Speech Is the Next Frontier for Local Search — TalkLocal
April 19, 2016 | Contributed by: Gurpreet Singh
Voice search is really picking up steam. In 2015, 60% of people who use voice search apps like Siri started doing so within the last 12 months according to research by MindMeld. Yet, so much of voice search is still about reading search results. With a promising market showing such strong growth facing such weak disruption activity, it should come as no surprise that the process of creating a truly intuitive voice based user experience is anything but.
Like many innovators, my company is among the many vying to impact the voice search landscape in a big way; but we face challenges that tempered our initial hopes of launching a multi-vertical voice enabled app covering 30+ service categories on various platforms in a little over a year. Now, although we’re proud just to have a single working prototype for a voice request app that helps consumers find plumbers in seconds; we know our full potential is a long way ahead – both as a company and as a tech space.
Here are a few of the most quagmire-prone aspects of building a voice enabled local search platform.
The Quagmire of Language Itself
Designing commands and rules for a voice-enabled AI can be like asking a powerful trickster genie to make you rich. Do it right and get millions of dollars; do it wrong and get turned into millions of calories – a double fudge chocolate cake with buttercream icing now rich in ways you hadn’t intended.
Take the term “bit” for example. It has distinct meanings in computer technology, carpentry, veterinary science, and locksmithing. So, an AI serving all of these industries would need to recognize context to match the consumers and businesses properly.
The broader and more varied the jargon an AI has to manage, the greater the margin of error. So, in order to manage the various paradigms of industry jargon and possible linguistic pitfalls, we had to build individual verticals over time. Still, even within the single plumbing vertical, AIs have a very complex decision to make: “Is this really a plumbing request or not?” The trick is giving the AI its maximum level of discernment using minimal rules.
For example, rather than add backflow, flowing, and flow control valve to the AIs lexicon, we used “flow” to signal a qualifying lead if paired with a similarly approved keyword. Not to say it’s fool proof. We’ve coded against exceptions like “water my flowers,” but there will undoubtedly be more such exceptions discovered once we launch in live beta. Human language, like humanity itself, is full of quirks and surprises.
The Human Element
Factoring for human error is standard practice in user-experience design and it’s especially challenging when pioneering new technologies. Even with our current web and app based services, our customer service reps routinely find and help customers who submitted a request incorrectly. Now, with a service category like plumbing reaching a voice based market through a platform dominated by party games and recreational interactions, even the least gutter-minded reader can easily imagine how incorrect submissions could be as often a matter of cunning as one of error.
It’s been kind of fun and sort of awkward coding against misappropriations of terms like pipe and all manner of double-entendres that users might try to sneak passed the AIs lead qualification rules and into the ears of our poor customer service reps. Of course, there’s no doubt that an especially creative troll will get through, and who knows, maybe we’ll send them a fruit basket.
Here’s what concerns us more than the risk of a troll submitting a request and assaulting the delicate nature of our staffers: it’s the risk of further frustrating or alienating our customers by keeping sincere submissions out. That’s why we had to take the additional step of creating rules distinguishing trolls from users who sincerely need help but are using profanity or humor to manage their stress. And, while we’ve decided against accepting profane or lascivious requests, the AI can respond with humor or empathy as needed while giving the user a chance to rephrase and resubmit their request.
It’s hard to implement a voice search strategy across competing brands like Apple, Google, Amazon Echo, and Microsoft Cortana. Exploring these 3rd party platforms is itself a complex multi-step process with an approval process far more elaborate than what many folks encounter when actually launching a typical mobile app. Afterall, how can anyone feel sufficiently knowledgeable about the voice search landscape when the industry leader (Apple) doesn’t feature an open API.
In a space where mere research is a networking and business development campaign all by itself, voice search innovation is compromising the cost-efficiency of the minimally viable product. Such barriers pump the brakes on innovation and stand in staunch opposition to the start-up guiding principles of test early, fail fast, and pivot when needed.
Still, we were grateful that one of the most start-up friendly brands was Amazon Echo. Alexa features an open API, unlike Siri; and, unlike Cortana and Google, Alexa operates as a distribution partner so we’re not left to merely add a voice feature to our app and rely on our own brand-power to build a user-base. Besides these facts, pioneering smart-home technology with Echo was indeed a perfect fit for a home services app.
However, since Alexa has fewer users than mobile app based Siri, for example, we have to look forward to a longer live beta testing period before we’ll have enough user-data to inform our next iteration. With the app pending approval, we’ll have to decide whether to begin the long road to another platform now or let proof of product gradually reveal itself before investing further resources.
And, I almost forgot to mention: it’s all totally worth it! The paradigm shift of giving voice to artificial intellects, the landmines of unpredictable human nature, and the speed-bumps of third party tech integrations are just the brush you’d expect when blazing a new trail in the local search space. Undoubtedly, startups innovating the voice space are scouting ahead and into the future. And, our collective efforts – no matter how fraught with difficulty – will make the path easier for the next iterations and generations of local search innovators.