A recent report showed the use of voice-activated devices has increased 129% in 2017. One-fifth of smartphone users now talk to a voice assistant at least once a month. Users are also increasingly adopting chatbots as a means for interacting with brands. Millennials, that all-powerful consumer segment, are especially willing to interact with bots first. Sixty-seven percent of them are even willing to purchase from chatbots.
While customers are increasingly comfortable talking to machines, many brands still aren’t sure how to get started with conversational AI (or what the best practices are). Conversational AI presents huge potential to improve efficiency and drive intelligence across all channels, especially for marketing and customer service teams. However, as with any innovation, potential is lost when this technology fails. Many users, 30% by some estimates, are disappointed with their bot engagements.
Often these failed conversations stem from flaws in a brand’s strategy for conversational AI: it’s not just about automation. Until now, this technology has been built to make processes move faster and at a greater scale. Luckily we’re starting to see a shift to using tools to uncover business value beyond deflection and cost savings. Speed shouldn’t be the only goal: ultimately this technology should be scaling our understanding as well as our service delivery, leading to better intelligence and ultimately customer satisfaction across the business.
What is Conversational AI
At its core, conversational AI is a system for managing data input and output. Your customer, the user, inputs data through a user interface (voice, chat, etc). Or, your customer may take an action like click on an offer that prompts the system to engage. Either way, the conversational AI system is incorporating multiple tools and technologies designed to understand and correctly act on user inputs in real time.
The system then needs to do two things simultaneously. First, it needs to understand the user’s intent in order to provide accurate information to your customer, or take the appropriate action. The way it does this has to match the user’s expectations of being accurate, personalized, and fast. Second, it needs to capture and analyze the data from the interaction. This machine learning step is both supervised and unsupervised to train the system while providing actionable insights. These insights can then be used to improve the system’s overall design and ultimately add value across the business.
The Building Blocks of Conversation Intelligence
Once we understand what a good conversational AI system should do, we can break down its implementation into logical steps and fine-tune our AI strategies. This was very difficult until recently– the technologies that comprise conversational AI systems have largely existed within a “black box” that provided a one-size-fits-all solution that may or may not actually fit business needs, let alone provide the ability to scale enhancement of the system over time. Now, open, distributed models of AI and ML tools are available. These open models give businesses the ability to analyze existing data while gathering new data to deploy exactly what’s needed, where and when it’s needed, in order to to build their own customized systems or enhance systems already in production. In short, we now have the tools to deploy conversational AI in a way that fits individual business needs.
When dissecting the black box, our focus should be leveraging the tools in a way that allows us to be strategic and anticipate what our customers want and what they need. Listening and understanding is the system’s goal, as that’s what will ultimately establish trust and earn your customer’s loyalty.
How to Listen to and Understand your Customers
The most effective way to start automating listening and understanding with conversational AI systems is to leverage technologies and ML tools that are capable of analyzing vast quantities of unstructured data. This data could include call transcripts, web chats, emails, unstructured texts, social media posts, and even conversations with existing bots. Using these technologies, you can identify the key purpose or motivation of your customers and see how they are engaging with the conversational system. You can also crowdsource the analysis of the user experience offered by your bot or assistant and gather insights about how and why your bot may have misunderstood a particular user’s intent and if the misunderstanding has any relevance to your business.
For example, let’s say you capture 200,000 conversations. Your system should indicate areas of greater or lesser engagement. It should also automatically surface gaps in your system’s understanding as it relates to your business goals. Ideally these tools would allow you to identify unstructured data that can be dragged and dropped into a database of identified business intents so you can test and deploy language understanding in real-time.
Key to all of this is understanding the customer’s intent – what they really care about. Accurately discovering intent for analysis requires the right mix of data and user input. This mix includes context: the particular conversation, situation, historical information, and the business all factor in when we’re trying to understand what a customer needs.
This level of analysis and insight will truly reveal how well your conversations are driving value for customers and business. Prior to leveraging conversational AI tools, many companies rely on perception and a combination of web analytics, call and chat routing, CX surveys, and audits to paint a “good enough” picture. Some measurement is better than none, but ultimately these methods are too easily influenced by interpretation bias, making them terribly imprecise: how can you infer what a user is really looking for from a page view, keyword search or high-level routing label?
Listening for Intent
AI enables us to push and pull conversational data to match user intent with the right outcomes and opportunities. This moves the needle from listening to understanding to action.
Every conversational AI system requires labeled data to create an intent model. To create labeled data, start by analyzing your customer interactions. Figure out where your customers are communicating with you, and look for opportunities for bots or Natural Language Understanding engines to plug in and create actionable, structured data in the process. The second step is to match your data to an existing intent model. One developed across industries is best: the richer the data, the better.
With these two components you have the building blocks: context and battle-tested data. Now we can train a machine learning system and implement an AI platform whose architecture can drive and grow integrated context, conversation state, and content.
It’s also important to choose a platform that supports the system at runtime and can be customized and extended across all your business for today and tomorrow as your system grows. It should utilize numerous integrations and NLU techniques to guarantee precision and accuracy. This combination of intelligence and experience tools will allow you to deliver a great customer experience no matter what channel a customer selects. At the same time, it will allow you to continuously understand more and more about your customers’ desires for your business.
For example, Dell, one of the largest technology companies in the world, recently improved their business performance and customer satisfaction by introducing an IVA on Dell.com to work with their live chat channel in supporting customer self-service. In just three months chat instances for both live chat and the IVA contact rate increased by 61%. At the same time, live chat costs decreased by 27%. The IVA was able to assist customers satisfactorily and close conversations at a greater rate than live agents alone while also reliably driving revenue from chat with product recommendations.
The IVA is not only able to effectively assist customers in navigating the ever-changing site and instantly support questions, but also accurately recognizes customer intent so well that it is able to compare ever-changing products by understanding language about new products and technology before customers do, in order to advise on purchasing decisions based on individual customer needs.
The End of Assumptions
Implemented in this way, conversational AI will be a new paradigm for your customers. It will most likely eliminate your website’s search bar and other time-consuming, often-inaccurate methods of getting information. It could go so far as to eliminate current websites altogether. In the Dell example, customers no longer had to learn to navigate the site, wait in a chat queue, or scroll through endless product options. With conversational AI, customers can get what they need in a way that feels most natural: a simple conversation.
There are no shortcuts for creating satisfying AI: the example above required iteration and analysis along the way, but the results for both the business and customers is worth it. On the business side, in addition to increasing customer satisfaction, the clarity and accuracy these solutions provide has long-reaching implications for understanding your customers and driving meaningful conversations. For example, many companies currently rely on KPIs like “time to resolution.” This can introduce bias and impatience into our customer interactions: in an effort to determine what a customer needs, we often make assumptions in the absence of information or experience.
Well-designed bots and IVAs aren’t built on assumptions. They operate on a scale and pace a human cannot. These technologies have the potential to eliminate assumptions, improve accuracy, increase satisfaction and drive intelligence across the business. By using them effectively we transition from a reactive to a proactive approach to customer service. If we can do this strategically, we leverage conversational AI not for automation alone, but to do what all successful business people do: consistently listen and take action.