Voice + Chatbots = Next Wave?
Anyone who’s had an existential conversation with Siri or had Google direct them to the nearest Chipotle, has interacted with one of these “omnibots”, that communicate via voice and text.
Voice-based digital assistants have recently become a staple within many homes. Equipped with the most basic AI, voices, and personalities, they can help users manage scheduling and handle simple tasks. In fact, IDC says smart speaker sales are going to rise from $4.4 B to $17.4 B by 2022.
Right now, giants dominating this space include Amazon’s Alexa and Google Assistant, with players like Apple's Siri, Microsoft’s Cortana and Samsung’s Bixby playing catch-up to capture a significant piece of the pie. This leaves some to wonder: are voice-bots the next big wave?
While there are notable successes, voice-based bots may or may not be the best fit depending on your business goals.
Things to Consider when Deciding to Build a Voice Chatbot
As you consider your bot structure and strategy, here are some things to consider:
1. Can you mirror a human-like “voice”?
Often, the measure of success for voice-based bots is how closely they mirror human conversation. The internet went wild this month when Google debuted their new “chatty” smartphones that can call and make appointments, in a voice that is virtually indistinguishable from a human. While this appears to be quite advanced (and we’ll believe it when we test it ourselves!) It’s important to pay attention to small details such as: conversational flow, to make as much effort as possible to build bots that mirror human conversation. Do you have the skills to create an effective conversational UI?
2. Do you have time to train your bot?
Artificial Intelligence (AI) and Natural Language Processing (NLP) are still in relatively nascent stages, and anyone who tells you that they have a great out-of-the-box NLP platform that doesn’t require training, will probably also try to sell you some magic beans. The fact is, text-based bots can rely more heavily on decision tree based structures, but voice-based bots, while able to use a couple verbal prompts, by their very nature allow users to say anything to them. Hence, if you want to prevent your bot from suffering major #fails like the dreaded, “I’m sorry! I don’t know what you’re saying” response, then you’ll need your bot to be highly-trained. Effective training can take anywhere from 2-5 months, so recognizing whether you have the resources to do this is key.
3. Is the solution to you users' needs optimized by using voice?
Will your platform (website, mobile app, etc) be conducive to host a voice-activated bot? And how do you users typically communicate with your bots? Voice-based bots are best used when the atmosphere is quiet and a person feels comfortable using their voice commands. If, for example, you have an iOT product that can be used in an office space-you may not want employees yelling over each other to use the bot. Also, in crowded spaces, this becomes confusing to use. If there is no indication that text conversations are inconvenient or cumbersome for users to efficiently communicate and have their questions answers, what would prompt the need to develop a voice component? Or, in other words, why fix what isn’t broken? At different bot conferences, we have spoken to multiple large brands who took advantage of Amazon’s offer to build out skills to augment their text based-content or services. Many marketing managers found, while this was nifty to say, they put extensive time and energy into it, for very little additional value-add. In other words, does a voice-based component create additional value (an interesting article from NPR on the topic)? Or does it simply replicate what you already have in text-form?
4. What platform should you build for: Amazon, Google, or Emerging Tech?
Unfortunately, you cannot build a voice-bot for all platforms, and the race to the winner of this category is still yet undetermined. If you go to any Bot/AI conference and ask people who will win this war you'll find that, at this point, the results are mixed. In Q1 of 2017, Amazon owned 79% of the home assistant market. By Q1 of 2018, Google was 36% and Amazon had dropped down to 27%, and the market is being inched out by competitors from abroad Alibaba and Xiaomi. (See more details here.) In a quickly evolving market, it’s challenging to know where to put your resources, and thoughtfulness about where the market is heading, as well as what your potential users will be purchasing, is tough. Moreover, because these devices are constantly listening, the data collected by these companies is incredibly valuable; one wonders if governments are not incentivized to help companies win this race just by owning the data inside of users homes both domestically and abroad—which suggests that this will be a tight race indeed.
via VoiceBot.ai/Canalys
5. Do you have a plan for the data collected?
Due to the nature of how these platforms work they collect an immense amount of data. The questions you should ask yourself are: how are you going to use this data, what goals will you achieve once you have this data, and moreover, what privacy methods will you put in place? For example, Amazon recently got into hot water because an Amazon Echo recorded a family’s conversation and sent it to a random person in their contacts. (Which goes back to Consideration #2: NLP has a ways to go…) For your first foray, you may want to put bumpers on the bowling alley to ensure that the data you are gathering has safeguards. With GDPR in full effect and serious backlash that has faced Facebook and Amazon, it would behoove you to be think strategically about what data you will gather and how you will use it.
6. Do you have the full scope of technological resources and budget?
Voice-bots have many nuances, not to be easily-dismissed. If you don’t have any experiences in bots, it’s helpful to get your feet wet and explore bots in general, before spending large sums of money with digital agencies. Start with a platform that allows you to easily create and launch bots at an affordable cost and experiment with NLP services like Wit.ai or Google Dialogflow—which can be set up quickly in a day. Gather research and learnings for a couple weeks, so that when you go to build out a more robust bot, you have the understanding and knowledge to make good decisions.
For more learnings on voice-based bots, check out some of our favorite resources: