In this article, I use the term "Help Center" to refer to all types of Help Center, Knowledge-base and FAQ solutions.
Help Centers are rarely a core part of the conversational AI discussion (or roadmap).
Why is that?
Did Help Centers miss the opportunity to evolve beyond what people associate them with today (i.e: long-form articles, organized in high-level topics, powered by keyword search and ctrl+F)?
Given almost every company invests in building a Help Center and keeping it up to date, it could have been expected to become the launchpad of the Conversational AI journey, not a parallel initiative.
As I discuss here and here, the best chatbot experiences are often those that act like search and provide a simple response, which can be anything that helps the user resolve his/her request - including linking the appropriate Help Center article; Ada (who just raised at a whopping $1.3B valuation!) built their business by developing this type of chatbot experience for their customers.
A great example of a chatbot linking Help Center content (Zendesk in this case) is Kraken's:
If some of the best user experiences are when the chatbot simply understands my request and points me to the Help Center, why doesn't the Help Center simply understand my question in the first place?
The only thing missing for this to happen would be for the Help Center to be powered by natural language understanding (NLU), and for the content to be organized in a way that provides easy consumption across "conversational" channels.
Today's Help Centers do provide chatbot-like widgets, but the resulting user-experience is often disappointing because of the gaps in the knowledge, and poor search results: Help Centers simply aren't trained as chatbots.
It feels like improving this Help Center widget experience could be a first, simple step towards conversational AI, that doesn't require building a completely separate chatbot.
For example, by training and mapping NLU intents (i.e: specific user questions) to existing articles, a Help Center could easily combine a light-weight "conversational" response (in pink below) to the traditional help article result, and provide the same experience as the bot:
Why did NLU become the realm of chatbots and not a core part of the Help Center development?
Why didn't developers hack the Help Center into being more bot-like?
As a developer, I want to say that it's because it's more fun to start over with a new, fresher solution :)
More certainly, I'd argue that it's because the most popular tools for building NLU (like Luis, Watson, DialogFlow, Rasa etc) simply don't make Help Centers a first-class use-case!
To make Help Centers a first-class use-case, these platforms would need to:
- Make it easier to manage the long-tail of user intents (currently, training and maintaining more than a few dozen intents becomes extremely painful - most teams end up managing the data in other tools like Excel)
- Allow users to structure intents hierarchically: this would make it easier to map intents to the Help Center (structured in hierarchies of sections & articles)
NLU for the Help Center would provide a strong foundation for any conversational AI roadmap
The Help Center is often part of a bigger customer support suite that ties communication channels together (live chat, email, ticketing etc): the data captured within these channels is perfect for training AI that understands the long-tail of customer intents.
The process of training NLU from unlabeled data is highly valuable in itself, as it inevitably leads to the discovery of new intents and gaps in the knowledge base
While a chatbot's scope might not be to handle all of the long-tail requests, the Help Center on the other hand should very much aim to address 100% of user queries; by applying NLU to power this type of search capability first, business priority would be given to uncover and map every user intent.
Far from being limited to the Help Center, the resulting long-tail NLU can be easily leveraged to power other parts of the AI roadmap (including "transactional" chatbot flows, AI-powered analytics etc).
Because ultimately, the biggest problem AI still has is understanding even the simplest of user requests :)
HumanFirst productizes the data engineering capabilities necessary for companies to build and improve Natural Language Understanding at scale.