Artificial Intelligence Chatbots Are Changing the Way You Do Business and May Impact Your Bottom Line

By Diana Ramos | March 14, 2018 (updated October 12, 2021)

Think back to the last time you chatted online with a customer service agent. Maybe you were complaining about receiving the wrong item in your order. It’s highly possible that the person on the other end trying to solve your problem wasn’t a person at all. You might have been speaking with an artificial intelligence chatbot, essentially a talking robot.

Artificial intelligence has made chatbots more lifelike than ever before, and they are becoming pervasive. Talking bots are taking pizza orders, reserving hotel rooms, and scheduling appointments. In short, these robots are all around us.

To maximize the ability of artificial intelligence (AI) chatbots to improve service, save money, and increase engagement, businesses and organizations need to understand how these programs work and what they can do. In this guide, you’ll get a crash course on talking bots, including the technology behind them, how they have transformed marketing and customer service, and how you can start putting them to work.

What Is a Chatbot?

Let’s start with the basics.

A chatbot is a computer program that imitates human conversation — spoken, written, or both. Chatbots conduct conversations with people, and developers typically hope that users will not realize they’re actually talking to a robot.

The term chatbot comes from “chatterbot,” a name coined by inventor Michael Mauldin in 1994. He created Julia, the first chatbot made with Verbot, a popular software program and development kit. Today, AI chatbots are also known by many other names: talkbot, bot, IM bot, intelligent chatbot, conversation bot, AI conversation bot, talking bot, interactive agent, artificial conversation entity, or virtual talk chatbot.

Artificial intelligence (in the form of natural-language processing, machine learning, and deep learning, which we will discuss later) makes it possible for chatbots to “learn” by discovering patterns in data. Without training, these chatbots can then apply the pattern to similar problems or slightly different questions. This ability gives them the “intelligence” to perform tasks, solve problems, and manage information without human intervention.

Of course, problems can and do arise, including instances in which chatbot limitations frustrate customers. Talk bots also raise some interesting ethical and philosophical questions that researchers have pondered from the start, and we’ll cover those later.

A History of Chatbots and Where They’re Headed

The history of chatbots dates to Joseph Weizenbaum's ELIZA program, which was released in 1966. Weizenbaum, a professor at the Massachusetts Institute of Technology (MIT), named the program after Eliza, a character in Pygmalion, a play about a Cockney girl who learns to speak and think like an upper-class lady. Weizenbaum’s computer program convinced many users that they were talking to a human being and not a machine at all.

By matching patterns and using scripts, ELIZA held “conversations” that gave humans the impression she understood them. The bot caused a stir because it was one of the first chatterbots to pass the so-called Turing Test. Legendary scientist Alan Turing proposed the test some 16 years earlier in an article entitled “Computing Machinery and Intelligence.” The test evaluates whether a machine can carry on a real-time conversation with a human being without the person being able to detect whether their conversation partner was human or machine.

The Turing Test looks for presence of mind, thought, and intelligence on the part of a “thinking” machine. The Loebner Prize, an ongoing competition, recognizes the most human-like machines using this test. ELIZA proved an interesting point: It was able to fool some users into believing that they were talking to a human being because they were quite willing to attribute human intelligence to answers that seemed halfway intelligent.


AI Chatbot Development

In the half century since ELIZA was released, chatbots have come a long way. The introduction of machine learning capabilities in bots has vastly improved the human-like quotient of their conversations. Most bots, though, still betray themselves as machines over short interactions. There are some exceptions, like AI chatbots Rose and Mitsuku, who are tougher to differentiate from real people, especially if the user doesn’t know they’re speaking to a bot.

But bots’ general inability to converse like humans doesn’t really bode ill for them because most bots are deployed in roles where they don’t need to converse like humans to be useful. Information-gathering bots don’t need to hold long conversations with you to be helpful; neither do banking bots, customer service bots, or bots taking purchase orders.

Landmarks in the Evolution of AI Chatbots

Name and Developer



“ELIZA,” created in 1966 at the MIT Artificial Intelligence Laboratory by Joseph Weizenbaum

This program was created to show how communication between humans and machines could never be more than superficial.

ELIZA used pattern matching to respond to prompts, but couldn’t contextualize and learn through interaction. Despite this — and much to its creator’s surprise — many of ELIZA’s users seemed to believe that ELIZA displayed real intelligence and understanding.

“PARRY,” implemented in 1972 by Kenneth Colby

PARRY simulated a person with paranoid schizophrenia.

