What are chatbots and why use them?

One thing worth noting is that, contrary to a common belief, not all chatbots are equipped with artificial intelligence (AI). New generation chats are increasingly based on conversational AI techniques, including natural language processing. Why? Because this technique delivers a greater understanding of what the customer or user is asking, along with a capability to learn better and faster. And that means it will provide higher quality answers.

Turning to two different sources, Nielsen Norman Group (NNG) and IMB, shows us that several types of chatbots can be identified. On the NNG website, Raluca Budiu distinguishes two types:

• customer-service bots. Designed to take over some of the human work involved in customer interaction, chats of this kind could be described as intended for the initial contact with a customer or user, providing them with basic information about services or products;
• interaction bots. These are more complex in terms of interaction. They enable customers or users to order products or services and are most frequently distinguished by a more advanced interface, with buttons or ready-made answers. 

IBM, on the other hand, pinpoints more types of chatbot, although not from the perspective of the client or user, but in terms of technological advancement and the technology used. An article on the IBM website tells us that there are:

• chatbots based on menus or ready-made answers. Chats of this kind use predefined options, but when faced with more complex queries, they may be insufficient;
• rules-based chatbots or, to use Oracle’s name for them, task-oriented (declarative) chatbots. These can be described as interactive FAQs. The conversation designer programmes specifically defined combinations of questions and answers, allowing the chatbot to understand the user’s or customer’s queries and answer them. Rules-based chatbots have their limitations, though. They operate by searching for keywords or combinations of words, so they can learn quite quickly. However, they are unable to answer more complex or advanced questions;
• AI chatbots. Machine learning (ML) and natural language understanding or interpretation (NLU or NLI) mean that these chatbots are capable of understanding more complex queries. Importantly, they can remember the context of a conversation and, if they are not sure what the customer or user is asking, they can ask an additional question to help them improve their answer. Like the GPT chat ‘conversation’ that many people have encountered, AI chatbots can provide several answer options.

The superiority of these chats is also rooted in deep learning (DL), which allows the bot to acquire more knowledge. How does this happen? The bot collects questions and data and draws conclusions from them on the basis of the user’s questions and answers. However, remembering the context or original questions can prove to be problematic at times by creating a loop.

Another noteworthy point is that user experience (UX) and interface design (UI) should be treated as elements that are integral to creating chats.

In chatbot design, UX focuses on the user’s experience when talking to the bot. Here’s a list of the aspects that should be taken into account.

1. Mapping the user journey. This involves identifying user goals, such as getting information or reporting a problem, and then creating logical conversation flows that lead to the rapid resolution of the problem.
2. The naturalness of the interaction. Here, the focus is on designing conversations that mimic natural speech, while taking the simplicity of the language and local cultural contexts into account in order to create a better experience of interacting with the chatbot. UX designers and writers should also avoid overly technical or artificial answers.
3. Personalisation. A chat conversation is no exception when it comes to this crucial aspect of creating offers and products. UX designers need to ensure that the chatbot recognises the user, for instance on the basis of interaction history, and adapts the answers accordingly.
4. Error management. This involves both designing for potential scenarios where there are questions that the bot does not understand and planning emergency exits such as connecting with a human on the service desk staff.
5. Usability testing. Iterations after the design process are the norm and should not be omitted in the case of chatbots. The best solution is to conduct UX research with users to check if the bot meets their expectations. 
    

The fact that the visual aspect of a project not only attracts attention, but also makes it possible to enhance the impact of a well-designed UX is nothing new and is just as relevant to designing a chat, as the list below illustrates.

1. Visual elements. This includes designing basic elements, such as layout and colours. It is best to adapt them to the branding of the entire page, while also focusing on accessibility as regards appropriate contrast and letter size. The naturalness effect can be built on through designing an avatar or adding a gaming element by creating a funny creature, for instance.
2. Micro interactions. These include elements like introducing animations, such as three pulsating dots to suggest that the chatbot is processing or creating a response, for example. It is also important for the UX writer to formulate visual confirmations of actions, such as ‘Your message has been sent’.
3. Additional visual functions. If the type of chatbot permits, quick response selection buttons can be added, shortening the user’s writing time.
4. Responsiveness and accessibility. Without these two qualities, there is no way of having a well-made UI. The design should be created bearing a range of devices in mind and ensuring readability, even on small screens. The chat should also be designed for use by people with disabilities, including possibilities like operation via screen readers, for instance.

In chatbot design, UX and UI should not only provide functionality, but also encompass aesthetics. UX covers architecture and logic, while UI deals with visual appeal and readability. Coherently created and combined, they can make interacting with a bot an enjoyable experience.

This article began by saying that chatbots are computer programmes that allow interaction with users via text or, in the case of voice chatbots, speech, and are intended to resemble a conversation with a human. They operate on the basis of several technologies and processes.

NLP allows chatbots to understand, interpret and generate responses in a language similar to human speech. The process begins with the analysis of a textual or vocal input prompt from the user. The chatbot first identifies the user’s intent; are they, for instance, asking a question or requesting help? Next, it extracts key information from the input, such as dates, locations or specific phrases. It then generates a response. Tools such as Google Dialogflow or IBM Watson are used here, as are AI models, including GPT-4 or BERT.

