Setting realistic expectations for AI driven virtual assistants.

In recent years, AI has made a leap from a distant futuristic dream to an almost tangible technology used around the world in personal and professional contexts. AI is the base for voice assistants on phones and at smart home systems (e.g. Siri from Apple and Alexa from Amazon), unmanned vehicles that are increasingly invading the streets, and chatbots that can be widely seen on messaging platforms (such as Facebook Messenger, Telegram and WeChat).

Businesses turn more and more often to virtual assistants that can help them in numerous business automation processes and increasing productivity. But more often than not people would get frustrated because the idea of an “all-powerful and all-capable AI” is far from reality. Our inflated expectations, are they becoming the stumbling blocks that prevent us from taking full advantage of a potentially beneficial technology? How can we set more realistic goals for a modern business chatbot?

To evaluate AI realistically, we need to take a step back for a moment from all the predictions, promises and hype and see exactly where we are standing.
While pure AI makes a positive contribution to analyzing data and processing huge amounts of information, it cannot be a self-learning solution for free communication, as we have seen in the example of Tay from Microsoft. Although intelligent virtual assistants and chatbots are hugely contributing to out day-to-day productivity, we need to admit that we are still far from Jarvis-style supercomputers. Don’t expect your virtual assistant to be omnipotent just yet. When it comes to introducing virtual agents into business processes, chatbots need a combination of self-learning and human content moderation.
One of the keys to overcoming frustration is being conscious about what to consider when choosing a chatbot and what to expect from its usage.

1) Proper data training. Most of the time a chatbot consists of ready-made answers, which a human operator had filled it with. Without examples, no neural network will be able to fully understand requests. Useful content is the key to everything. The better a company defines an objective for a particular use case, the more information it can provide about the topic, the smarter the bot will be.

2) The technology used. Today, most of the chatbots still operate on the basis of a set of rules and behaviour scenarios. However, the natural language being vague and ambiguous, one thought can be expressed in many ways, so the commercial success of conversational systems depends on solving those language processing problems. The machine needs to be taught to clearly classify and interpret all the variety of incoming requests. Opt for a chatbot with a developed NLU system to help you with this process. And never skip the proper testing phase.

3) Supervision learning, not ‘self-learning’. The key principle of successful chatbots is the control and the potential processing of the bot responses. This is why pure AI is not suitable for business context – the technology has not yet reached a level where fully self-learning and self-sufficient systems can provide truly qualitative responses. The hybrid approach of self-learning and human moderation of content allows the chat system to continually improve depending on how it is used, and also allows companies to control the quality of their responses. A properly installed feedback cycle allows receiving comments on outdated or incomplete content in real-time; this paired with human content moderation can give companies absolute confidence that their chatbots respond to user requests in a predictable, consistent and legally approved manner.

4) A right governance platform. When the chatbot knowledge base is integrated within the ‘right’ platform, the company can set up the necessary approval procedures to ensure that moderation meets its organizational requirements and internal norms. This platform should also allow the company to decide which types of “training” the system will automatically apply to future conversations without human intervention and which areas of content require human approval or further development. A proper governance platform that centralizes the bot knowledge with the human ability to improve it is one of the essentials for a happy bot experience.

5) Channel agnostic assistant. The number of messaging channels is growing in staggering numbers. Channels are being created, scooched over, replaced and recreated. Businesses constantly migrate from older communication platforms to new and more sophisticated ones. In order to avoid being stuck with a chatbot running exclusively on your outdated channel, opt for a channel-agnostic chatbot, that will be able to adapt to whatever environment you might be using in on.

Where does it take us?

The current capabilities of AI-based chatbots and virtual agents allow these solutions to perfectly complement human efforts. Chatbots can be used on customer-centric support channels (website, mobile communications, SMS, social networks, messaging applications, etc.) as well as on internal business channels (team messengers, intranets, internal communication solutions) and be managed and monitored from a centralized knowledge base platform. A single centralized knowledge governance platform is the key to the consistency and relevance of information across all channels.

It is also important to implement a feedback loop, which allows people to provide real-time feedback and suggestions on the contents of a chatbot, in order to create a better conversational experience. When this feedback loop is linked to the centralized knowledge base and functionalities mentioned earlier, “live” agents can continuously improve their chatbots without any additional effort.

This harmonious cooperation between people and machines working together also benefits businesses by reducing costs, reducing staff and increasing the motivation, qualifications and satisfaction of people involved. In addition to automating business processes, chatbots can be incredibly good at providing quick information, skill training and rapid access to various documents, to mention just a few.

Closing thoughts

With regard to AI, chatbots and virtual assistants, companies should make decisions based on realistic expectations, but they should not be afraid to include this technology in their business process automation strategy or plan for deeper integration in the future. As companies continue to migrate to digital channels, the decision to integrate a SaaS virtual assistant into their processes seems as a logical follow-up of the digitalization of a modern workplace. And while companies should not worry about the AI being not yet perfectly autonomous, they should be more conscious about being possibly left behind if they cannot accept the reality of people and machines working in harmony as part of their approach to business automation.