“How can I help you?” Millions of users of conversational artificial intelligence encounter this greeting with hopes for a wide scope of automation to follow. “Wow, it can answer anything? Let’s test its limits.” It’s never long before they find those limits. An interface designed to help a policyholder file a claim or a viewer change a TV subscription isn’t necessarily trained to help choose a pizza topping. Too many users find themselves trapped in frustrating loops or interacting with dumbed-down experiences such as simple FAQ navigations.
These offerings lack minimum viable scope (MVS), undermining the potential of artificial intelligence (AI)-driven self-service to free users from the need to call contact centers or navigate clunky web pages and mobile apps. The conversational AI paradigm started with businesses running the equivalent of high-school science projects, testing different platforms in a race to showcase innovation before they could offer their audiences real utility.
There’s good reason for hope, however. Shining stars have risen by deploying well-architected, integrated, immersive experiences. They knew better than to simply upload swaths of training data and hit the “go” button. The recipes for quality conversational AI deployments increasingly include the same ingredients, and enterprises can now carve out competitive advantages according to how well they conversationalize user experiences.
Not All Use Cases Are Created Equal
Many “science projects” started by automating a nearly random set of topics that reflected a business’s own web services more than the needs of its customers and prospects. Use cases matter. It’s important to set clear benchmarks for success, such as “cutting contact center costs by $3 million in the next year” or “elevating the brand score by 20 points.” From there, data-driven decisions can be made about which use cases will produce the highest dividends. A good place to start may be automating those topics that account for the highest volume in existing communication channels, increasing the likelihood that users will navigate more than one topic and perceive MVS.
Design With Empathy For The User
What frustrates users today? The exercise of defining three or four personas — giving them names and demographics so they are believable and mapping out what each of them might say, think, feel or do within your existing communication channels — can uncover pain points that conversational AI can solve. A user persona of a single mother named Sylvia who, at 2 a.m., has a newborn baby in one arm and a smartphone in the other, may suggest a design that requires limited typing and more of a guided journey with option buttons, list pickers and carousels to navigate with the tap of a finger. It may combine a friendly tone of voice with messages that get right to the point.
Don’t Obsess Over Natural Language Understanding
Too many conversational AI deployments go overboard chasing the intelligent elements of the technology, and end up focusing their design solely on natural language understanding (NLU). While this intent-driven approach can free a user to quickly leapfrog to the point in an automated use case that may best serve them, it can also prove limiting. Pairing NLU with guided journeys, like interactive step-by-steps for technical troubleshooting, can graduate experiences beyond the single question-and-answer approach we find with Siri or Alexa. It’s also wise to set expectations upfront: Not “How can I help you?” but “I can help you with XYZ, how would you like to proceed?”
Transact Or Bust
Transactional capabilities distinguish “stupid bots,” like many seen on social media platforms, from those that lead to actual resolution. Airline passengers will get far more value out of an interface that enables them to book, change or cancel flights than one that merely tells them about aircraft amenities. API integrations and robotic process automation can open up these powerful uses by pulling data from the back end to the front end. Authentication is another integration that can transform a conversational experience via personalization, such as pulling a user’s purchase history to recommend new products.
Yes, There’s Still A Place For Humans
Few deployments will automate 100% of user interactions. Positioning a seamless escalation from automation to a human agent can ensure full coverage of all possible conversational topics. When first prioritizing use cases, businesses may examine a Pareto chart of common inquiries with a long tail of curveball topics that won’t make sense to automate. Training agents to quickly review the context of a user’s interaction with AI and pick up where the automation left off enables businesses to deploy to a wide audience with MVS.
Incentivize Usage In The Right Places
Businesses have spent untold millions to deploy AI, but their wait-and-see approach to engagement resulted in low usage, hurting their return on investment. Piggybacking conversational AI atop a high-traffic landing page is rather different than burying an interface 11 clicks away. Avoid the temptation to blindly go where users are, such as Facebook or Twitter. Deploy through secure platforms that do not sell conversational data to third-party ad buyers.
Conversational Data Is Gold
By A/B testing user journeys on a small audience, companies can create actionable insights that will optimize the journey to a full rollout. Visualization techniques like heat maps can identify user drop-off points in conversational flows and lead to broader business process re-engineering efforts. A subtle addition of content can make a massive difference in engagement. For example, when asked to select dining options for an upcoming cruise, customers may respond more favorably to the upscale choice once they learn about the chef and the wine pairings.
With the right planning, conversational AI can delight users as it creates business value. That’s why it will increasingly replace websites, mobile apps and call centers as businesses learn to deploy it strategically. Users get to vote by way of the data their interactions produce. It’s this feedback loop that will tell businesses whether the design and utility of their conversational AI reflects a mere science project or a UX transformation.
Originally published in Forbes