In an early version of the Turing test, experienced psychiatrists were unable to distinguish reliably between actual human patients and PARRY.

“Julia,” developed by Michael Mauldin in the early 1990s

Julia was developed as a companion for text-based role-playing games, such as TinyMUD.

Julia participated in the first editions of the Loebner Prize competition and has existed on the internet in one form or another for nearly 30 years.

“A.L.I.C.E.,” composed by Richard Wallace and launched in 1995

This bot applied pattern-matching heuristics to input received from a human conversation partner in order to engage in conversation.

Although A.L.I.C.E. has not passed the Turing test, it is nevertheless a three-time winner of the Loebner Prize and is considered one of the strongest programs of its type.

“Cleverbot,” launched by Rollo Carpenter in 1997

Cleverbot learned from human input to conduct conversations with human users, instead of relying on a set of pre-programmed responses.

Cleverbot is the algorithm behind Eviebot, Boibot, PewDieBot, and Chimbot. Since its development, Cleverbot has participated in something like 8 billion conversational interactions, though it only draws from a tiny fraction of these in conversation. Eviebot, Boibot, PewDieBot, and Chimbot feature human, or chimpanzee, avatars.

“Kiyana,” created in 2005

Kiyana was developed to be a flirtatious female conversation partner.

Kiyana, who is apparently twenty years old, sings and likes cats. She’s popular with male users who seek an intimate, even romantic, chatbot conversation experience.

“Mitsuku,” developed by Steve Worswick in 2005.

Created to approximate an 18-year-old girl from Leeds, including the capacity to reason with specific objects, Mitsuku comes as close to a general-purpose conversation bot as any of the entries on this list.

A three-time winner of the Loebner Prize, Mitsuku claims to be the world’s best conversation chatbot and is popular with young users who discuss dating, being bullied, and their careers with the bot. Mitsuku runs on PandoraBots, which is one of the most powerful platforms for conversational AI chatbots.

“Rose,” created by Bruce Wilcox

Designed as a conversation bot, Rose poses as a 31-year-old security analyst and hacker from San Francisco.

Rose and an earlier rendition of the conversation bot named Rosette have placed in the Loebner Prize competition four times since 2011.

Right Click

This AI chatbot creates a website based on input it gathers from a customer.

The chatbot asks questions about the industry for which you’re creating a website. Perhaps its most impressive feature is its refusal to be pulled into a tangent by human users while gathering the information it needs to create a website.

Answer Bot

Designed to help businesses improve the quality of their customer relationships, this is a customizable customer service bot.

Answer bot, created by Zendesk, saves businesses having to develop customer service bots from scratch. If the bot is unable to answer a customer’s service request, the customer is directed to a human agent.


Replika creates a doppelganger of the user that learns and mimics speech patterns, moods, and mannerisms.

A personal chatbot, Replika was built by Eugenia Kuyda to memorialize a friend who died in an accident. It’s meant to serve more as a confidante than in any functional role and encourages users to talk their way through emotion or stress.


Poncho is a weather bot.

Poncho merges the dry role of weatherman with a large dose of sass, which makes him fun to talk to even if you’re not looking for weather updates.


This program keeps you company if you can’t fall or stay asleep.

Created in what might be a stroke of marketing genius by the mattress company Casper, Insomno only works at nighttime. While it’s not the best conversationalist, it does make a great curiosity for Casper.

Dr. A.I. and Melody by Baidu

Both apps fulfill medical functions, but in different ways. Dr. A.I. collects medical histories, systems, and body parameters, analyzes these for probable cause, and recommends a course of action. Melody, on the other hand, simply collects information to help doctors make diagnoses.

Both Dr. A.I. and Melody are interesting applications of AI in an industry where the human touch remains vital to the user’s experience and even to outcomes. It’s worth noting that both chatbots are designed primarily to complement or supplement the work of human doctors — not to replace it.

Obviously, the older bots on this list don’t really hold a candle to the new entries as far as appearances go; try comparing ELIZA’s clunky, old black-and-white interface with Mitsuku’s avatar and sleek text interface. But, when considering what’s under the hood, we have to ask ourselves how “intelligent” the bot is. We know that many of ELIZA’s early users were quite willing to attribute intelligence to the program, but what makes a bot “intelligent” today?

While bot technology has been around for over half a century, chatbots’ machine learning capabilities have been progressing by leaps and bounds recently. AI chatbots have been a focus of active research and investment by Silicon Valley, which has helped accelerate development.  

