January 23, 2020
According to Gartner, the market for Conversational Artificial Intelligence (AI) is predicted to grow between 25% to 45% annually between 2020-2023. 31% of CIOs surveyed from industry-leading, global organizations have already deployed or are in short-term planning to implement conversational platforms in 2018. This acceleration in the deployment rate reflects an increase of 28% from 2018.
Regardless of industry, e-commerce, Healthcare, Telecommunication, or Banking / Insurance/ Investment, organizations' maturation of their digital transformation has resulted in the adoption of Conversational AI. The geographical landscape of the vendors investing in Cognitive Virtual Assistants (CVA) is split with the Americas have the largest market, closely followed by EMEA, as quoted by Gartner.
This transformation of Conversational AI is propelled by drivers such as; user experience improvements, mandates for productivity increases, and flexible but controlled accessibility, to name a few. The heightened interest and advancement in human interactive technology has led to organizations utilizing Machine Learning (ML) and Natural Language Processing (NLP).
Amidst all the brouhaha of impressive adoption and growth statistics, several questions arise -
- Why is it that organizations still hit a wall when it comes to CVA's delivering value and experience?
- Are CVA's only fancy decision trees with responses based on If-Else statements?
- Why was the "intention of the conversation" missed?
- Why is it that during your conversations, you get either a "Sorry! I did not understand your issue" or an unintended Warm Transfer to a live agent?
Based on our experiences working with numerous clients, delivering Proof of Concepts and conducting client workshops, we envision that the below checklist will ensure that your CVA delivers on promises made:
Setting the right expectations
Often prospects have a grandiose image as far as the capabilities of a CVA are concerned. A CVA can only be the Jarvis for your organization if it can interact with different constituents of your ecosystem. For example, this could be achieved through Webhooks or customized Out of The Box (OTB) integrations, as well as ensuring the CVA has the AI to handle context/intent-based personalization's.
Clearly, there have been advancements in NLP that are at our disposal, but there are some limitations and dependencies that need to be well understood, alerted on, and managed to set the right expectations. For example, yes, NLP has advanced, but limitations concerning understanding verbalism such as sarcasm, subtle humor, or multiple contexts still exist. It is essential that buyers are sensitized to the usage of CVA with defined problem statements and desired outcomes.
Proof of Concepts (PoC)
PoC or a working prototype is the best way to explore any environment for potential CVA success. At a high level, it helps to identify potential gaps, an opportunity to solicit feedback from users and proofread the functionality flow. At a granular level, from a CVA standpoint, it helps to estimate, finalize, and size use cases.
Organizations get an opportunity to test out the Training and Performance phases of the CVA deployment. These tests entail creating data models consisting of relevant data and devising ways in which that data will be leveraged during user conversations. The recipe for a successful PoC includes an effective environment that provides training data, API function, cloud services, etc.
CVA Adoption Strategy
Digital Communication has permeated our everyday lives. Therein lies the opportunity with CVAs. However, the success of a CVA or any product for that matter depends on whether it performs as expected.
It is often seen in scenarios where the use cases are clearly defined, the scope is outlined, and the integrations are in place, yet the adoption rate of the CVA is found wanting. This becomes crucial where the CVA is not the only way to get a task done, but merely an option. Chatbot Adoption strategies should involve considerations like debunking the myths associated with the use of CVA, marketing campaigns to generate interest and awareness, user training so that skills are acquired enough to extract value and monitor adoption metrics on an iterative basis.
NLP is the heart of a CVA. All the top NLP engines have their own terminologies regarding Intents, Utterances, Entities, Sessions, and Context.
However, the bottom line is that your dialog flow should map to your CVA vision. For it to be a successful journey, you need to take measures like avoiding conversational dead-ends by providing fallbacks or user evaluations. If you do not have your own proprietary NLP, an effortless way out would be to follow best practices for the ones you do use.
The plot is straight forward; whether your CVA is an open domain or closed domain, it should converse with the user to achieve a goal, and for that, effective Dialog Management is indispensable.
Last but not least are Conversational Insights. CVA's often hit a wall when insights are either ignored or not acted upon. Conversational Insights are crucial as they enable continuous improvement through rectification of existing flow, identification of potential use cases, and enhancing the overall conversational experience.
Shinning a bright light on Conversational Insights, post-implementation, yields three distinct classifications:
- User Insights
- Interaction Insights
User Insights: Aimed at your CVA users, they are quantifiable insights such as count of users, active users, new users, active geographies, etc. The numbers crunched are all aimed at how many and what kinds of users are talking to your CVA
Interaction Insights: As the name suggests, these are derived from the conversations or interactions users have with your CVA. These Insights can be measured in conversation numbers, use cases triggered, and conversation length, to name but a few.
Miscellaneous: These include diverse Key Performance Indicators (KPI) such as user feedback, warm transfer numbers, APIs hit, etc. Miscellaneous Insights These are subjective to the ecosystem the CVA exists within, and the value to be derived from them.
The good news is that the market is awash with third-party tools and platforms which can track a wide range of above-given metrics for retention and engagement.
CVA's have passed through different stages of the Hype Cycle and, as a result, are everywhere around us. They still have a long way to go before they become the cornerstone of business functions, other than customer service, where they are considered an obvious fit. As the technology matures, each new generation of CVA is bound to be better than the last.
In the next part of this series, we will cover how DRYiCE Lucy ensures the above checklist, making it an enterprise-ready CVA.
- Market Guide for Virtual Customer Assistants - Gartner
- Driving Enterprise Chatbot Adoption – Everest Group
Himadri has over 5 years of experience in Product Management, Pre-Sales, and Business Development for user centric product solutions. He is passionate about products that showcase proactive Machine Learning intelligence based on user data and analytics. An advocate of the qualitative view while building products, his current work as a Product Manager entails product planning and execution throughout the product lifecycle.