The market for virtual assistants (VAs) continues to demonstrate strong positive momentum, driven forward by a combination of rising demand and improved accessibility across the enabling technologies. Global shipments for ‘built-in’ VAs, covering products offering direct access via microphones and speakers, will exceed 2.3 billion units by 2023. This is up from a projected 1.1 billion units throughout 2019, representing a CAGR of 20% over the forecast period.
There is industry-wide opinion that virtual assistants will rapidly become commonplace across consumer electronics. Declining cloud compute and hosting costs, open-sourced machine learning algorithms like TensorFlow, Caffe and integrated neural network accelerators within SoCs mean it has never been easier to integrate virtual assistants. Most conspicuously, virtual assistants offer an intuitive way for consumers of all ages to engage with products and services, since voice presents a natural mechanism of interaction that effectively shortcuts otherwise complex user interfaces.
However, there are potential headwinds and challenges to overcome. Voice control is viewed ubiquitously as a technology with durability, and one that will certainly feature across a diverse range of applications because the costs of adding virtual assistants are falling considerably, with module costs in the region of $8 to $10. From smart homes to self-driving cars, the race is on to add virtual assistants to enhance the user experience. But there isn’t a ‘one size fits all’ solution and so we expect use-cases to mature in the longer term. Device manufacturers have the option to consider a ‘works with’ strategy alongside the alternative of offering the technology ‘built-in’. Smart Home vendors prefer ‘works with’ because it offsets the risks of technology obsolescence, whilst enabling them to work with all ecosystems; meantime others seek to offer ‘built-in’ solutions that provide virtual assistants with specific capabilities for individual application. For example, the automotive sector favours something more bespoke that is built into the vehicle, processing voice commands locally and using cognitive arbitration to hand off generic requests to the cloud.
Most importantly, voice interfaces must move beyond the mechanics of ‘command and control’, effectively akin to replacing a button press in an app. Research suggests that users rarely explore the full potential that virtual assistants can offer, often reverting to known or trusted functions, such as playing music or setting timers. Whilst smart displays regularly encourage users to interact with them through visual prompts – “Ask Alexa what the weather will be like” – screenless devices, such as smart speakers, do not have the same opportunity, instead sitting resolutely dormant until summoned by the user.
So the industry must move towards more intelligent virtual assistants that improve consumer engagement and, with voice processing capability now moving into the devices themselves to reduce latency, there is opportunity to extend the virtual assistant ecosystem to offer conversational platforms that avoid the risk of stagnation. The major VA platform vendors know this: Amazon, Google, Microsoft and Samsung are all focused on improving language and conversational support this year, whereas primary vendors in China, notably Alibaba and Baidu, are focused on developing the utility and reach of their platforms by extending the AI capabilities. All are aiming to offer an effortless and, above all, trusted experience that truly begins to drive an uplift in engagement. It’s one thing for consumers to ‘use’ their virtual assistant, but it’s another level entirely when they ‘trust’ virtual assistants to handle more complex tasks and those that present risk, such as large financial transactions. Once this happens, the utility of virtual assistants will expand to encompass a broader range of applications and from this new monetisable opportunities will emerge.