Artificial Intelligence (AI) chatbots enable companies to provide immediate and efficient customer service and engagement. AI chatbots leverage natural language processing (NLP) techniques to understand and engage in human-like conversation. Brands using AI chatbots report significant results, such as 25% increase in online bookings, 300% increase in ROI and 91% positive sentiment ratings.
In this article, you will learn:
- What is an AI chatbot
- How AI chatbots work
- How leading AI brands use chatbots
- Key trends in chatbot technology
AI chatbots are software that use natural language processing (NLP) to engage in conversations with users. You can include these bots in mobile applications, messaging apps like Facebook Messenger, on websites, email, and even voice platforms like Alexa. Many companies today use chatbots for customer service as a way to provide immediate resolutions to common issues.
Difference Between AI and Pattern-Based Chatbots
It’s important to understand why modern AI chatbots differ greatly from first-generation chatbots. The first chatbots adopted by companies were based on stringent rules and rigid decision trees and often led to frustrating user experiences. On the other hand are modern chatbots. Modern AI chatbots are more forgiving when it comes to following strict rules, enabling users to engage naturally in conversation. Here are the core differences between modern AI chatbots and basic bot technology:
Pattern-based chatbots (first-generation bots)
Pattern or rule-based chatbots are designed to follow conversation decision trees. These decision trees outline how chatbots respond based on keywords in the response from a user. Typically, the bot will ask a user a question and display a few responses in which a person can select from or it will identify a specific keyword in a user’s question. Based on a person’s input, the conversation moves forward on a specific path. With pattern-based bots, what a user says must explicitly match with how a bot was pre-trained in order for it to understand and move the conversation forward.
If a user’s statement matches a topic that has been pre-trained, the bot responds with the associated message. If no match is found, the chatbot may prompt a user to re-ask their question or respond in a different way. Alternatively, the bot might simply respond that it cannot answer the user’s question.
Pattern-based chatbots are very accurate when a user’s input matches what was pre-trained. However, these strict rules also prevent bots scaling since you need to account for all possible inputs. A person might ask “Where is my order?” by saying “my package isn’t here yet” or “Will my dress get here today?” Pre-training with all possible inputs (or “utterances”) is nearly impossible. In one example, Comcast found that its customers had asked the simple question “I want to see my bill” in 7,500 unique word and phrase combinations.
Learning-based AI chatbots (modern bots)
Learning or AI-based chatbots use machine learning technologies to understand, contextualize and predict to accurately respond to user inputs. The algorithms used in these chatbots are trained using historical data from real user responses and can be optimized with ongoing user feedback (reinforcement learning).
There are two subclasses of learning-based chatbots: generative chatbots and retrieval chatbots. Generative chatbots can dynamically create responses in real-time. Retrieval chatbots select from a pool of responses based on the person’s message to the bot.
AI-based chatbots leverage semantics to understand the context of what a person is saying. Therefore, these bots can engage more naturally in conversation, and respond to more inputs without being explicitly trained on every single way a person might phrase their question. Traditionally, these bots may not have been as accurate as pattern-based methods and used to take a long time to train. However, there are a few companies, like Netomi, that have built robust NLP engines that accurately understand user inputs up to 95% of the time, which means scaling and training are now exponentially easier and the end-user experience is much better than pattern-based bots.
AI chatbots use NLP engines and machine learning to interpret user inputs. This involves extracting user entities and determining user intents. These NLP methods are used widely in the technology industry, including for machine translation, sentiment analysis and user behavior analytics (UBA) in cybersecurity.
When chatbots use NLP, they are leveraging the following capabilities:
- Intent recognition—involves a semantic understanding of text-based and AI chatbots that leverage general syntactical and semantic knowledge which they learn from a large corpus of language data and business-specific training samples. Knowledge learned by AI chatbots from large data corpora helps for the expansion and transfer of vocabulary which helps to improve interpretations with fewer business-specific training samples.
- Extraction of entities—information that relates to a specific object or concept. For example, dates, places, times, descriptions, names, items, or numbers. These bits of data are the building blocks from which inputs are interpreted and defined.
- Dialogue management—Based on intent and entities, AI Chatbots use the next best action to trigger various actions required to capture appropriate details from users and business systems for meaningful resolution. AI chatbots learn user preferences in its long and short term memory to take contextually relevant smart actions.
