Our early virtual assistants are kind of janky. That’s okay!

It’s Not A Two Horse Race For Voice Assistants And Chatbots Aren’t Dying

Three predictions on virtual assistants in a week full of predictions — from folks like us who actually build them.

As voice assistants dominated CES last week, people are asking “is every toaster, microwave and car really going to talk to me now?” Or are virtual assistants just a fad since many are being shutdown or proclaimed dead?

It’s hard to tell from the news this week. Having spent the last 1.5 years building virtual assistants (and 20+ years building consumer tech), here’s the bets I’d make:

Prediction #1: Google and Amazon will be in a lot of devices, but neither will “own” the voice entry point to customers.

That’s because the data and brand interactions created by conversations with customers will be so valuable that companies will not want to share this data with potential future competitors. Companies might experiment with Alexa and Google, but they are not going to totally give up on building their own voice platforms as Ina Fried at Axios observed this week at CES:

It’s a fierce battle between Amazon and Google to get their assistants included on other companies’ devices. At the same time, hardware makers including Samsung, LG and Roku are (also) putting their own voice assistants into their products.

As these companies watch their users interact conversationally with their hardware, they may find that the verboseness required to be a “smart speaker” like Alexa that can answer everything may not be required to enable useful conversational navigation of, say, a microwave. A smart speaker needs to be an omniscient service that can answer everything for everyone. Often a few basic user intents like ”Yes”, ”No” and some base navigational phrases can help the user accomplish most tasks — especially if the device has a screen that can convey information as well.

I’m completely biased on this idea, but I believe companies will start turning to dedicated (plug!) conversational developers to build holistic voice and chatbot features for their businesses outside of just embedding inside the big tech NLU platforms like Alexa or Google. To be clear, I believe these proprietary services will have interoperability with the major NLUs, but will be something more than “just an Alexa skill” in the near future.

Also, the companies that are experimenting with their own services now will be way ahead of the curve when their customers expect having a personalized conversation with a brand as a primary feature. Experimentation while the market is still growing and the bar to wow the end user is low is important.

That’s because even though Google, Amazon and Apple have a huge ASR lead on most speech recognition services, the most practical assistants do not need to understand a massive vocabulary to accomplish most tasks. Again, that’s because a single domain application can manage to handle the recognition of limited entity names (cocktails, movie names, etc) within a reasonable amount of time. It can be a bit rough at first, but better to work out these issues now while the number of total users is small. If you were late to the web or mobile, don’t blow it this time by waiting to find out how to present your brand in the conversational internet. Do it now while the stakes are low.

So expect even more voice and chat platforms outside of Amazon, Google and Facebook to exist and thrive in the marketplace over the next few years.

Prediction #2: The conversational internet will expand and interoperability between conversational platforms will accelerate as consumers demand consistent, state-aware conversational relationships with their favorite brands across platforms.

Even if I’m wrong and there are not dozens of successful conversational platforms and only 3–5 conversational platforms dominate, the consumer will demand that the relationship with their favorite brands transfer state to whatever platform is convenient for them. Think streaming a movie on Netflix on multiple devices for conversations.

For example, I may start talking to Alexa about a recipe in my kitchen in the morning, but I may want to pick back up the conversation on a Slackbot while at work to confirm I want to make that recipe. Then I might want to pull it up on my phone when I’m shopping for ingredients later. If I have to start the conversation over again or the AI doesn’t remember what we last spoke about — even if that’s on another device or platform — that user will be irritated with the brand and think it is dumb.

There’s no reason for the consumer to ever feel like a brand is dumb just because it can’t remember what the last interaction was on a different platform. Platform lock-in will not work on the conversational internet, much like it didn’t on the web or mobile. Brands and their consumers will force openness onto these conversational platforms. The platforms that try to keep brands locked in to their platform will ultimately fade while the open ones expand…as usually happens. Can you imagine how pervasive Siri would be if it was launched as an open platform? MBA’s will be writing cases about that missed opportunity for years.

The thing that is really missing for conversational services to explode is an open standards body that will enable developers and companies to build interoperability for the conversational services. More on that in another post.

Prediction #3: Conversational assistants and chatbots are not overhyped or dying.

Sure, conversational assistants and chatbots have been kludgy (or downright offensive) in this first wave, but that doesn’t mean they are “dead” or dying. It means they are evolving. Yep, this is evolution and we’re seeing the extinction of things that aren’t quite right or fully baked.

It has a shell and hair! Everything for eveyone! Glyptodons (above) did not make it, but armadillos did.

Every major platform change starts with weird experiments or just plain bad ideas. 98% of startup products deserve to be evolutionary fodder. Another 1.5% were genius ideas that were too early. The other .5% that survive become monster businesses.

In Wired’s reporting on the “death” of Facebook’s M assistant, they lopped in “and so are chatbots” in the headline. In the vicious hype cycle of new technology, it wouldn’t be a cycle if chatbots and voice apps didn’t suffer a bit of a blowback after last few years of exuberance for those technologies. But to say that virtual assistants and chatbots are dead because the first wave of these applications are a bit wonky would be short-sighted.

The Wired article by Erin Griffith and Tom Simonite actually lays out the real issues with not only M, but Siri and the whole first wave of all-in-one “Pangea” assistants:

M's core problem: Facebook put no bounds on what M could be asked to do. Alexa has proven adept at handling a narrower range of questions, many tied to facts, or Amazon's core strength in shopping.

Another challenge: When M could complete tasks, users asked for progressively harder tasks. A fully automated M would have to do things far beyond the capabilities of existing machine learning technology. Today's best algorithms are a long way from being able to really understand all the nuances of natural language.

These two paragraphs completely wrap up both the promise and the problem with assistants. As users, we so want them to work! The user immediately goes to superuser mode with conversational assistants -whether chat or voice-based and ask them to solve all kinds of problems out of the scope of what’s currently possible. Inevitably, the user then curses the assistant when it fails and claims “this is stupid”.

I remember the feeling the first few times I tried to get a modem to connect to the internet or using my first cell phone. You gotta look past a lot of fail to see the future.

So while the Pangea “all things for everyone” assistant phase ends, I believe we will move to a “continental drift” phase of assistants where smaller assistants will breakout and tackle complex domain-specific problems successfully for end users. There are already quite a few productivity and work-related chatbots that are effectively solving problems for customers. As more companies have their assistants focus on domain-specific or single purpose assistants we will see more consumers asking “why can’t I do that for (X problem)?” in their lives. Once this happens, I really believe every website, app, device and brand had better be conversational — or start the slow fade to oblivion.


A lot of these predictions came from the work we are doing at Pylon ai. Our first two beta products, Tasted and The Bartender, are popular voice apps on both Alexa and Google. Our apps are built to work across voice and text platforms, so they also work on FB Messenger and Slack. Our apps are also “multimodal”, which means you can use them with a screen when it’s easier than talking to them. You can see a video of how that works here. If you would like your own cross-platform, multimodal assistant for your business, please email me.

And… if you made it this far, we owe you some schwag or a Google Mini! If you’re interested, send us your address! Or if stuff is not your thing, please sign up for our newsletter here for updates on conversational assistants, Elixir, React and other stuff we talk about at Pylon. Thanks!