Given the advancements in artificial intelligence and the considerable amount of time most customers spend on messaging platforms — where many chatbots are deployed — it’s not an exaggeration to say that AI chatbots are becoming a necessity in some industries. Talking bots are a good choice for systematic, time-consuming tasks.

Technology research company, Gartner, has predicted that 85 percent of all customer interactions will be automated by 2020, and consultancy Servion believes that artificial intelligence will power 95 percent of all customer interactions by 2025.

Chatbots are surpassing the functionality of interactive voice response systems, which simply help users progress through a decision tree of potential options. AI chatbots constitute a different kind of conversation agent. Rather than being confined by a series of possible paths, they can generate relevant, contextually informed responses to a variety of prompts.

Artificial Intelligence Chatbots Provide Therapy, Grade Exams, and More

Talking bots are deployed through lots of channels, including on websites, as an app or part of one, and, perhaps most significantly, on messaging platforms. They’re used in applications ranging from digital commerce and banking to research, sales, and the development of brand awareness. They’re a good fit for business functions that have historically involved people performing time-consuming, repetitive tasks, such as handling customer services, listening to complaints, or taking orders, and they’re likely to reduce personnel costs while improving efficiency and standardization of functions and services.

Grand View Research projects the global chatbot market will reach $1.25 billion by 2025, with a compound annual growth rate of 24.3 percent. Moreover, 80 percent of business decision-maker respondents to a 2016 survey by Oracle said they already used chatbots or plan to use them by 2020.

Chatbots are already part of virtual assistants, such as Siri, Alexa, Cortana, and Google Assistant. In addition, so-called “social bots” promote issues, products, or candidates. People also use AI chatbots for entertainment. The Hello Barbie doll employs a chatbot created by a company named ToyTalk to talk back to her child owner. And, chatbots can even do less glamorous things, like encouraging people to get their flu shots.

At the cutting edge of the chatbot spectrum are bots performing more complex, nuanced functions in fields where the human touch remains important, such as law. Legal chatbots can  help you file appeals against traffic violations. In education, talking bots fulfill a number of functions from serving as course assistants to grading essays and eliciting student feedback. They act as conversation partners for people seeking companionship and even as talk therapists, such as the mental health chatbot Woebot. The ground rules for relationships between people and machines are still evolving, as the 2013 movie Her illustrated.

In marketing, the belief that AI and chatbots are the next big thing is widespread. Many  marketers believe artificial intelligence chatbots are revolutionizing business by providing better customer service, keeping customers engaged after sales, and adding “personality” to a company’s brand. AI chatbots can also help companies create more personalized experiences for customers, tailoring responses and content to the user’s questions and interests. Furthermore, talk bots are cheap, able to work 24/7, and never lose their tempers.

The hype notwithstanding, there’s a stark disconnect between those who foresee an AI marketing revolution by 2020 (80 percent of marketing executive respondents in a 2016 study by Demandbase), those who have a very confident understanding of AI (26 percent), and those who actually use AI at work (10 percent). Findings like these have led some experts to believe that AI and talk bots will not transform marketing to the extent advertised. Most skeptics also feel that AI chatbots will never have a truly human touch and that over-enthusiastic use of them can backfire with customers.

In fact, chatbots can even create distrust. The security firm Distil Networks says that about 40 percent of all bot traffic is malicious. Chat and messaging platforms, such as Yahoo Messenger, Windows Live Messenger, and AOL Instant Messenger, have gained reputations as being populated with bots that, at best, fill chat rooms with spam, and, at worst, try to entice people to reveal sensitive personal information. 

From Disney to Whole Foods: Which Companies Are Using AI Chatbots? lists more than 1,350 chatbots and virtual agents in use around the world. Many large brands have created fascinating bots that have hit it off with their customers and their target demographics.

Disney, for example, created an Officer Judy bot on Facebook Messenger to promote the 2016 movie Zootopia. Users helped Officer Judy solve cases, spending an average of 10 minutes talking to the bot. A large number of participants completed multiple cases. Fandango’s Facebook Messenger bot provides showtimes, theater locations, and links to movie trailers.

Pizza Hut’s chatbot takes orders, answers questions about food items, and tells customers about promotions. If you don’t like swiping through the extensive menu of the Starbucks app, you can place orders by talking to a chabot built into the app. And, the Whole Foods bot provides food recipes; you can prompt it to do this by using emojis.

