Opus Research has just issued a comprehensive assessment of the intelligent assistant (IA) and bot ecosystem and vendor landscape. We focused on the 28 vendors we identified as leaders in offering enterprise-grade solutions.
By our estimate, this group accounts for more than 2,700 deployments of intelligent assistants for a customer base approaching 1,200. In terms of dollars spent on intelligent assistant software, services, and professional services, we're still in a heavy growth period. Starting from a modest $225 million in 2015, Opus Research estimates that figure to reach $1.2 billion this year on its way to $5 billion in 2021.
Global Intelligent Assistance Spending (in millions)
To be honest, around 90 percent of the 2,700 intelligent assistants in the enterprise setting are close relatives to virtual chat. The earliest ones, introduced around the turn of the century, were natural language front ends to a fixed set of static answers. Companies made it easier for customers to find the FAQ section of their websites, or they provided conversational shortcuts to the answers to questions from the "Contact Us" link.
These limited-function offerings were able to handle a high percentage of the questions sent their way and have given other customer care and digital experience professionals the confidence to teach old assistants new tricks. The natural language processing became core to more accurate understanding of queries. At the same time, the knowledge bases that informed the intelligent assistance added depth and currency. That's why these text-only cousins to virtual agents comprise an important branch in the intelligent assistance family tree.
Speech professionals have kept their eye on the 10 percent of implementations (equating to about 270 enterprise-based virtual agents) that incorporate speech processing and dialog management resources. Nuance and Interactions (with AT&T Watson's core speech processing resources) took the lead in making their speech-enabled virtual assistants a starting point for phone-based self-service. During the past two years Google, Apple, Amazon, and Microsoft have joined a collective effort that has improved core recognition accuracy dramatically. It is routine to see quotes of more than 90 percent accuracy or, more dramatically stated (for some reason) a sub-10 percent error rate.
Tight coupling between automated speech recognition (ASR), transcription, and natural language processing has further improved the ability for automated virtual assistants to recognize the intent of a call and very quickly resolve issues for the caller. This ability has been enhanced by improvements in enterprise search, interaction analytics, CRM and knowledge management. All are systems that bring highly dynamic, and often unstructured, data into the mix of possible responses.
The third dimension along which intelligent assistance has evolved is in the human realm. In the five years since Apple introduced Siri on the iPhone 5s, people have become accustomed to the idea of taking command of devices and the services they offer by using their own words. While it doesn't always work, the introduction of voice-first assistants like Amazon Alexa, Google Assistant, Microsoft Cortana and, soon, Samsung's Bixby (based on Viv) attest to the fact that the general public has gained sufficient levels of confidence to turn to their virtual assistants on a regular basis. User acceptance and repeated use has gone hand in hand with improved accuracy and successful completion of tasks. It's a virtuous cycle.
When you add the 10,000 skills developed for Amazon's Alexa and the 5,000 or so actions put under the voice command of Google Assistant, we can start to consult the wisdom of the crowds of developers, deployers, and technology providers to figure out what's working, what isn't, and how best to get started and succeed in the world of intelligent assistance. This is definitely a space to watch.
Dan Miller is founder and lead analyst at Opus Research.