Originally published on the Infopulse blog.
“Talk to a robot” is no longer a line from a science fiction novel. At present, AI has finally reached a state when we can effortlessly speak to the “technology” — our smartphones, websites and even our cars. This revolution in human interactions with technology has massive implications for businesses across different industries. Clearly, progressive companies are already “gearing up” their cognitive powers as the chatbot market is expected to witness a 24% growth rate in the next six years and reach $1.25 billion by 2025.
What exactly does that multi-million dollar opportunity entail? Following is the breakdown of the current state of the conversational AI technology, along with use cases of chatbots and the benefits businesses can expect to obtain.
Trending AI Articles:
1. Making a Simple Neural Network
2. From Perceptron to Deep Neural Nets
3. Neural networks for solving differential equations
4. Turn your Raspberry Pi into homemade Google Home
What is Conversational AI?
Conversational AI is a set of technologies, enabling the software to understand and to naturally enter in conversations with people, using either spoken or written language. Siri and Google Assistant — the trusted friends of many — are two prime examples of voice conversational AI in action.
Chatbots are another type of conversational assistants, interacting with users through text. You may have dealt with these “conversationalists” either on social media or through on-site chats. In the course of this post, we’ll use the term “chatbot” to refer to all the cognitive applications that enable conversations between humans and machines.
In the past few years, chatbots became a hyped technology. The proponents say that chatbots may replace 99% of apps in a few years time. The skeptics say that this technology has not lived up to the hype.
In reality, both sides are correct. Indeed, by 2022 chatbots are said to save businesses more than $8 billion per year in customer support costs alone. In the banking sector, the success rate of interacting with bots is expected to reach 90% — meaning that just 10% of customer queries will land in front of human agents. At the same time, there are still certain limitations to what conversational AI can accomplish. We are yet to see autonomous AI robots greeting guests at hotels or conversing with shoppers in-store.
To understand the full spectrum of current chatbot possibilities, let’s take a look at the key technologies powering conversational AI. After all, advancements in those areas will ultimately enable conversational bots to handle more complex queries and tasks.
Key Technologies Underpinning Conversational AI
Machine Learning (ML) is a method of training systems to learn automatically and improve performance over time, without being explicitly programmed to do so. In the context of conversational AI, various machine learning methods allow developers to train algorithms to capture data from conversations, operationalize it and turn it into some action e.g. a booked appointment.
Deep Learning (DL) is a subdivision of machine learning. Deep learning approach assumes a lesser extent of human involvement in the algorithm training process. Instead, the systems are programmed to learn by example and determine whether their output is accurate or not on their own. Artificial neural networks are the baseline method of deep learning, enabling machines to operationalize available data with higher accuracy and draw better conclusions. ANNs are one of the key technologies enabling the “intelligence” component of a cognitive app. Such algorithms are also better suited to recognize and reproduce human speech.
Artificial neural networks are now used by Google Translator to deliver better translations, including real-time voice ones. LinkedIn uses neural networks to detect spam or abusive content in users’ feeds. Zalando uses a neural network to enhance on-site product search for misspelled queries.
Natural Language Processing (NLP) stands for a set of programming techniques powering the analysis and synthesis of natural language and speech. NLP allows AI algorithm to better understand the context and meaning of users’ requests. By relying on NLP, chatbots can do the following:
- Process search queries such as “show me the latest products in my cart”.
- Interpret the basic information from the conversation e.g. the topic being discussed, whether the response was negative or positive and so on.
- Correctly understand and render synonyms, jargon and common abbreviations.
- Format the information they have collected in natural language. For instance, spell out the weather forecast for the requested destination, instead of throwing in some numbers.
