July 15, 2022
To quote the famous Honda advert of the early 2000’s, ‘wouldn’t it be nice if things just worked?’ If you haven’t seen it or would like to again, you can find it at the bottom of the page. Nowhere is this more applicable than when it comes to chatbots. As consumers, we’ve all been on the end of a pretty unsatisfactory experience that has left us feeling anywhere from a little annoyed to totally exasperated. The question is, what does it take to get a chatbot to work as it ought to? What are the elements of an effective human to machine conversation?
Whether it is the technology used in design, the inability to facilitate natural, context-rich human to machine conversation or accurately capture this information, cross-platform and pick up the conversation with a returning customer further down the line, there are a series of fundamental, yet incredibly common problems synonymous with chatbots.
Many are failing miserably to connect with their users or to perform simple actions. In a worst-case scenario, their failures make headline news. For example, Scatter Lab’s Luda Lee, gained attention with her straight talking style, attracting 750,000 users and logging 70m chats on Facebook. However, in late 2021 she made homophobic comments and shared user data, leading over 400 people to sue the firm.
Nabla is another example from a Parisian healthcare facility. It tested GPT-3, a text-generator, for giving fake patients medical advice. So a “patient” told it that they were feeling very bad and wanted to kill themselves. The GPT-3 answered that “it could help [them] with that.” Once the patient sought affirmation for whether they should kill themselves or not, GPT-3 responded with, “I think you should.”
While it’s easy to see the funny side, the subtext is critical – that it’s vital to get human to machine conversations right. That being the case, there are six critical elements for businesses to consider to avoid being added to the list of famous chatbot failures.
A conversation, regardless of whether between humans or with a virtual assistant, must have a purpose and should end with a clear conclusion. While this is the primary purpose of communication, it also needs the ability to sway and fluctuate, much like we do as humans. If a virtual assistant isn’t designed to do this from the get-go, it’s a near impossible task to get it right. A virtual assistant also represents the brand as a digital ambassador so how it speaks and the language and tone of voice it uses should be reflective of the brand to deliver the same experience as a human would.
As humans, rarely do we have one linear chain of communication. For example, if we go into a shop to buy a pair of shoes we often try on several, ask about sizes, consider how the shoes might go with our clothes, ask about different styles or even leave the shop after looking and come back later. We shift through different contexts in a conversation and expect the person at the other side of the transaction to recall and react accordingly without having to start from scratch each time. It’s the same for virtual assistants. Using conversational AI they should have access to past knowledge to leverage relevant information at the right time to create dynamic conversations and take the most appropriate action. And they should know the same sentence – “can you help me?” for example – will have a different meaning and require a different reaction at different points (we call them scenes) within a conversation.
With the complexities of human language, no solution is ever going to be 100 percent accurate, and the virtual assistant needs to be designed to handle errors in dialog flows during a chat. One of the main frustrations with virtual assistants today is that they’re difficult to reverse if a customer needs to correct a question or seek further clarification on a previous point. Neither should an error automatically mean the user is bumped to a help centre or simply given a number to call – it rapidly undermines the point of the virtual assistant, which is to deflect conversations away from expensive human resources and provide an enhanced customer experience
Chatbots have historically been built in a rigid fashion, often requiring short-term patches and fixes to address changes in circumstance. This has inevitably led to many resembling a kind of digital Frankenstein that, over time, becomes a shadow of what it was intended. There needs to be a baked-in acceptance that improvement is constant. Creating successful virtual assistants requires a recognition that they can be adapted over time quickly and easily to mitigate ongoing change management requirements.
Business must have the autonomy to implement change easily and without the need of deep technical skills and/or hours of ML training. This means the ownership of a conversational application should sit within the business – not a third party agency or designer – and require as little coding and developer expertise as possible. Ideally, none at all. Similarly, any virtual assistant needs to integrate with external or legacy services quickly. If it can’t do this, it’s a project that is almost certain to fail.
Finally, they need to have the ability to scale and grow with the business. Virtual assistants, by their nature, are not a flash-in-the-pan quick-fix but a strategic platform to automate conversations and complex processes across the enterprise. They must be viewed as such. If a business is investing in a platform for one part of the operation, then there’s no reason why this shouldn’t be replicated in others. If a virtual assistant can’t scale on-demand and where required, it will quickly turn from a help to a hindrance.
Businesses need a virtual assistant to be fit-for-purpose. Many build on yesterday’s technologies have failed to deliver on the significant potential, leading to abandoned projects and orphaned deployments. But the correct platform – one that delivers on these six elements – is a significant asset, and one that can and will realise a bold vision; changing forever the way humans interact with machines.