Seekar Technologies CEO, Kordel France, joins us to share Seekar Tech’s approach to explainable AI. read more
Guy Munro, Head of Innovation at TalkVia, joins us to share his perspectives on voice assistant adoption down under. read more
How to train an ASR (automatic speech recognition) engine with Esteban Gorupicz and Alejandro Heredia, Atexto
Atexto join us to dive under the hood of training automatic speech recognition systems (ASR).
Product Owner, Dialogue Management, AI and ML Group at Swisscom, Roger Dill, and CEO of Artificial Solutions, Per Ottosson, share insights on how to scale a conversational AI practice.
Sam Danby shares insights on conversational AI tooling and Boost.ai. read more
Kane Simms and Dustin Coates are joined by Elaine Lee, Principal Product Designer at Twilio, to discuss the ins and outs of Twilio’s Autopilot bot builder and how you can build trust with users through dialogue design. read more
Why do all skills start with ‘Welcome to xyz’? Is an ‘assistant’ the right mental model for voice experiences? Mark Webster of Adobe XD joins us to tackle some of the biggest challenges in voice and discusses how design can play a role in solving them. read more
Invocable announced today that it’ll be closing its doors, just 5 months after pivoting from Storyline and putting itself behind a pay wall.
VoiceFlow are working with Invocable to offer a migration service for users wanting to port their existing Invocable skills to the VoiceFlow platform.
That’s the news, but the question is:
Why is invocable closing?
If you don’t have time to read the full article, then Vasili summarises it as:
- The market of a tool for creating voice applications relies on the success of voice applications, which is not there yet.
- A voice app works well as an integration — very short, concise request that is correctly recognized and processed. “play the latest album from Eminem” is a good example. But there’s nothing to design here; all these applications are custom integrations, sometimes made by vertical players (like DoorDash skill for ordering).
- There are voice apps that need to be designed, but NLP and NLU quality are not good enough now to support their growth. They’re like IVRs from the 90s, but on Alexa.
There are a couple of other reasons why I’m not that surprised by this news.
What went wrong?
Aside from what Invocable report, there are three other reasons why I think Invocable struggled.
- Premature monetisation
- Target market pivot
- Tree-based design
Firstly, without significant funding, it’s extremely difficult to sustain a startup. Invocable landed a $770k of funding in July 2018 back when it was Storyline, but that’s not a huge amount considering they were powering about 10% of the Alexa skill store at the time.
Its core target market at the time was hobbyists. Bedroom skill builders creating and publishing Alexa skills in Storyline. Some of them, like Kids Court, won competitions and cashed in on developer rewards.
It’s understandable then, that Storyline would want to cash in on their product. If it’s making money for the people using it, then why shouldn’t it make money for the founders?
However, what I suspect the team found is that hobbyists are less likely to pay $60 per month for software. Far less likely than full-time designers and agencies.
So the company pivoted into providing prototyping tools for designers.
The problem I think the team might have found is that there just isn’t enough full-time VUI designers or agencies out there willing to pay that kind of price for that kind of tool. Yet.
Maybe it’s too early.
Target market shift
The second thing that might not have helped, in hindsight, is that pivoting the company might have alienated some of their core customer base.
I have no doubts that the tool was used heavily by designers for prototyping purposes. I was one of them. But I wouldn’t have classed myself as a core user. I didn’t actually use it that much. Just in workshops and here and there for ideation.
I wasn’t someone who published a skill, nurtured a skill, updated it and had success with it.
I’m not sure of the numbers, but I’m sure there would have been a split between people publishing skills through the platform and people logging in regularly, but not hitting publish. Furthermore, I’m sure there would also be a difference between those who publish one-time skills and those who’re creating things so good that they’re getting developer rewards for it.
I’m not saying that things would definitely have turned out differently, and I’m sure the guys knew what they were doing at the time, I just wonder whether they might have picked the wrong group to hone-in on.
The last thing that hinders Invocable, and many other prototyping tools for that matter, is the tree-based nature of its design. It forces you to think about your voice app as a decision tree.
With a voice experience, a user should be able to ask for whatever they want, whenever they want it and have the app respond to them. With a decision tree style design, you’re only ever going to be able to provide the answer that matches your specified next steps.
I’ve written about the Virgin Trains skill and the perils of decision tree style design recently in our 4 ways to take your voice strategy to the next level article.
Tree-based designs are fine for games and interactive stories where the assistant is in control and leading the interaction, but in situations where the user is in the driving seat, tree structures don’t work as well.
I would have thought that this would have been one of the first things Invocable would have fixed after going behind the paywall. But to my knowledge, it didn’t. Nor did it update much else, aside from multi-modal support.
Or am I wrong?
I’m not writing this with any motivation other than trying to understand how the top voice design tool of 2018 has ended up folding. And this is just my thoughts on it. Maybe others can learn from it, or set me straight if I’ve missed something.
I’m more than aware that I could be completely wrong. It’s entirely possible that Vasili and Maksim have figured out the eventual end game for all of the prototyping tools out there. Maybe they’re all destined to become interactive story or game design tools.
Or perhaps the guys are just ahead of the game and they’ve kept the tool backed-up until NLP and NLU advances to the point where they feel it’ll be useful again.
Time will tell.
For now, though, I certainly need a prototyping tool.
This week, Dustin and I catch up with John Kelvie, CEO and founder of Bespoken, and learn all about the three types of testing that can help you create and sustain great voice experiences.
- Unit testing: how to test your code locally without having to deploy into the cloud and test through your smart speaker or phone. This can save developers a whole load of time and effort in the development phase.
- End to end testing: how to automate testing of utterances and intents to make sure you’re returning the correct response to the various utterances that can be fed through your skill or action. This saves the QA folks time as you no longer need to fire up your skill or action and physically test every possible utterance.
- Continuous testing: making sure that your continue to keep on top of the ever-changing AI operating systems and ensuring your skill or action is always operating as intended.
We also discuss the convergence of usability testing and technical testing and how they can play together, as well as hear John’s take on the future of voice.
Where to listen
This week, we’re finding out how content creators can have their podcasts and YouTube content indexed and searchable on voice, with Bryan Colligan of Alpha Voice. read more