficiency.
AI assisted agent technology has been compared to cyborgs. Human agents handle the conversation, but they are assisted by an AI that can help locate the right macro, suggest next steps, analyze past behaviour, draw attention to things humans might miss. While the AI will still suggest an answer, humans are able to approve and adjust the message before sending it. Some programs will also take care of the tagging and filing for you, so human agents can focus on creative problem solving.
More efficient agents reduces the cost of contact, and gives teams more time to provide better support. LivePerson, a customer service platform provider, reported their agents saw up to a 35 percent efficiency gain with “AI-assisted human agent” mode. It’s all about putting the computer to work and getting the busy work out of the way of the humans.
The third type of AI technology customer support teams are using is for data analysis. There’s so much data locked up in the conversations support teams have with customers. But this data is rarely well organized and even less frequently analyzed. If it is analyzed, it’s usually through the eyes of humans looking to prove a point.
Think about the steps you’d need to go through to unlock meaningful insights from your support conversations manually. Someone would have to tag the tickets, export the ticket data, and live in Excel Hell for a couple weeks cross referencing churn figures with tags, reply times and customer satisfaction scores. Plus, the tags might not even be applied correctly in the first place! Instead of spending time tagging and analyzing an AI can spot the trends that you didn’t even know to look for.
Ben MacAskill, VP of Operations at SmugMug, has recently invested in an AI insights tool that crunches conversations to surface problematic trends. “At a certain size,” Ben explains, “human insight stops being capable of scaling. If you have 30 agents, each person only sees about 1/30th of the conversations and that’s simply not enough. Turning to AI to analyze all of your conversations helps you surface trends to understand the whole picture.”
Ben says that when they turned the analysis over to an unbiased, all seeing AI, suddenly certain trends became crystal clear. “Many of these issues were never brought up by our excellent support team. They simply couldn’t see the forest from where they sat.”
Find more time for what matters
All of these different AI tools have the same goal. They help customer support teams be more efficient and effective at the work they already do. It’s all about maximizing your available resources and delegating repetitive, computational tasks to the computers.