Are You Ready for Marketing Made Better?
We can help you understand your customers, stay ahead of regulations, and grow your business.
The prevalence of conversational AI solutions such as chatbots, voice assistants, and virtual agentsin customer service has grown significantly over the past year, in line with the arrival of consumer‐facing LLMs like ChatGPT and Bard.
The global chatbot market size is growing at a CAGR of almost 20% and is expected to be worth close to $5 billion by 203234. This rapid growth is primarily due to the capacity of AI-powered chatbots to deliver a proactive and personalized service across digital channels at scale, which caters to the evolving demands and expectations of constantly connected consumers. This shift toward automated customer service enables businesses to offer always-on customer service on the platforms preferred by their customers, while also streamlining operations and reducing customer service costs.
Generative AI has the potential to further revolutionize customer support through a combination of digital self-service and enhanced agent interactions. In fact, there’s a growing recognition that customer service should seamlessly blend digital technologies and automation with human support to deliver a frictionless and highly personalized experience.
In 2024, we’ll see a shift from reactive to proactive customer service. Businesses that use AI-powered chatbots and virtual assistants to solve problems before customers even know they exist, or those that proactively make suggestions to prevent potential issues, will be the ones that retain long-term customers. Businesses that get this right will forge ahead of those that respond reactively.
Case Study
Expedia’s Conversational Trip Planning
In 2023, Expedia launched a new chatbot application powered by OpenAI’s ChatGPT. In the Expedia app, users can start a conversation with ChatGPT and receive recommendations on destinations, hotels, transportation, and activities. The ChatGPT-powered feature will then automatically favorite the hotels it recommends, which saves users time and effort.
The integration of ChatGPT enables Expedia to offer a more personalized travel experience. The application analyzes user data and preferences so that, if a user asks for hotel recommendations in a particular place, ChatGPT can tailor its recommendation to the individual and suggest hotels that match their travel history, preferred amenities, and price range.
Travelers can also ask questions about their upcoming trip and receive information, such as flight schedules, weather, recommended transport routes, and relevant news that may impact their trip.
“Expedia introduced a chatbot with AI integrated through the ChatGPT plugin earlier this year, and we’ve seen great success. We use our ChatGPT chatbot during the discovery stage to drive new acquisitions and conversions. We have found that, once the traveler enters the shopping funnel through our chatbot, their propensity to complete the purchase journey is much higher than those who don’t.”
Raghav Gupta, Global Product Manager, Expedia
Consumer perspective:
A chatbot can save time and energy but can add friction to the customer experience if it’s not sophisticated enough
However, when we asked consumers about more advanced customer service applications, we found lower uptake. Only 36% have used a biometric ID to complete a transaction, and 33% have used voice recognition to interact with a company37.
Promisingly, 51% stated they’d had an interaction
with a chatbot that felt like natural human conversation, which shows the progress these chatbots have made in recent years. But suffice to say, not every chatbot is made equal. There are varying levels of conversational AI maturity for customer service, ranging from basic question-and-response bots to those that use the advanced natural language generation required to conduct a human-like conversation.
Currently, most chatbots for business do not perform beyond question and answer. They are pre-programmed with a specific set of data and are primarily used to provide quick responses to common customer questions. Most do not use natural language processing (NLP), dialog management, or machine learning (ML) to provide a 360-degree customer service experience. In many cases complex queries (including those that involve multiple concerns) are escalated to a human operator.
Consumers who have used chatbots for customer service often encounter a variety of challenges:
Of the customers who have interacted with chatbots for customer support38:
Consumer survey: Base = 1,305
Customers who prefer to message with a chatbot or virtual assistant/agent than with a human customer service agent39:
Consumer survey: Base = 1,305
While chatbots still have some challenges to overcome, they are already a key part of the customer support journey for many people. Overall, 35% of customers actually prefer to message with a chatbot or virtual agent instead of calling a human customer service agent, with this preference particularly pronounced among 35-54 year olds.
As more brands incorporate AI into their chatbots, we can expect to see their popularity and use increase. AI-powered chatbots can learn and improve over time, which will address concerns about their reliability, adaptability, and even empathy, as new use cases continue to emerge.
Business perspective:
AI is increasingly embedded into customer support with significant benefits, but human interaction still has its place
Many businesses are already integrating AI into their customer support functions and, by doing so, are seeing notable improvements in understanding customer issues and resolving them quickly. A significant 44% of the businesses we surveyed said they are currently using AI for customer service40, and these businesses noted the following benefits:
Business survey: Base of companies using AI in CX = 88
Example use cases for AI in customer service include:
Chatbots and virtual assistants: AI-powered chatbots can answer common questions, troubleshoot, and perform routine account management tasks.
Customer self‐service bots: AI-powered bots can identify or predict friction points, based on the customer’s immediate browsing activities, and suggest actions before the customer reaches out for support.
Customer authentication: AI can be used for customer identity verification through voice recognition and biometric IDs.
Tailored customer experiences: When ML is used to analyze and interpret behavioral and demographic data, customers can receive personalized recommendations or timely promotions.
Multilingual queries: AI translation applications can automatically detect the customer’s language and facilitate a conversation in this language or translate for an agent.
Enhancing agent interactions: Using data points, such as ticket type, past resolution processes, customer sentiment, and customer history, AI can recommend actions to agents. Agents can also use AI content generators to respond to customers.
Intelligent triage: Using NLP and sentiment analysis, AI can apply tags and labels to tickets, sort them into categories, and funnel them to the correct agent.
