Tech is the Best Way

Alexa is inconsistent with it's skills in India

It has been a year since Alexa came to India. Thanks to a community of 40,000 developers who have developed over 19,000 voice skills for Alexa, it can now do more than answer what is the factorial of sixty (product of an integer and all integers below it). Voice skills are to Alexa what apps are to Android or iOS.

“Schools struggle to find English teachers, which is why we decided to build this English Teacher skill for pre-primary and primary children,” says Vishnu Saran, founder of Voice Qube, a company that makes voice skills. Saran is looking to do a beta launch of this skill by the end of this month at the Gokavaram Govt School in East Godavari district, Andhra Pradesh.

However, Alexa’s most popular use cases continue to be rather basic. Games, music, and questions about the weather are mostly what people currently use the device for.

In the long run, though, it could prove useful for skills that make it worth the Amazon’s while. Skills like shopping, making payments, or discovering new products. This is part of Amazon’s Star Trek vision of Alexa being able to take voice commands for just about everything.

Bezos' pillars

Jeff Bezos called Alexa the fourth pillar for the company after e-commerce, its cloud business Amazon Web Services (AWS) and the subscription service Prime
But right now, what stands between Amazon and its vision is the kind of skills being built for Alexa. As one of the partners working with Amazon said, “Most developers are approaching voice skill building as a hobby. They are solving their own problems rather than creating a user base.”

As a result, the skills in the store have no consistency. And they can be a bit tricky to navigate. For instance, “Alexa, I want to play Kaun Banega Crorepati (KBC),” has no response, but “Open KBC” launches the popular quiz game.

Users also need to remember how to invoke different skills, and the experience is not seamless. A flight booking task via online travel aggregator Goibibo, for example, has to be enabled over the Alexa app on the phone. It can’t be enabled directly through Echo, Amazon’s smart speaker. Not just that, it also requires you to have the Goibibo app, separately. The inconsistency in experience has resulted in patchy usage of Alexa. And that’s a pain point for Amazon.

Voice skills on Alexa are expected to give Amazon a 360-degree visibility of its users and get a piece of all the economic activities in a home. But right now, Alexa India’s focus is simply to get developers creating skills on Alexa. For winning their mindshare is half the job done, something Amazon learnt from Amazon Web Services (AWS). When it launched AWS nearly seven years ago, every single developer interested in cloud was trained on it. Due to the availability of talent, companies automatically picked AWS as their cloud provider.

With Google, Apple and Microsoft snapping at Amazon’s heels in the voice assistant race, Amazon is making it as easy as possible for developers to build skills. When Apple launched the App Store 10 years ago, overnight, developers got access to millions of customers with their credit cards linked to the iTunes store. Even one-off app developers became part of the improbable success run of mobile apps. Voice skill is different from an app but Amazon is making sure even one-off skills get shelf space. What gives?

The complexity of building for voice

In the last 10 years, smartphone users have developed an intuitive understanding of how to navigate an app. Pinch the screen to zoom out, expand to zoom in; when to scroll, swipe or double tap. Developers did this by giving visual cues to guide users. But in the world of voice, none of that exists. A voice user interface, as developers call it, is all about the ability to have a flowing conversation. And also identifying the many different ways in which people can say the same thing.

The simple task of ordering food can be said in so many different ways.

I want South Indian lunch.

Find me South Indian options for lunch.

Give me some quick lunch options?

Recommend the closest lunch options.

Ashish Jha, a 20-year-old computer science student and developer, says that before he gets down to creating a skill, he writes down five-six pages of how a conversation can go and then writes code. Alexa lets developers code in any programming language they want—from JavaScript to Python.

In fact, Amazon has done all the heavy lifting here. It built the AI engine so that Alexa can translate the incoming speech to text. It made Alexa smart enough to understand the context through natural language understanding capabilities. It then created APIs—application programming interfaces; these allow two software programs to communicate—for different use cases. For smart homes, for weather, for games, for commerce. Now it is up to the developers to scoop up those APIs and build skills using it.