Machine learning, deep learning, and NLP are the standard triumvirate powering the majority of conversational AI apps and chatbots, now available on the market, as well as end-to-end conversational AI platforms for enterprises:
Why Businesses Rush To Implement Chatbots and AI Assistants
AI chatbots have already won over younger consumers. Millennials, in particular, have a stronger affinity towards these “clever conversationalists”. Per a recent survey, over half of them (55%) stated that their perception of business improved after an interaction with the company’s assistant. In particular, a brand that uses chatbots is viewed as more efficient (47%), innovative (40%) and helpful (36%) by consumers, according to another survey conducted by PSFK.
The most sought after benefits customers expect to receive by engaging with chatbots are as follows:
By employing intelligent bots businesses, in turn, can effortlessly meet these customers’ demands and solve some of their operational challenges as well.
1. Streamlined Customer Support.
Customer experience is the new battlefield for businesses: 89% of companies are already competing on CX. Customer Care and Support are the two main components of superior CX. And yet, they also tend to be the most problematic areas for businesses. With high turnover rates, low morale and constantly increasing customer expectations about response time, few businesses manage to excel in these areas, especially when their CS teams are small.
The new generation of AI bots is taking over the customer service sector. Unlike the earlier predecessors, capable to handle just a few simple FAQ queries, the current state of conversational AI enables you to employ truly smart and helpful, albeit artificial, CS team players.
KLM, one of the busiest airlines in Europe, has recognized the benefits of chatbots for customer support early on. Since 2016, the company has been working on its AI learning model, trained on years of historical data such as routing history, metadata history, chat logs and CRM data.
Their AI chatbot is available on Facebook, WhatsApp, Twitter and integrates with Google Assistant, proactively helping customers round the clock. Every customer support interaction is now augmented and/or largely managed by AI. The algorithm suggests human agents the best possible answers and actions based on the conversation context, reducing the number of mundane repetitive actions.
The best part? Their AI conversation engine improves over time and witnesses an accuracy growth rate of approx 2.2% or 55% between Jan 2018 and July 2018. Beyond that, KLM managed to achieve the following positive results:
- 50% of all customer interactions on social media are now supported by AI.
- Over 5 million clicks have been saved in the last five years.
- 100% increase in case volume.
- Customer interactions speed up by 50%, meaning that more staff is now available for more complex cases.
2. Conversational Commerce.
Chatbots enable a completely new sales channel for brands. AI assistants can now manage the entire customer journey from awareness to consideration to purchase and post-sales experience management. They provide an entire shopping experience inside a single conversation, happening within a mobile app or on a social media platform.
Think of travel assistants:
Infopulse team has recently developed Chat&Fly — a multilingual travel assistant that helps users find and purchase the best airfare. Chat&Fly will make casual chit-chat (this feature is currently available only in English) while processing the travel request and prompt the user to check out the weather forecast at their destination as a quick after-service delight.
How’s Chat&Fly different from other AI travel assistants?
- It’s truly intelligent. We applied Deep Neural Networks to power-up Chat&Fly conversational skills and overall user experience.
- To accurately capture user request and intent, we relied on innovative Microsoft technology stack (Bot Service and LUIS) that delivers natural language processing capabilities.
- Chat&Fly is multilingual and can connect with customers all around the globe.
- It is integrated into Skype, Facebook Messenger and can also operate on other popular messaging platforms.
Contact us for more information or to request a Chat&Fly demo.
Chatbot sales funnels now generate the same and sometimes even higher conversion rates than other sales channels including mobile and website. They alleviate the common barriers between consumers and buying products online: multi-page order forms, payment processing, poor on-site search experience. Chatbots bring the online shopping experience closer to the physical realm: instead of pressing buttons, you engage in a friendly chit-chat and receive personalized suggestions from an assistant who understands what you need. Clearly, such an approach works as the following examples demonstrate.
Tobi, Vodafone’s chatbot, tasked to handle end-to-end SIM card sales, has proved to be a tremendously profitable sales channel for the company:
- Tobi conversion rate is 100% higher compared to the company’s website.
- Abandonment rates are significantly low and the transaction time is 50% less compared to the website.
