How to Scale Conversational Artificial Intelligence?
Assistants and bots have reached a new adoption high. Opus Research found spending on intelligent assistants will cross the $2 billion mark in 2018 and will exceed $5 billion by 2021. However, many businesses are finding their projects harder to scale than they expected. Gartner predicted that, by 2020, 40% of virtual assistants launched in 2018 will have been abandoned. The disappointment with some deployments triggered controversy whether to use Artificial Intelligence (AI)-powered assistants or to stick with rule-based bots. Let’s explore what needs to be done to get the most out of conversational AI.
Enterprises are still experimenting with Virtual Customer Assistants (VCAs). Assistants and bots are relatively easy to set-up. Companies can start rapidly but are finding it more difficult to scale and can become disappointed. Challenges maximizing conversational AI should not be construed as the technology not being ready for large deployments. At the same time Gartner issued its prediction on the fate of VCAs being launched, it acknowledged the strategic nature of conversational AI. It stressed that VCAs consistently reduce the volume of inquiries by up to 70 percent, increase customer satisfaction, and save 33 percent over equivalent resolutions by call center agents. It also forecasted that the combination of messaging and conversational AI will drive 20% of brands to abandon their mobile apps by 2019.
First-wave assistants and bots have been deployed as point projects. They started targeting a small set of topics on a single channel. Businesses then expanded iteratively their scope. Such an agile approach remains the way to go, but important questions must be addressed to get the full benefit of the technology.
Today’s bots are good at dealing with a focused set of situations. Scoping and design are critical. The agile approach may tempt people to expedite the project definition phase. Important considerations include:
- What level of complexity should be handled by the assistant?
- Where to draw the line between self and human-assisted service?
- Should the bot aim at providing answers, make suggestions, or just uncover intent to direct questions?
- How should different types of inquiries be guided to digital or voice channels?
- Should the assistant experience be branded?
The technology landscape is pretty fragmented with almost 150 providers. Solutions tend to shine in different areas. The job at hand and the increased software specialization should guide the options to consider:
- Some platforms provide APIs for enterprises to build their own bots while others include Natural Language Processing/Understanding (NLP/U).
- Some solutions with embedded NLP/U capabilities come with built-in optimizations for specific use cases while others require more tuning.
- Assistants are getting more specialized with new categories such as answer bots starting to emerge.
Eventually, bots and assistants can no longer be considered in isolation. A conversational AI roadmap must include integration with the other elements of the customer interaction stack, in particular:
- What analytics are required to perfect the experience and guide the next best actions?
- How to prioritize channel activation?
- How to leverage conversational AI to drive the adoption of digital channels?
- Which integrations to CRM and other applications are needed to personalize the experience?
- How to combine conversational AI with knowledge?
We will discuss these issues in a panel at the Conversational Commerce Conference next week. I am thrilled by the caliber of the participants with Linda Crawford, CEO at Helpshift, Marina Kalika, who leads Enterprise product and solution marketing at Nuance, Richard Kimber, Daisee founder and CEO, and Tobias Goebel, Vice president of product marketing at Sparkcentral. I hope you can join us!