1-800-Flowers, one of the first to deploy bots on Facebook Messenger, makes gift suggestions, takes orders, and provides shipping updates. H&M’s bot on Kik Messenger will style outfits for users using H&M products. Sephora’s bot, also on Kik, shows you how to do makeup, recommending Sephora products that you can buy online.

Lyft’s Facebook Messenger bot integration lets you book rides and share expected arrival times from within the Messenger app. Capital One has an Amazon Echo Skill that handles balance inquiries, pulls up transaction records, and enables customers to make payments via voice command.

What Is the Technology of Artificial Intelligence Chatbots?

To grasp the potential of chatbots, you need to understand how they work and what their capabilities are.

Chatbots come in two main varieties: those based on fixed rules and those based on machine learning. The former only respond to specified commands, and they only display a fixed level of smartness. Give this kind of bot a command that it doesn’t understand, and it won’t know what to do. It does not get any smarter with more interactions or information.

The second type of chatbot incorporates artificial intelligence, the ability to understand language, not just commands, and the capacity to learn. This technology has led to intelligent chatbots that can discover new patterns and get smarter as they encounter more situations.

Put simply, a chatbot’s job is to receive input data, interpret it, and translate it into a relevant output value. Upon receiving the input data, it must analyze and contextualize in order to determine the appropriate “reaction” to whatever prompt it has received.

A chatbot’s artificial intelligence has two components: machine learning and natural-language processing (NLP).


How an AI Chatbot Works

Machine learning is the ability of systems to learn from experience without human intervention and then use what they learn.

While, in theory, it might be possible to give a machine instructions for how to deal with every language prompt that it could conceivably face, it would take forever to write these instructions. By the time programmers were done, human language might have evolved, necessitating new programming. A machine programmed this way would understand language perfectly, but be impractical.

With machine learning, the computer system learns by being exposed to lots of examples (rather than by being given more rules). This approach is patterned on how the brain learns and is called neural networks.

Machine learning employs algorithms, which in their simplest form are sequences of instructions telling computers what to do. Algorithms can be combined and sequenced in complex ways. When a chatbot receives an input prompt, it must analyze the prompt and form context so that it can determine the desired output. As the chatbot is trained by having data input, it searches for patterns, which it can save for reference. This is the “learning” process.

To carry this a step further, deep learning is a type of machine learning that uses layered algorithms called an artificial neural network. Rather than task-specific algorithms, deep learning involves techniques in which the system discovers so-called representations in the data that allow it to make sense of raw data. Each layer of algorithms, in turn, comprises interconnected artificial neurons. The connections between these neurons are weighted by the prior learning patterns and events. The algorithms find patterns in vast quantities of data and infer how to respond to new data from these patterns. This approach is used in AI chatbots, where a predefined set of responses is not desirable or workable.

The chatbot’s algorithms are given a huge quantity of data that they explore for the model (or models) that enables them to convert to the output that the programmers tell them is correct. (For example, programmers might input different ways to ask a certain question and validate only appropriate answers.)

The data can thus be thought of as a set of examples that specify the correct outputs for a vast number of given inputs. The machine’s job is to extract and store the patterns from the data. The chatbot continues to “learn” — that is, extract and save patterns — with each subsequent input of data.

Natural-language processing (NLP) is the other component of a chatbot’s intelligence and refers to its analysis and synthesis of human languages. NLP makes use of predictive analytics, a combination of statistical, data mining, and data modeling techniques aimed at proactively generating information, without having to wait for a prompt from a human.

NLP gives a chatbot the ability to learn and mimic the styles and patterns of human conversation. It helps create the illusion that the chatbot is another human, not a robot. There are a number of developers of NLP tools that intelligent chatbots utilize. These include IBM Watson,, and

But, NLP faces challenges, such as enabling the chatbot to deduce how the human user is feeling, a capability called sentiment analysis that employs language analytics. Since speech and emotion are fraught with nuance, developers have struggled with how to teach chatbots to infer those cues from context.


Ben Virdee Chapman

“The quality of the training data that allows algorithms to 'learn' is key,” notes Ben Virdee-Chapman, Head of Product at, an AI company specializing in face recognition, including emotional state.

“The context in which emotion is being measured has a great impact on the application of the results. For example, it is now relatively simple to collect emotion data from text, audio, or visual data. Applying it in the context of stimuli is where the real challenge lies. And, written words or spoken words often come from the rational mind…so, understanding the differences between rational and irrational responses can help us weight the results,” says Virdee-Chapman.

