IDC Opinion

Almost all types of organizations have customers, and most of them struggle to effectively scale communications with customers that have questions or need actions taken. Significant amounts of money are spent on contact centers, interactive voice recognition (IVR) systems and websites meant to support a company’s interaction with customers. When the COVID-19 pandemic struck, companies found themselves with an even larger challenge than they had faced previously. COVID-19 placed massive stress on communities and economies around the globe. Healthcare providers, companies, and government agencies were overwhelmed by the number of urgent inquiries and concerns caused by this pandemic. Across the globe, companies in virtually every industry — whether banks, hotels, manufacturers, or retailers — were exploring ways to communicate with customers while their physical locations were closed to the public. Call centers were swamped, as anxious consumers looked for answers about the virus, purchased needed supplies and goods, or attempted to cancel or change travel plans. This situation caused long wait times and dropped connections, preventing callers from getting help when they needed it most. At the same time, call centers were short-staffed as agents called in sick or attempted to transition to remote work. 

Throughout all of this, conversational artificial intelligence (AI) tools and technologies proved their worth during the COVID-19 crisis. Organizations turned to conversational AI software platforms, developing customer service applications for help during these unprecedented times. For years, organizations have wished for “virtual” or “digital” assistants that could carry some of this conversational load working with customers. The good news is that recent breakthroughs and improvements in speech recognition, natural language processing(NLP)/natural language understanding (NLU), and conversational artificial intelligence made that wish come true for an ever-increasing number of organizations over the past two years.

Over the past three to five years, due to advances in deep and machine learning, conversational AI applications can understand and respond to conversation in all its various forms, including telephony, voice, text messaging, web messaging, WhatsApp, and Facebook Messenger. The conversational AI software platforms that IDC has evaluated as part of this IDC MarketScape have shown that organizations can develop and deploy sophisticated AI-based conversational agents that can interact with customers, consumers, and the public at large to answer their questions, help them conduct transactions, and provide a wide range of self-service that wasn’t possible only a few years ago.

As part of this evaluation, IDC spoke with dozens of organizations using these conversational AI software platforms to develop and deploy applications that are providing great customer service and generating significant return on investment (ROI). Among the technology buyers IDC spoke with, we noted a range of maturity in implementations of conversational AI, from point solutions to enterprisewide deployments. If your organization is not using or evaluating conversational AI applications for customer service use cases, it should be. 

The technologies behind these conversational AI software platforms are good and getting better by the day, but that shouldn’t stop organizations from evaluating and implementing these solutions as soon as they can. Conversational AI tools and technologies are rapidly evolving, and new vendors, products, technical innovations, and acquisitions are a frequent occurrence. Conversational AI platforms can be used for a broad set of use cases related to customer service, for both customer-facing applications and internal-facing agent assistance, and these platforms can vary greatly in terms of prepackaged offerings and templates, low-code/no-code tools for business analysts and line-of-business (LOB) subject matter experts, and customizable developer tools. For example, some organizations will benefit from vendors that offer low-code/no-code tools and other features that can eliminate the need for one or more of the initial three steps in the conversational AI-build process (see Figure 2). Other organizations will need the ability to work directly with one or more of these areas to customize aspects such as language, conversation flows, and workflows.

Tech Buyer Advice

IDC offers the following advice to technology buyers considering conversational AI:

• Assess your starting point. If your organization is new to conversational AI or embarking on a new use case or channel, decide up front how you will train and develop the system, educate users, and measure success. Consider the following questions:
  • Does the organization already have a knowledge base and/or archive of past conversations that can be used for training? If not, will the vendor or public sources provide adequate training data?
  • To what extent will you want to use in-house resources for training and development versus having the vendor manage this for you? Depending on your organization, you may want to consider working with the vendor to train internal resources.
  • How important is customizability (i.e., the ability for in-house developers to tinker directly with code to make changes wherever they see fit) versus low code/no code (i.e., tools that often include a graphical user interface [GUI] and are aimed at enabling business analysts and LOB experts to develop with little to no coding expertise)? While some platforms offer both, there are also a number that focus on one of these over the other. Buyers should also be aware that so-called “low code/no code” tools may vary in terms of user-friendliness for those with no developer experience or expertise. Many vendors will offer training on these tools, where and as needed. 
• Develop requirements. Regardless of whether you are focused on voice- or text-based channels, consider the type of knowledge, interactions, question answering, and task completion that the conversational AI application will need to handle. Important questions to ask include:
  • What languages will the platform need to support? How flexible will it need to be in moving between languages?
  • What channels of interaction are most important for users? 
  • How domain specific will the application need to be? How accurate will it need to be on domain-specific terms?
  • What other back-end systems will the application need to connect with to answer questions and complete tasks?
• Decide what reporting capabilities you want. Consider whether and what reporting capabilities you will need to both monitor conversational AI performance and build on the capabilities of the conversational AI. Some of the buyers that IDC spoke with wished they had spent more time up front ensuring they could use reporting and analytics to capitalize on the rich conversational data generated by user interactions with conversational AI.
• Test conversational AI for customer service in one use case, then embrace it across the entirety of your customer service operation. Conversational AI systems can provide value across a variety of use cases in the customer service domain. These systems can be costly and time consuming to develop, so it is worth considering early how you will build on initial use cases and adjust business processes to expand the reach of your conversational AI investment.
• Educate customers and employees. For some, conversational AI applications will be a natural interface for asking questions or completing actions, but others may require some education about the benefits, limitations, and best uses of these applications. When rolling out a virtual assistant to a customer base that may be used to, for instance, communicating with a human by phone, communicate these aspects early and often.

Methodology