- Average usability score is 90 out of 100 — the highest of any other asset.
- Touchpoint NPS >65. The number is getting close to contact center levels.
Julie, an online chatbot, deployed by The National Railroad Passenger Corporation (Amtrak) in the US, was trained to handle both customer support queries and facilitate the sales process. Per the latest report, the reported post-adoption benefits were as follows:
- A 25% increase in bookings.
- 30% more revenue (monthly average) generated per booking.
- 50% year-over-year growth in users’ engagement with Julie.
- Nearly $1 million in customer service email costs saved per year.
3. Intuitive Onboarding and Improved Retention.
Better onboarding experience for new customers means less churn in the future. And better retention and engagement mechanisms have a direct impact on your bottom lines. Customer.io estimated that a 1% improvement in customer acquisition boosts your revenues by approx. 3.3%. Improving retention by 1% boosts your bottom line by 7%. No wonder, as existing customers are 50% more likely to try a new product/service and they are going to spend on average 31% more than a new buyer.
Expensify — one of the leading business expense management tools — recently deployed an AI assistant app to assist new users with onboarding and product discovery. Concierge bot operates on the company’s website, mobile app and will soon become available in Slack. It helps users with the setup processes and troubleshoots common payment issues e.g. verifies the connection of users’ credit cards and proactively suggests fixes before anything breaks.
Concierge bot can also look up real-time pricing on flights, hotel, and other travel services and notify users about the best deals. Post-adoption, the company reported a 75% reduction in banking problems and significant growth in free trials for the company’s product.
AI chat assistants can be as well deployed in the SaaS niche to activate customers at different stages:
- Week 1: Short-Term Retention: An AI assistant can prompt the user to discover new product features by either suggesting walkthroughs or try new actions.
- Week 1 to 4: Mid-Term Retention: A chatbot can help establish a consistent pattern of usage by proactively helping customers accomplish new tasks, sending relevant deals, notifications, and reminders.
- Week 4: Long-Term Retention: Proactive assistance offered by chatbots can further enhance the value of your core services. As a result, customers will view your product as an indispensable tool and continuously engage with it.
4. Co-Automated Workflows.
On a daily basis, your teams are dealing with the mundane repetitive tasks. By incorporating AI and chatbot assistance into everyday routines, businesses can streamline the low-value processes and decrease productivity loss.
AI platforms can be now set up and customized to help your teams with lead scoring, data collection, follow-ups and a number of common admin tasks. Perhaps, you are already familiar with Slack bots — the simple algorithms automating various tasks in the app such as sending reminders, scheduling tasks and so on. Thanks to recent advances in ML and NLP, businesses can now develop more intelligent algorithms, capable of assisting with more complex tasks — for instance, streamline the hiring process.
Mya conversational AI platform helps HR teams to prescreen a wide pool of candidates through various channels: Facebook, email, Skype, SMS or chat. The assistant asks a series of predefined questions, captures responses, answers candidate’s questions and delivers tips and progress updates to potential hires.
Recruiters, in turn, receive actionable insights from Mya — a list of candidates for the position ranked from the most to the least qualified, based on a variety of collected metrics. According to Mya developers, the platform can automate up to 75% of the qualifying and engagement process. HR teams who tried Mya reported a 38% increase in efficiency, as well as a 150% increase in candidate engagement.
Up to date, some of the best AI chatbots on the market are those automating the common internal workflows such as:
- Scheduling and organizing meetings: Amy AI, Clara, or Microsoft’s Cortana.
- Assisting during the sales process: Conversica (conversational prompts for engaging with leads), Clari (sales forecasts, deal progress reports and daily analytics), Olono (real-time guided selling for B2B teams).
- Managing legal documents: LISA (helps to create standard legal documents such as NDAs), Ailira (AI-assisted database research for legal and tax documents), ROSS (AI-powered legal research platform).