While many researchers hold out hope for a truly intelligent virtual conversation agent, anyone today who has actually talked to a chatbot knows they’re all too fallible. Many betray their robot identities quite easily. As such, human supervision of bots remains an integral part of the chatbot ecosystem.

For example, bots work much better when they perform a small number of recognizable tasks, and they make for a better user experience when conversations resemble a set of commands or instructions. They can then hand off more sophisticated operations to human supervisors. And, if bots interact with the public, they must be able to deflect or ignore the nonsensical, abusive, discriminatory, or otherwise inappropriate conversations humans seem so tempted to conduct with them.

A More Detailed Breakdown of Artificial Intelligence in Chatbots

We have discussed in general how machine learning works in chatbots. At a more granular level, machine learning incorporates a number of different approaches.

  • Supervised Machine Learning Algorithms: These are most like the technique detailed above. They involve the use of a training dataset with inputs and outputs for which a machine creates an inferred predictive function that allows it to turn inputs into desired outputs after sufficient training. This type of algorithm also enables a machine to learn from mistakes — that is, when an input isn’t converted into the desired output.

  • Unsupervised Machine Learning Algorithms: These are similar, but use training data that is not classified or labeled. The idea here is not to generate outputs, but to describe hidden structures that might be present in unlabeled data, so the machine’s goal is to infer a function that can describe these hidden structures.

  • Reinforcement Machine Learning Algorithms: These use a behavior-reward learning method by which a machine learns which “behaviors” will earn it rewards — including delayed rewards — through trial and error. The goal is for a machine to be able to determine the most suitable behavior in a given context.

Roboticist Igor Mordatch doubts whether deep neural networks will ever truly succeed in teaching bots to mimic human language. He has sought to use reinforcement learning to empower bots to create and learn their own language. The bot languages that arise would be translated into human languages. The eventual goal, real conversation between humans and bots, will likely require a combination of techniques. Mordatch’s work is supported by OpenAI, a nonprofit artificial intelligence research firm founded in part by Tesla CEO Elon Musk that is dedicated to developing smart technologies that benefit humanity.

Natural-language processing comprises two related subfields: natural-language understanding (NLU) and natural-language generation (NLG).

Natural-language understanding is the complex, nuanced process of taking human-language inputs — with all their variability — and translating them into forms understandable by a machine. Natural-language generation is the reverse process: taking the computer’s generated output to a prompt and turning it into a form that is understandable by the human user.

In the fast-moving world of artificial intelligence, chatbots occupy a pretty low rung on the ladder. They’re classified as weak AI, which means they’re only supposed to learn specific, very narrow functions, like making reservations or locating books in a library. Strong AI, on the other hand, would boast intelligence that equals or surpasses that of human and be capable of a similar diversity of intellectual processes.

For example, the chatterbot A.L.I.C.E (Artificial Linguistic Internet Computer Entity), a weak AI entity, uses a markup language called AIML (Artificial Intelligence Markup Language) that is designed specifically for use as a conversation agent and has been incorporated into other Alicebots. Yet, it boasts no capacity for reasoning and relies solely on its ability to match patterns.

Some chatbots rank slightly higher on the AI strength scale. Jabberwacky, for example, learns how to construct responses and parse context based on user interactions, instead of relying on an unchanging database. Other newer chatbots combine the ability to learn with evolutionary algorithms that continually develop their communication abilities. Nevertheless, no chatbot is capable of carrying on the fluid, far-ranging conversations typical of human-to-human interactions, and chatbot development usually remains narrowly focused.

So What Exactly Is a Smart AI Chatbot?

So, what constitutes intelligence within the context of talking bots?

In layman’s terms, a chatbot’s level of intelligence is an umbrella term for its performance in several interconnected areas. An intelligent chatbot is typically one that’s capable of “learning” and using its experience to improve its performance.

Bots created for narrower functions are intelligent if they’re able to identify desirable outcomes (a challenging task), plan a multi-step process to achieve these outcomes, and then work autonomously to meet them. This sequence of activities goes by the name of the “sense-think-act” process. In addition, bots need to know how to ask questions that will enable them to gather the information required to undertake all the steps.

