What Makes A Great Customer Service Chatbot?
Done well, AI-powered chatbots have huge potential to disrupt customer service in target areas like ticket deflection. But what separates great chatbots from the rest? First, beginning with a deep understanding of the customer and second, an algorithm capable of handling real-world “messiness”.
It’s no secret that customer care can make or break a brand, but it’s getting harder and harder to “wow” customers. Forrester’s 2018 Customer Experience Index (CX) for US Brands reported that there are no clear brand winners, and overall, CX is trending downward.
Depressing for consumers, but a huge opportunity for proactive companies. What’s the key to customer service excellence? It may seem counterintuitive, but there’s a case for full-scale digitalization. If done right, McKinsey reported that digitalization leads to higher customer satisfaction while lowering costs. And in a big way – customer satisfaction jumped almost 20% when switching from traditional channels (phone, email, etc.) to digital.
Customers want what they’ve always wanted – easy access to high-quality information. But now they expect to have their pick of medium – whether this means customer forums and FAQs, trouble ticket systems, or phone calls. But don’t think you’re off the hook for providing a quality personalized experience.
A great place to begin your journey into digital customer care is ticket deflection through well designed chatbots. Today’s AI-powered chatbots are light-years ahead of the clunky bots of the past, and companies are pouring resources into AI for customer care. What separates well-done chatbots from the rest? First, deep understanding of customer behavior, and second, flexible algorithms that can handle messy real-world interactions.
You must understand your customers to make a great chatbot, and a great chatbot can help you better understand customers.
It’s the virtuous cycle of AI. Well-functioning chatbots reward the user by providing the right information quickly, which drives repeat traffic to the AI, which trains it better, in turn making the system more accurate.
Of course, you’ll need some amount of customer data in order to train the initial system, but with modern NLP, you’ll need much less data to arrive at a system that works well. Some customer service functions, like ticket deflection, are a perfect fit for an pilot AI-chatbot project because they are rich in customer data already. The same questions pop up again and again in different configurations, which helps the system understand context and scope of queries, and provides data for advanced sentiment analysis, etc.
Ultimately, your goal is to better serve your customer. So you’ll need to deploy your chatbot on an application for which you know (generally) what will be asked, and how to help. From there, your AI solution should perform while always learning from experience – and as a bonus, deliver insights you wouldn’t otherwise discover.
Self-teaching chatbots make life easier for your customers and your team
Many enterprise AI applications rely on supervised learning. This is like learning to ride a bike with training wheels. The algorithm ingests training data with the answers already known – “training” is the process of mapping input to output until the machine produces correct outcomes regularly. If it spits out “wrong” answers, a data scientist steps in and tweaks the algorithm. This can take a long time.
Unsupervised learning is more complicated. The algorithm doesn’t receive any answers upfront – it must learn on its own, guided by the strength and flexibility of its programming.
This is important in digital customer service because of the variety in customer requests. In some cases it may be possible to make a conversational flowchart which captures most possible questions. For example, your cable provider likely has a flowchart for billing questions or service requests. But this method isn’t scalable – imagine what a conversational flowchart would look like for TurboTax. The work would be endless, ROI would never appear, and you’d always have a frustrated customer out there. It’s impossible for your team to code for each and every query. But luckily, Natural Language Search (with unsupervised learning capabilities) solves the problem.
Assuming your customer will create a help ticket when they have an issue, NLS can analyze the intent and sentiment of the question as it’s typed, and return possible solutions before a ticket is created. Because it’s not keyword-based, question variation doesn’t matter – the AI can “understand” the intent of the user and return actionable knowledge, not just a document dump. NLS with unsupervised learning will continue to add to its own knowledge base with each query, so your team isn’t playing catch-up, anticipating and training for new customer issues. This approach means less work for everyone, happier customers, and immediate ROI.
The chatbots of the future can do so much more than just chat. They’re self-taught subject matter experts who understand sentiment and intent to provide real service to your customers. They never get frustrated or give up on a difficult customer, and they learn from every interaction.
If you’re thinking about implementing an AI-powered customer care chatbot, remember: going in with a deep understanding of your customers and their behavior is key, and your AI solution must be chosen carefully; NLP-based AI with unsupervised learning capabilities will deliver on ROI and customer satisfaction if it’s capable of unsupervised learning.
If you’re interested in learning more about how AI-based customer service can take ticket deflection from 10% to 60% or more, read our case study here.