Over the past decade plus, chatbots have dominated the conversation (no pun intended) when it comes to digital engagement. You’ve undoubtedly had experiences interacting with them, some helpful while others underwhelming, and perhaps even fiddled around with building one on your own. That’s because the chatbot promise is grand: businesses and their customers can speak to each other 1:1, at scale. And it’s this attractive potential that has led us to an abundantly crowded space — chatbots are everywhere. There were 300,000 individual bots deployed this past year, and that’s just for Facebook alone.
But alarmingly, amidst such rapid growth the hype is dwindling. Why? To be completely frank — most chatbots today suck :/
They support only limited basic questions, try to obtain our emails instead of actually fulfilling our requests and fail (often miserably) to understand us as users. But ultimately, the underlying reason is that people underestimate just how challenging it would be to create an open conversational interface.
If you’re in charge of developing your organization’s digital roadmap, the chatbot space is likely important to you. As you think about how to implement conversational AI in your business and assess its impact across your industry, be it healthcare, travel, retail and beyond, there are a few factors to consider during your due diligence.
If “x”, reply “y”
Let’s rewind a bit and use websites as an example. Today, just about anybody can build their own website with user-friendly platforms like Wix or Squarespace — basically, you don’t need to be a developer to craft a visually stunning and/or highly functional site. Well, a similar phenomenon of “DIY” (Do-It-Yourself) has emerged in the world of chatbots, boasting the promise of building and deploying a conversational agent for a fraction of what is typically a large investment. But there’s one key (and potentially obvious) difference — chatbots are not websites.
Bots are conversational, not graphical. And therefore, some of the rules of DIY simply cannot carry over.
For a conversation to seem real, it has to feel organic. Bad conversational bots (voice or chat) are the notorious punchline of many jokes, but what makes for a funny commercial is also a convenient illustration of what happens when a bot is constructed through a template. If a conversational agent is meant to solve a business problem, then conversation has to be more open-ended.
A controlled experience in which users select from a limited set of options isn’t actually a conversation, it’s a sequence, which makes chatbots not really chatty at all.
According to the rules of a visual-only world, design can be templatized. But if a user is in charge (as they are in a conversation), those templates are too restrictive, lead to bad or even embarrassing conversation, and might turn the user off of chatbots for good.
Maintenance is key
In order for a bot to become continuously smarter overtime, the data has to be constantly updated. With a small team or limited resources, it can be difficult to maintain the flow of information needed to successfully launch and scale a proper chatbot experience in-house.
Businesses themselves are also inherently dynamic, and information about a company doesn’t stay ripe for stretched periods of time — phone numbers and addresses change, employees come and go, and APIs are added and subtracted often. A visual experience (i.e. website) needs meticulous maintenance, as does a quality conversational experience.
Unfortunately, today chatbots are completely separated from other content management systems (CMS) deployed by the company which creates a cumbersome maintenance process and often leads to accidently providing outdated information to customers.
Your own data is not good enough
If a bot is meant to understand natural language — real, human conversation — then its data diet poses another deep challenge.
The majority of today’s solutions in the chatbot space are built from limited, intent-based flows and heavily relies on machine learning (ML) techniques — which means that “teaching” a chatbot requires tons of training data and thousands of examples per intent. So essentially, the AI, and by extension the resulting conversational experience, is only as healthy as the data you feed it.
There are countless well-known examples of AI “learning” and becoming more advanced, and if you’re Google, you can afford to take that to another level. But through a strictly machine learning approach that relies heavily (if not solely) on digesting information, a bot’s ability to carry out a successful interaction is unfortunately limited to its own universe of data.
Power in expansion
Since the data being fed to a bot should never be stale, building your own invites constraints in quantity and quality of information, with data narrowly confined to that of your own customer base. More so, upon a successful deployment of a chatbot in a specific use case — you may rightfully decide to support other use cases as well, which might require completely different natural language understanding (i.e. scheduling an appointment vs. resolving IT issues).
You may also want to go omni-channel and deploy a conversational AI agent on platforms that don’t have a graphical user interface such as call centers or smart speakers (Alexa, Google Home, etc.). It’s essential to think long term and consider the possibility of adding other use cases and platforms in the future.
When evaluating chatbot or voice assistant solutions, identify the potential walls you’ll run into when expanding conversational AI as part of your digital strategy.
At Airbud, we’re sporting a new school of natural language understanding (NLU) that renders the classic “if X, then Y” as obsolete. Instead of building predefined conversational flows and guessing user behavior before it happens — we use a knowledge graph capable of aggregating content collected from your own digital channels and beyond, all in real-time. Our zero maintenance and code-free solution makes the conversations easily scalable, so when your content updates — your conversations update. Finally, using computational linguistics methods allows us to easily add omni-channel support and use cases that grow with your business needs.