AI – Beyond Simple Messaging
By: Rob McDougall, CEO, Upstream Works Software
Many contact centers have a large focus on using Artificial Intelligence (AI) applications to provide a level of AI-assisted services via the website. Many companies, when they think of conversational AI, think of chatbots as a default. But – changing demographics aside – one of your largest communication channels is likely still email. At Upstream Works, we have seen email volumes come as a complete surprise to some of our customers; we have seen email completely ignored as a channel in the contact center, to the degree that there is no concept of how many email communications come in a month or what percentage of the contact volume that represents.
And apparently, the industry is missing the point, too. Conversational AI tools abound; searching for chatbots is an easy exercise for locating providers and a difficult one to choose from all the available options.
But conversational AI applications have been trained to emulate conversations. They are designed to support a two-way dialogue (Google’s AI is even called Dialogflow!). They have not generally been trained on how to deal with a paragraph or two at a time – which is generally how email works.
A dialogue via chat allows the AI to systematically narrow down the choices for what the customer is asking for and determine the best probability outcome. They would even be trained to work optimally with multiple short responses – fragments, sentences, or short paragraphs. These are natural to a chat (or voice) conversation. Stray from this paradigm, and you will quickly find out that AI has its limitations. Most conversational AI, for example, cannot parse out two elements in a request. “Can I have information on my house insurance and car insurance,” will generally give you information on one or the other, but not both.
Email, on the other hand, uses a substantially different paradigm. Email is constructed free form, with multiple paragraphs used so that all information required is typically provided in one shot.
Your typical chatbot provider just hasn’t been trained to deal with this type of information.
And given the lack of focus on email as a channel, it’s not even that easy to find emailbot providers to help you automate your email workload. But, they do exist.
To work, your email AI application will be functionally different from your chatbot and will likely come from a different vendor. Operationalizing these multiple platforms requires planning and focus to ensure that the customer and agent experience is not degraded. But many of the same rules still need to apply when selecting your email AI application. KISS is still an important factor – use AI only for the basic routine email requests, and don’t try to get fancy. Route the hard stuff to the agents.
And without good planning, then you won’t be able to determine if your email AI is working – you need to measure its success in the same way you would measure an agent or a chatbot. You need to provide the ability to easily escalate to an agent for email responses that have a low confidence level. And you can’t add new desktop application silos to further confuse your agents.
Providing a usable email AI application in conjunction with a separate conversational AI application will, without further consideration, create more new silos of customer service for agents and consumers.
Think about how to integrate those different solutions into a single framework within your business to standardize the workflows, simplify the agent experience and provide consistency across all channels to your valuable customer base.