This technology is already widely implemented in digital products and it plays an important role in chatbots. Thanks to machine learning and deep learning, these bots have the capability of recognising patterns in data, predicting intentions and providing increasingly better, more accurate answers. In generative chatbots such as the well-known ChatGPT, AI enables the creation of answers in real time on the basis of context and knowledge gained from training on huge data sets.

Chatbots are based on integration with databases and external systems. Using APIs, the more sophisticated ones can obtain information from CRM systems, knowledge bases or other applications. Simpler chatbots, such as rules-based types, can operate on the basis of pre-programmed rules and scenarios, which is useful for repetitive tasks such as handling FAQs.

The conversation engine, which is the central component that understands and generates answers to user queries, uses data from databases. The data is stored and managed by the backend. In simple terms, the conversation engine is the surface, and the backend is the technical layer that provides the data and response logic.

Chatbots typically perform simple tasks such as answering questions in an interactive type of FAQs and conducting basic conversations. They operate on basic NLP models or, if they are rules-based, on a set of rules, and are effective in simple scenarios but their capability is limited when it comes to understanding context or conducting more complex conversations.

Simpler chats are most often used for simple interactions, such as providing quick answers or reminders or handling tickets. A chat of this kind can respond quickly to a customer about the status of an order, for example, on an online store website. However, they are limited to what they have programmed with by their creators.

What are some of the other limitations of chatbots? One is their restricted capability in terms of understanding the context of a conversation. If it deviates from the predefined scenarios, the chatbot may lose the thread or provide an inadequate response. It is also important to remember that they do not usually learn on their own. On the contrary, their development requires their rules or scripts to be updated by their programmers and designers.

Chatbots typically operate on a single channel, such as a website, mobile app or communication platform like Messenger or WhatsApp, and their interface is simple and mainly text based.

Examples of applications

• FAQ support for online stores
• Booking appointments, for example at beauty salons
• Simple reminders and messages in applications

Virtual agents are more advanced systems that use technology, deep learning methods and language models. They can understand context, interpret complex questions, learn from interactions and conduct conversations on multiple topics at once and they offer a more personalised experience.

These tools provide a more comprehensive support, performing tasks of higher complexity, such as changing a reservation, and providing responses that outstrip pre-designed ones in intricacy. They also have the capability of conducting long-term and multi-threaded conversations and personalising responses in line with the user’s needs, even if the user changes the subject. They can do this because they are created using machine learning technology, thanks to which they can improve on the basis of analysing previous interactions.

Another major facet of virtual agents is their capability of operating on multiple channels simultaneously, integrating with CRM systems, emails, telephones and even augmented reality (AR). An additional bonus is the fact that their speech-to-text and text-to-speech processing systems enable them to handle written text and speech.

Examples of applications

• Bank assistants that advise customers on financial matters
• Customer service systems which analyse reports and send them onwards to the appropriate departments
• Technical support for diagnosing problems and guiding the user through the repair process step by step

This question could be considered rhetorical. Why? Because there are so many benefits that accrue from introducing chatbots to a business as they can significantly improve customer service, increase operational efficiency and reduce costs. 

Let’s take a brief look at some of the major advantages.

Chatbots are available 24/7, meaning that companies can provide non-stop client or customer support. Automating routine queries like answering frequently asked questions frees staff from having to spend time providing simple answers and allows them to focus on more complex issues by serving as a second point of contact with the customer when the chatbot is unable to meet the user’s expectations.

At the same time, automation in this area of business significantly reduces staffing requirements and reduces costs, since the expenses of maintaining a chatbot are much lower than staff salaries.

Artificial intelligence and integration with databases mean that chatbots have the capability of providing personalised answers, which increases user engagement. For example, a chatbot can remember a customer’s purchase history, offer recommendations tailored to their needs or adapt the language of communication to their preferences.

Chatbots can actively support sales processes by answering questions about products, advising customers on their choices and guiding them through the purchasing process. When it comes to marketing, chatbots can be tasked with collecting user data, conducting surveys and even engaging customers in interactive campaigns, all of which translates into higher conversions.

Customers value quick and effective answers to their questions. Chatbots eliminate long waiting times and can be designed to be interactive and user-friendly, which has a positive effect on brand perception.

Chatbots collect data on user interactions, enabling companies to understand customer needs better. Analysing these data can reveal common problems, questions and/or preferences, which often ties in with customer satisfaction.

Another major advantage is that chatbots provide support not only for customers, but also for staff, acting as in-house assistants. They can help with planning meetings, managing schedules, supplying quick answers to questions about procedures and providing data from company systems.

This article has introduced the concept of chatbots, explaining their functionality, types and benefits and set out the design considerations. It has looked at how chatbots simulate human conversation and examined their technological underpinnings, including AI and NLP, their applications in business and their use for customer interactions. Basic chatbots have been contrasted with advanced virtual agents and their role in increasing efficiency, personalising user experiences and reducing costs has been highlighted.

1. Raluca Budiu, The User Experience of Chatbots, Nielsen Norman Group;
2. Oracle, What is a chatbot?
3. IBM, What is a chatbot?
4. IBM, 5 types of chatbot and how to choose the right one for your business
5. Coursera, What Is a Chatbot? Definition, Types, and Examples

Translated from the Polish by Caryl Swift