Alexa, by providing the blueprint, made it a breeze for developers to create skills. On an average, for a developer, creating and launching an app for something like recipes could take around six to seven weeks, while launching the same recipe content as a voice skill would take half the time, says Sreeraman Thiagarajan of Agrahyah Technologies, an agency that develops voice skills and the content for it. In fact, what takes time in creating a voice skill is building the voice user interface.

An app can lay out all the options available and leave the decision-making to the user. In a voice interface, Alexa has to make the decision. It only says “here is the top option”. And if it’s not the right one, you could be in for a long day.

Voice assistants also run the risk of being annoying. “The threshold for frustration is very high with a machine,” says Dilip RS, country manager for Alexa Skills India. It cannot risk asking too many questions nor can it afford to provide too little information.

But these complexities are not holding back developers, even though the rewards are not very apparent. Why?

Develop first, find takers later

Jha was among the first to develop for Alexa when it reached Indian shores. “Developing apps is very subjective. There are rules one must follow. And in the end, what I may design the user may not even like. But with voice, since it is all about conversation, I can design a skill that meets everyone’s taste. I love this platform,” he gushes. Jha has built about 75 skills, 60 of which are live on the Skills store.

Right now, Alexa only rewards developers with t-shirts and Echo devices that cost between Rs 5000 to Rs 15,000. There is no way for the developers to monetise voice skills now. Alexa would have irate users if it tried to play ads in the middle of a task. In the US, Amazon sends cheques ($5,000 to $10,000) to developers based on the engagement seen on the skills. It also now allows in-skill purchases much like in-app purchases. But it’s not a predictable revenue stream, unlike apps.

For now, the hip-factor of a voice-enabled future is what draws developers in.

While Alexa has successfully captured the developer community’s imagination, there is also a lot of junk swirling around in the skills store. Most skills are games or quizzes, which people would use on a one-off basis and forget about. For instance, Jha built a quiz game for Alexa based on the TV show The Walking Dead. In the first 24 hours, it got 1,000 users, and for the next 10 months, just 1,500.

Similarly, Anil Atyam, founder of VoyceFirst, which builds enterprise and consumer applications, released a KBC game on Alexa. “We expected it to go viral but all we got was 500 invocations over a period of seven months. If it were over an app, we would have easily gotten, 40,000-50,000 installations a year,” says Atyam, whose company now mostly makes skills for enterprises.

The low traction was because enough people didn’t discover the skill, as it is hard for users to know what skills sit inside the skills store. Just like the KBC voice skill’s keyword problem.

Not just that, a lot of skills on the store are repetitive. For instance, there are 4,000 fact-based skills. Of them, 128 skills give out facts about space, and another 109 skills are just facts about cats, according to the Skills store. While the problem of plenty also applies to apps, here, Alexa picks out the skill for you. “And Alexa doesn’t pull up the same skill each time,” says Atyam.

“So of the 19,000 skills, at best, only a few hundred are actually useful,” he adds.

Amazon did not share its numbers on the engagement of skills across different categories but said the kids category sees high traction. As a function, most of the skills getting built are for kids.

Amazon, meanwhile, is trying to nudge developers. It rolls out monthly themes like financial services, cricket, football to encourage developers into building skills in those areas. But it’s just gentle nudging. “We want to be a marketplace for skills,” says Dilip of Alexa Skills India. “Even if a skill is going to be used only by one person, we are not going to stop that from getting built because that skill is important for that user. And it is this selection that would get people to come back to it for more,” he adds.

But users will come back to it for more only if they find value.

“Alexa, you can’t do it all”
One of the best experiences on Alexa is shopping over Amazon.

“Alexa I want to buy towels on Amazon.”

“Amazon’s choice for towels is Amazon brand Solimo towels costing Rs 599 for a set of two towels. Would you like to add it to your cart?”