- Expansion and transfer of vocabulary—algorithms can capture and refine vocabulary, including synonyms to improve interpretations. These refinements are tied to subsets of users to generate more natural responses and be passed to new bots.
As chatbots get smarter, the adoption rate by big brands and industry leaders grows exponentially. Below are five of the best chatbot examples:
WestJet, the only 3-peat winner of TripAdvisor’s Best Airline in Canada, has incorporated a chatbot named Juliet, to help serve its millions of monthly website visitors. With its chatbot “Juliet”, users can book travel plans, ask questions and get resolutions to common customer service questions.
When WestJet’s bot first got started, it could automatically resolve about 30% of all customer service tickets. In less than two years, that number has jumped to over 87%. Not only is this the highest rate of automated ticket resolution ever recorded – making WestJet’s Juliet the most powerful chatbot in the world – it also speaks to the sophistication of how true artificial intelligence can learn and get better over time. As a result, the WestJet customer service agents are able to work side-by-side with the AI bot and handle over 5X the normal load of customer support.
2. Charter Communications
Charter Communications, a top cable and phone service provider in the U.S. has incorporated a chatbot into its customer service operations. Before launching its bot, Charter’s customer support agents were answering around 200k live chats per month, a large portion of these for common use cases including forgotten passwords or usernames.
After bringing a chatbot into its customer support team, Charter Communications was able to handle 83% of chat tickets without human intervention. This significantly lightened their customer service load and resulted in a 300% increase in ROI.
Covergirl, a popular makeup brand, has taken a different approach. They are leveraging chatbots to engage with teens by providing product information and disseminating coupons. The Covergirl bot was designed to help the brand address the role that social media influencers play in young customer’s lives. Customers can interact with the bot to get product information and coupons for items.
As a result of the bot, Covergirl has seen social media engagement increase by a factor of 14. They have also experienced 91% positive sentiment ratings and a 51% click through on coupons.
Amtrak, a nationwide rail provider in the U.S., launched a chatbot named to provide support to its 375k daily website visitors. With the Amtrak chatbot, users can book travel, ask common questions and seek assistance modeled on the company’s best customer service representatives.
Currently, Amtrak’s bot is responding to around 5 million requests per year. This has led to a 25% increase in bookings and a 30% increase in revenue. Overall this has meant an 800% increase in ROI.
5. Johnson and Johnson
Johnson and Johnson owns a pregnancy and childcare resource called The BabyCentre UK. The company introduced a bot on Facebook Messenger to provide information and content to new or expecting moms. The bot uses targeting and personalization to deliver relevant information and to answer popular queries from users.
As a result of the bot, BabyCentre saw an 84% engagement rate and a 53% click through rate. This represented a 1,428% increase over email communications despite using the same personalization methods.
There are a few emerging trends that are propelling the sharp rise in the adoption of AI Chatbots. Take a look at these key chatbot trends:
- Personalization – personalization involves chatbots tailoring the interaction based on customer profile and behavior. For example, an AI bot could provide a hyper-relevant cross-sell recommendation by learning that a customer prefers certain brands or types of products. By incorporating customer experience personalization, chatbots respond on an individual level, providing more meaningful interactions.
- Voice recognition – voice recognition enables faster, hands-free interactions for users, making AI bots even more convenient. Examples of voice recognition can be found in the array of personal assistants, including Google Assistant, Siri and Alexa. Companies, including WestJet, are also launching skills on voice platforms to provide yet even more choice with how customers receive support.
- Machine learning operations (MLOps) – MLOps is a strategy used to automate and operationalize machine learning workflows. This strategy plays a role in chatbots by improving the speed and ease with which bots can be trained and improved. With such automation, bots are ready for market faster and can be more frequent, and easily, updated.
- Memory and context – many brands store customer information in customer relationship management systems (CRMs). When integrated into chatbots, CRMs can provide valuable information that enables chatbots to continue previous conversations with customers or look up specific details about the user. While this is often limited to profile details for privacy, chatbot engineers are working on ways to make queries more secure to enable broader interactions.
AI chatbots are continuously learning how to better interact with users, by expanding the software’s vocabulary, utterance training, entity graphs and contextual intelligence. By leveraging AI chatbots, companies can provide immediate assistance to customers, resulting in higher customer engagement and a positive user experience.
Interested in learning more about AI Chatbot technology can work for your team? Let’s chat!