Writing exclusively for AIBusiness.org is Ed Smith, Chief Product Officer at Humley, who is giving our audience an insight into how we can realise the true potential in Virtual Assistants. 

Humley is a company focused on removing the challenges of deploying and training AI systems. They focus on automated training, industry agnostic technology, and tools which help a business understand, manage and launch AI powered services. Humley believes the real power of AI is its ability to improve everybody’s daily life. Whether that is helping workplaces be smarter, engagements with brands easier and quicker or an experience more individually tailored.


Ed Smith

Smith brings over 20 years experience launching D2C and B2B products and services globally; including 10 years with mobile technology startups such as TIS (acquired by Scottish Telecom/Thus), AirMedia, Elata (acquired by Qualcomm), and a further 7 years based in the USA with Qualcomm Inc. the worlds largest mobile chipset vendor.

Smith is currently Chief Product Officer at Humley, leading a team that applies Artificial Intelligence to human engagement technologies through natural language communication with a focus on context and actionable responses.

How do we realise the potential of Virtual Assistants?

The success of automated assistants crafted from a range of scripted conversation tools is inherently limited and produces narrow scope bots or novelty bots. The problem is that a script is only good if the user knows/or is willing to follow it. Virtual assistants that attempt to serve the public with a broad service find scripting (albeit with a degree of understanding) is inherently too narrow and does not scale.

So how do we bridge the chasm between narrow bot technology and the expected potential of virtual assistants? An IBM survey indicated that 73% of users would prefer to solve their own problems than speak to a human to resolve them. The answer from Humley is to start with a deep understanding of the knowledge that will power an assistant. By understanding the knowledge it is possible to handle a critical mass of user queries and therefore provide inherent value and customer acceptance. As illustrated below, starting with a broad knowledge base, user behaviour can be examined and additional narrow scripted capabilities can be added to that broad knowledge based assistant to perform narrow scripted tasks.


Take a virtual assistant for a new phone for example. Using the Humley approach, a user can launch the assistant and ask questions. The assistant will be able to use the broad knowledge base to answer a wide range of those questions. Humley can also launch the user into the right screen on the device, video tutorial, app or webpage.

The Humley knowledge approach provides an assistant that will point the users in the right direction and offer instructions, advice, or direct solutions, without any scripted interactions. Once it is understood where the peak areas of concern are, scripted interactions can be added where appropriate to help the user further, or seamlessly hand over to a human (with the background information) to get the user issue resolved.

The access to information from a knowledge base can present a sterile user interaction. The User asks a question, and an answer being returned does not create a compelling user experience.

To address this, Humley developed a range of conversational components that augment the interaction to simulate a two-way conversation without scripting (see blue section in illustration above). These modules include small talk, clarification, context follow through, alternative suggestions and many more small strategies to smooth the communication around knowledge interrogation.

Humley launched it’s conversation capabilities early 2016 and now boasts a range of conversation components that allow it to respond to the user when confident, partially confident, or where there is little confidence in the answer. It assists the user to find an answer where they may have been misunderstanding, or by holding conversation context through a dialogue. Each of these conversational components individually simplify the conversation only a fraction, but together they instil a confidence within the user that they, and this virtual assistant, will be able to get the right answer.

As users become more familiar with Siri, Alexa, Cortana the pressure for individual brands to provide user interaction via natural language increases.

To deploy these services there will always be space for carefully scripted interactions or triaging smooth handovers to human agents. However, Humley’s experience has demonstrated that the route to scalable, deployable virtual assistants begins with a knowledge-first approach.

Once that knowledge foundation is laid we will find more virtual assistants breaking through from PoC/trial status to live services offering valuable and vital customer touch points.

For the latest news and conversations about AI in business, follow us on Twitter, join our community on LinkedIn and like us on Facebook