An intelligent chatbot also understands what a user wants and is prepared to meet users’ requests. Conversation bots are intelligent if they can handle different styles and topics of conversation with ease. So-called generative models of chatbots are able to come up with new responses rather than just retrieve from predefined responses. Generative models can take part in complex, longer conversations and deal with multiple questions from the customer. As the tasks chatbots perform become more complex, raising their IQ becomes more difficult.

How to Tell if You Are Talking to a Chatbot or a Person

Given the improvements in chatbot technology and how commonly we encounter bots, you may now be eager to spot the next virtual conversation agent you talk to.

Here are a few signs. Chatbots will often talk and respond at superhuman speeds, especially when processing mathematical or simple language inputs like “yes” or “no.” A chatbot’s speech output may be unnatural in context: They may speak in complete, formulaic sentence constructions or, conversely, be too informal. But, these characteristics aren’t a sure giveaway. What’s more of a tell is their habit of using repetitive outputs, like repeatedly telling you they don’t understand what you’re saying.

Chatbots may also give themselves away by trying a little too hard to sell you things, name-dropping products, and sending you links you didn’t ask for. Malicious chatbots may also request sensitive information without your indicating that you want to make a purchase.

And, lastly, the vast majority of chatbots will give themselves away if you just keep talking to them. Ask them what they think about current events or how they feel, or simply let a conversation stagnate by using filler words and monosyllabic responses, and the limitations of the chatbot will eventually become obvious to you.

What a Message Bot Is and How to Start Using Chatbots in Your Business

If you’re thinking about deploying chatbots for your business, chances are you’re looking at getting started with a conventional bot. A message bot’s primary function is exchanging messages with users, and the term should not be confused with bots that are built into messaging platforms like Facebook Messenger, where they may have functions besides exchanging messages.

Message bots are the classic chatbot. The term message bot is also not synonymous with conversation bot; the latter describes a specific type of message bot that is designed primarily to converse with users. Message bots use messaging interfaces to take on a variety of functions. They will likely replace some of the apps on your phone in the near future. You won’t need a pizza delivery or ride sharing app if you can just send a message instead.

Facebook Messenger is the biggest platform for chatbots. When Messenger was opened for developers to deploy in 2016, 30,000 bots were created in the first six months and 100,000 in the first year. Chatbots are also featured on WeChat and Kik, where they appear as individual contacts or as participants in group chats. Non-message bots on these platforms do things like play games, help you learn, send you curated content, and provide personal task assistance.

If you want to try out a few message bots, here are some popular ones:

  • Alaska Airlines’ Jenn: This bot answers typed questions about flights and the company.

  • Rare Carat’s Rocky: This bot answers your questions about diamonds and helps you search for a stone that’s within your budget and in line with your specifications.

  • Expedia’s Bot on Facebook: This app searches for hotel options and guides users to the site to make bookings.

  • Mila: This internal bot at coordinates communications when workers need to take a sick day.


Pamela Pavliscak

Pamela Pavliscak, Founder of technology innovation and insights firm Change Sciences, suggests starting simply when using chatbots. “At best, a chatbot can answer a few questions about the weather or schedule a meeting. At worst, a chatbot can feel like a customer service phone tree. For now, narrow the focus on supporting a repetitive, or dreaded,  task  first,” she says.

A good place to get started with AI chatbots is by deploying them on social media platforms and messaging apps where potential customers spend time. Messaging apps have surpassed social networking apps in terms of monthly active users, and both the number of sessions and the time spent on social and messaging apps have skyrocketed in the last few years. So, if you want chatbots to be effective, they must engage with your customers where your customers hang out.


Jonathan Duarte

“When talking with clients, I suggest they consider a chatbot as a new employee,” says Jonathan Duarte, CEO of chatbot development company GoHire. “If that employee is going to help answer customer questions, they would need to be trained on what the most common questions are and where to find the answers. You wouldn't put them on the phone on day one without training. Same thing with a chatbot. You need to put them through product training, so they know what the user's concerns are,” adds Duarte.

Mitul Makadia, Founder of bot development firm Maruti Techlabs, says that chatbots can help businesses reduce costs, increase efficiency, automate internal functions, and scale customer service in retention.


Mitul Makadia

“In order to get started, you can work on identifying certain areas within your business that would benefit the most. For example, if you happen to be in e-commerce, you can set up a Facebook Messenger bot to showcase your product and have interactive conversations with your customers. Best part? You don’t need an exclusive app for your products/store — the bot helps you efficiently bypass that long-winded process,” Makadia notes.