And voila! It’s on the Amazon cart on your phone. (You still can’t make a transaction over Alexa in India as it is constrained by India’s two-factor authentication norms.)

But this doesn’t apply to all use cases and skills.

Anshuman Bapna, the chief product officer of Goibibo, part of the MakeMyTrip group, launched the Goibibo skill on Alexa last year. The expectation was that it would serve as an acquisition channel for new customers. Through the website and mobile apps, the MakeMyTrip platform—MakeMyTrip, Goibibo and Redbus— has served about 33 million customers.

“All of us at-large e-commerce companies feel that there is a glass ceiling at about 60-70 million transacting users. Once you hit that mark, growth isn’t as fast as we would like–although online businesses have clearly established that pricing strategy, distribution models and convenience ensure there are more new converts,” says Bapna.

He believes that the paradigm of searching where people apply filters for a purchase while adding things to their cart is still a foreign concept. It is only shown by an evolved e-commerce user. But, he says, voice interactions are part of daily life–Indians are doing it already using chat apps. Moreover, over the past year, he shares, India has reached an inflexion point for voice-based searches. 28% of all Google’s searches in India are voice-based searches, and it has seen a 270% year-on-year growth.

With that belief, Goibibo launched an Alexa skill for looking up flights and hotels last year. But it didn’t go as well as planned.

Sample an interaction.

“Alexa, open Goibibo.”

“Welcome to Goibibo, discover flight fares, bus fares and hotel bookings. What would you like to know?”

“Show me flight options to Chennai.”

“When do you wish to travel?”

“Nov 2.”

“Which city are you leaving from?”


For the next 20 seconds, it regurgitates the details.

“The best economy fare from Bangalore to Chennai on 2 November is for Rs 1479, and this is provided by SpiceJet. It departs at 12 o’ five and takes one hour, zero minutes to reach the destination. Should I send details to the GoIbibo mobile app?”

“Alexa, find me other options.”

“Sorry, I was not able to to do that.”

“Voice is not a medium on which you can communicate complex or multiple pieces of information,” says Bapna. And there’s no booking a flight without time and price deliberations. What Alexa, he says, is good for is for figuring flight timings, if it’s on schedule; but that’s about it. “We saw only about a few hundred interactions in a week, while on web and mobile app we see a few million interactions per day,” he says. As a result, they have not invested further to promote it.

While Alexa comes out badly in this interaction, there are areas where it does well.

It is efficient with repeat actions, like booking a cab to ‘Work’ from ‘Home’. Of course, this is only when destinations are saved.

Right now, having a voice skill out is also about looking good. Case in point, IndusInd Bank. “When Amazon approached us, it didn’t take us long to agree to partner,” says Ritesh Saxena, head – savings, digital & payments for the bank. “If we didn’t accept the offer to have a banking skill, they would have gone with some other bank.”

So the bank launched an IndusAssist skill that allows customers to check bank balance, transfer money and pay bills. It follows the RBI’s mandatory two-factor authentication rule for this, says Saxena. But the form of authentication is perplexing as the second factor is a voice pin. It is a 4-digit pin that one will have to announce to Alexa. That flies in the face of logic as discretion is necessary.

“Yes, it has the limitation that no one should be around when announcing the pin,” admits Saxena. “But, in the future, this will be solved by voice biometrics that Alexa is developing,” he adds.

For the Alexa Skills ecosystem to thrive, finding the right use cases for voice is key. While the onus is mostly on how businesses choose to use voice skills, the implementation is important for Alexa. “When a skill doesn’t work, most users think it is Alexa’s fault. While it is actually the developers who have not implemented it well,” says Atyam. And that is not something Amazon would want. After all, Alexa has a leading role to play in the retailer’s big plan of selling Alexa-powered devices. And well, just selling.

No comments:

Post a Comment

Search This Blog

Powered by Blogger.

Blog Archive


Recent Posts