Sentiment analysis: Based on the tone and context of the customer interaction, AI can detect how the customer is feeling, and tailor responses accordingly. It can also generate valuable insights from customer interactions across support channels
Performance monitoring: AI can monitor wait times, identify gaps in knowledge-base content, and determine when it’s necessary to transfer customers to a human agent.
Inventory management: AI-based analytics of product inventory, logistics, and historical sales trends can improve customer support with real-time feedback on inventory levels.
Of these many AI use cases, chatbots are the one people are likely to be most familiar with. But, although many businesses do use chatbots for customer service, our survey revealed that few are currently using AI-powered bots.
Almost half (47%) of the businesses we surveyed say they use some form of chatbot or virtual assistant in their operations. But only 15% of businesses use a chatbot or virtual assistant that is powered by AI42.
Regular chatbots that aren’t powered by AI have a number of limitations that restrict their functionality and value, including one-dimensionality and an inability to engage in conversations beyond binary interactions. These one-dimension chatbots aren’t necessarily sophisticated enough to handle the entire spectrum of customer queries, and if they can’t answer questions or provide help in a timely or accurate manner, this can lead to customer dissatisfaction, increased workload for customer service reps, lost sales, and bad publicity from negative reviews. What’s more, using regular chatbots limits some of the proactive capability use cases we previously mentioned.
AI-powered bots can address these limitations by drastically expanding the range of use cases and complexity of tasks that chatbots can support. And with improvements in the accessibility of LLMs such as ChatGPT and Bard, as well as advances in their capabilities, we expect to see much wider adoption of generative AI-powered chatbots in the coming years. In fact, several big name brands, including Stripe, DuoLingo, and Coca-Cola, have already recognized this potential and have integrated the ChatGPT API into their operations.
But companies will need to carefully tread the line between streamlining customer service operations and outright replacing valuable human-to-customer interactions. As it stands, AI-powered chatbots can’t always provide the level of empathy, understanding, and personalization that a human can. Furthermore, replacing entire customer service units with AI-powered bots can have serious consequences for a company if the bots aren’t implemented or managed in a way that provides people with accurate and up-to-date information.
“On the flip side of all of this, it’s very early in all of these endeavors to think that the computer is smart enough to get it right all the time. The thing is, math doesn’t have morals. I think we’re on the cusp of letting the computer do some things faster and better for us, but we’re not at a point to trust it to be the sole arbiter of the path forward in all scenarios.”
Brady Gadberry, SVP Head of Data Products, Acxiom
The most successful brands will retain a set of skilled and experienced customer service representatives to handle the more complex issues that their AI-powered bots cannot, and provide customers with a personal touch when necessary.
Making it work:
Sophisticated AI‐powered chatbots need to be trained on vast volumes of high‐quality data to perform effectively
The success of conversational AI does not depend solely on technology. Good user experience (UX) design and a wealth of training data are also crucial. A chatbot that is not correctly implemented or not trained on sophisticated enough data will not be able to engage in the natural human conversations that customers expect. In fact, insufficiently trained chatbots run the risk of alienating customers by delivering empty responses or answers that don’t solve their problem.
And, given AI-powered chatbots need to be trained on first-party data, brands must make sure the tools they use to develop their chatbot solutions prioritize data security. A brand’s data is incredibly valuable and no one wants to risk jeopardizing customer privacy in any way.
Key pillars that businesses need to get right to maximize the value of their AI integration for customer service include:
Data and analytics capabilities:
- The ability to ingest large volumes of customer data to build intelligent customized experiences
- A nerve center for descriptive and predictive analytics
- CRM and cloud-based call centers (if relevant) embedded into an AI-driven tech stack
- Significant data storage capabilities
Implementation:
- Seamless integration of AI chatbot tech with legacy systems such as CRM, commerce platforms, and enterprise resource planning software
Training:
- An AI-enabled automated intent recognition and resolution layer (i.e., intelligent triage) that can analyze the customer request and determine the intent, sentiment, and language – including making proactive recommendations
- The ability to train AI on specific models
- Prompting and tuning LLMs to each service channel to ensure interactions are contextually appropriate
Management and optimization:
- Executing regular updates and maintenance to models
Privacy‐centric design:
- Ensuring all privacy policies are updated with current use cases
- Ensuring security requirements are met
- Ensuring data use is permissioned and meets all privacy regulations
Where next?
As businesses continue to integrate AI across customer support functions, they’ll need to be mindful of increasing customer expectations for positive, seamless experiences that deliver answers quickly and effectively. To facilitate this expectational shift, businesses will be prioritizing developments in four key areas of conversational AI:
Handling more complex requests: Ultimately, advances
in LLMs will enable chatbots to complete more complex tasks. In the meantime, businesses may enable their bots to handle more specialized tasks by adopting a multibot architecture where specialized chatbots function as an ensemble, accessed through a unified interface. After inferring the user’s intent, requests are directed to the appropriate specialist bot.
Improving personalization: We expect further advances
in personalization, with bots adjusting their communication style in real time according to individual customer characteristics, such as formality of communication, and emotional and cultural context.
Introducing visual capabilities: Businesses will use visual AI to solve customer problems in real-time. While basic visual AI – for example reading text – is already widely adopted, using visual AI to identify an issue with a specific make or model of a device or product – perhaps using video – requires functionality that is currently only accessible to businesses with advanced capabilities.
Investing in voice: Differences in accents and cadences of speech, as well as the unpredictability of human conversations, mean AI is only just beginning to be used for verbal interactions. However, recent advances in text-to-speech and speech-to-text technology are making it easier for LLMs to conduct verbal conversations, which will encourage adoption for customer service interactions.