Do Your Chatbots and Voice Assistants Have the Testing Support They Need to Succeed?
It’s hard to imagine where today’s businesses would be without conversational AI. This technology, which powers both chatbots and conversational IVR systems, proved essential for navigating a changing service economy through a global pandemic.
Even before COVID-19, Gartner predicted that 70% of white-collar workers would interact with conversational AI platforms every day by 2022. The market for this technology is now expected to grow at a compound annual growth rate (CAGR) of 21.8%, reaching $18.4 billion by 2026.
This is thanks, in no small part, to how much this technology has improved in recent years. Chatbots, in particular, can now support the customer experience in many ways, enabling more customer self-service and reducing the demand on human agents.
Nonetheless, success is not a given when contact centers deploy chatbots and other conversational AI solutions. A chatbot comes with powerful AI capabilities, but it still hasn’t been tailored to fit your needs or tested in your business. Before contact centers take the plunge, they must consider what it really takes to ensure their conversational AI solutions will support and enhance the customer experience.
The growing demand for chatbots in the contact center
In large part, contact center executives don’t need to be convinced that they should adopt conversational AI in the form of either chatbots or intelligent voice assistants. Most are overly eager to bring these solutions into the mix. According to Canam Research, 78% of contact centers planned to deploy AI by 2023, with the largest portion (55%) pointing to chatbots as their primary AI solution. The CAGR for chatbots is expected to grow even faster than conversational AI in general, at 30.29% from 2022–2027.
There are good reasons for this, too. Across the board, contact center executives see the fruits of deploying chatbot solutions. A recent survey of Fast Company Executive Board members noted that adding a chatbot solution to their website enhanced customer engagement, accelerated service, enhanced personalized support, and increased customer satisfaction — just to name a few outcomes.
These positive results are encouraging, but that doesn’t mean chatbots and other conversational AI technologies are now flawless. They still fall short in many ways, from misinterpreted customer intents to delayed handoffs and security failures. And the resulting poor customer experiences can lead to customer churn and other negative impacts on a brand. These possibilities should make any contact center executive pause before jumping on the chatbot bandwagon unprepared.
The chatbot testing conundrum
That’s not to say contact center leaders shouldn’t embrace this technology — only that they should do it in the right way. As responsive and smart as AI is, it’s still limited by its programming. Ultimately, chatbot misfires still occur because bots can’t possibly account for all potential human interactions. The nuances and quirks of human communication are so vast and varied that there’s no way to prepare a chatbot for all possibilities out of the box.
Consider, for instance, how many possible ways someone could ask a chatbot to order a vegetarian pizza. They may ask for a “veggie pizza,” a “pizza with no meat,” a “meatless pizza,” or use one of any number of other phrases. On top of that, any given person might bring their own quirks, like spelling errors, colloquial ways of saying something, limited tech capabilities — you name it. How do you know if your bot is capable of handling all these variations and nuances? You need to test it.
But truly testing for all these and the many other options for how someone could order pizza is an extensive job. Doing it manually would require many hours, or possibly even days, first to come up with the types of tests to run and then to run them. To do it efficiently, you need a solution that can accomplish all the necessary steps for you — a testing platform that allows you to quickly and efficiently expose these limitations so you can send the bot back to development and teach it new skills.
AI testing AI: the true path to flawless CX
Fundamentally, this kind of testing must cover the entire process so your testers don’t have to test your chatbots manually or spend hours developing test cases.
It means testing from end to end with automated natural language processing (NLP) score testing, conversational flow testing, security testing, performance testing, and chatbot monitoring. Ideally, the testing process should be simple and intuitive, with no coding, scripting, or programming involved.
Let’s return to the veggie pizza example. It would take a person (or a team of people) an incredibly long time to come up with all the ways someone could order their veggie pizza; and even then, they’d probably miss some. The only way to effectively come up with all possibilities would be to leverage AI to generate the test data. AI could select a question, such as “Can I have a vegetarian pizza,” and then automatically generate a list of ways to say the same thing. It could then automatically test the chatbot with those variations to see how it responds.
Going a step further, how many different ways could a person actually say each of those variations? AI can be used to further drill into the unique human quirks that different customers might bring to an interaction. For instance, AI could add layers to testing for customers who type sloppily, type in all caps, misuse homophones, add extra spaces or emojis, and more. “Pizza with no meat” could then become “pizza with no meet,” “PIZZA NO MEAT,” and any number of other possibilities.
These are just examples, but what’s important is that your testers don’t have to come up with all these options or run the tests themselves. You need a testing solution that will do it for them, with minimal manual effort. What you want is, effectively, AI testing AI so you can run these kinds of comprehensive, detailed tests much more quickly and frequently. This allows your testers to expose more chatbot weaknesses so your developers can teach and improve your bots more often and with greater precision, ultimately providing a better-quality experience for your chatbot using customers.
Contact center executives’ instincts are right: Investing in chatbots is a smart move. But doing so without adequate testing support could lead to more harm than good. Cyara Botium does exactly what we have described here and can provide the testing support your contact center’s chatbot technology needs. Learn more and try a demo to see for yourself.
Artificial Intelligence, Machine Learning