At present state conversational artificial intelligence can be effectively applied to enhance any arduous process or business task into one that allows all parties to become more productive and engaged. On a less positive note, cognitive services adoption will require careful planning and perhaps even operational and technological changes within your organization.
The Common Chatbot Development Pitfalls and How to Avoid Them
The concept of automating low-value processes and scaling one-to-one conversation using AI appeals to a lot of businesses. However, designing effective conversation workflows and anticipating human behavior is a complex task. Despite the best intentions, sometimes chatbots fail and deliver less than satisfactory user experience.
To design and develop a truly delightful and efficient assistant, businesses need to carefully consider several facets:
- Tech stack and integration with current systems;
- Implementation and usage strategy;
- Conversational flows and overall UX.
1. User Value and Predicted Use Cases are Not Fully Understood.
The majority of users today may have a positive attitude towards conversational AI platforms, but they have a low tolerance for annoying, unhelpful and glitchy chatbots. 32% of recently surveyed consumers stated that their biggest pain point with chatbot interactions is when a bot gets stuck in the conversation. 59% also express their frustration with the fact that they need to repeat the same information to a human assistant when the “stuck” bot redirects them to one. Furthermore, some users just don’t find the chatbot to be helpful at all.
Oftentimes, these scenarios happen when development teams deploy something that users don’t actually need. The goal of a chatbot is to reduce friction in the user experience. Yet, many chatbots increase the number of steps or time required to complete an action. To avoid this trap, ponder over the following:
- What is the exact value delivered to the end user/customer?
- What pain point does it solve? What type of friction will it remove from the current process?
- Is it a simple process you are planning to automate or are users expecting something more sophisticated, that will be interactive, and learn over time?
2. Assuming That One Chatbot Means One Use Case.
Powering your chatbot with ML/DL means that your system can learn over time. Unlike the simplistic chatbots deployed on Facebook Messenger, AI-powered apps can handle more than one task well enough and drive more engagement.
Instead of developing multiple bots to perform similar actions for a different group of users (e.g. mobile and television customers), or operating on different platforms (Messenger, on-site chat, Slack), consider creating a unified platform that will handle all the use cases at once. The current artificial intelligence chatbot algorithms already allow accomplishing that.
3. Overlooking the Value of Data Gathered by the Bot.
Chatbots are a powerful tool for customer insights mining as all conversations assume sharing information. Every time a user engages with your AI assistant, they are open to giving away an enormous amount of valuable information you could apply towards product development, personalization, and marketing.
However, businesses make several mistakes here:
- They forget to create an effective process for capturing that information and sending it over for further analysis.
- Flaws in conversation design can result in the bot asking the wrong questions and collecting unnecessary information.
- GDPR compliance presents certain challenges when it comes to customer data collection via this avenue.
4. Neglecting the Context in Conversation Flow Design.
You cannot yet talk to artificial intelligence in the same fashion as to another person. AI assistants are not proficient in capturing sarcasm, conversational undertones or contextual cues. NLP has allowed us to make a significant leap in language processing, but AI cannot predict with 100% accuracy what a user will say next or how they’ll behave when no boundaries are present.
Chatbot conversations still require a certain structure — prompts, buttons, etc. — to ensure that users are indeed getting the outcome they expect. Free flow conversations should be carefully curated and structured by designers to ensure that your assistant is asking the right questions and receives the necessary data for processing a request. Hence, consider the following:
- What are the logical conversation path(s) for users to follow?
- Is the user expected input clear enough at every step of the conversation?
- Did you notify the user about the chatbot capability limits?
Designing conversational interfaces represents a significant shift in how we are used to thinking about online interactions. To create a delightful conversation experience, brands should rely on professional designers to assist them with conversational UI content creation and optimization.
Schedule a discovery session with Infopulse AI team to learn more about conversational design and development. We’d be delighted to assist you with discovering the ultimate set of use cases for your chatbot; advise you on the technology stack and work out effective conversation flows.