Among other strong use cases, Makadia points to having chatbots handle lead generation on your website or notify you when inventory needs restocking. Chatbots can make it easier for you to offer product suggestions and push personalized deals to customers. They’re also capable of accepting payments, so the customer engagement can translate into real bottom-line gains.

Duarte of GoHire says that customer support and sales are now the primary use cases for chatbots, but that traction in HR has been strong too.

“Chatbots can have immediate effects on qualifying sales prospects, reducing the cost of having high-paid sales teams who manually respond to non-sales qualified prospects. In customer support, it's a volume-efficiency solution. Chatbots can scale significantly for level one support issues,” Duarte points out.

“In the recruiting and human resources departments, chatbots are being used to engage and acquire candidates, while also pre-screening them for required skills. We've seen clients generate 500 percent increases in candidate apply rates using chatbots because of the engaging process, compared to job descriptions and web-page application processes,” Duarte continues.

Chatbots are also a great way to solicit customer feedback, which is an expensive, time-consuming process if done manually. Having a machine collect feedback expedites the analysis of feedback. Plus, it can help build brand image. Chatbots can help streamline and standardize customer service and improve the help-seeking experience for customers, meaning less frustration all around.

Chatbots can also be used internally to automate business headaches, like coordinating schedules for meetings. Chatbots stick with the task and, because they have a friendly demeanor, can elicit responses more quickly.

But chatbots — especially the ones designed to be conversational — can wear out their welcome quickly if they don’t display the ability to contextualize or respond in a human-like fashion. We know from ELIZA that unsuspecting chatbot users are willing to ascribe human intelligence to chatbots who display only a semblance of the same, but a bot that is repetitive, not satisfyingly responsive, or out of date will become a liability.

To avoid these problems, make sure information databases that the bot is drawing from are kept updated, and use a bot designed with the ability to learn. Chatbots that can accurately gauge human emotion through NLP and other user behaviors are still largely a work in progress, but, when they are perfected, they’re going to be a big step forward for user engagement.

While chatbots are designed to be human-like in their interactions, actual human beings who know they’re interacting with a machine and not a person don’t usually return the favor, especially if they’re using a chatbot anonymously. Your chatbot is likely to have all sorts of odd prompts thrown at it, from the vague to the tangential to the obscene. That’s why an effective chatbot needs the ability to steer users gently but firmly in the direction of what it’s meant to accomplish.

And, remember that with customer service bots, customers ultimately care about whether they’re getting the service they came for. The novelty of talking to a robot, no matter how cute it is, is no substitute for good service. Put the customer at the center of your customer service bot strategy by making sure that the bot is actually an improvement on your existing service experience. Don’t keep people waiting to speak to a bot that turns out to be unhelpful.

How to Build AI Chatbots

Traditional bot development is a three-stage process that includes design, build, and analytics. Designing involves defining what the chatbot will look like and how it will interact with users. Building is creating the brain of the chatbot, i.e., its ability to understand input and generate output. Analytics monitor how the chatbot performs and offer guidance on improving the way it works.

Chatbot development platforms, or chatbot builders, simplify the chatbot-building process. People unfamiliar with coding can create a chatbot using simple drag-and-drop. Cloud-based chatbot development platforms, such as Oracle Cloud and IBM Watson, provide NLP processing and AI capabilities that businesses can leverage. Other platforms include Chatfuel and Pandorabots. This article on 10 of the most popular tools can help you choose.

If you want something more advanced, you’ll need to program it, and this requires picking a programming language. The programming language has nothing to do with the language in which the bot converses with users; it’s the language in which the bot’s inner workings are coded.   

Platforms that host chatbots, such as Facebook, Telegram, and Slack, support most popular languages, including Python, PH, and Java. But, since the choice of language has ramifications for the bot, here are a few basic questions you should ask.

  • What language is the developer most comfortable using? The obvious answer is to go with whichever language the developer prefers, but for new bot programmers, Python is a good choice: It’s versatile and easy to pick up.

  • What sort of AI capabilities do you want built into the bot? Python’s a winner here too, given that it’s one of the most widely used languages in the field of Artificial Intelligence. Natural Language Toolkit (NLTK), called the “grandfather of NLP integration,” was written in Python, and the language features a wide selection of libraries for machine learning algorithms.

  • How fast do you want the bot to function? Java and C++ execute faster than Python, but you’ll have to ask whether the increased speed is a worthwhile tradeoff for the lesser intelligence you’ll be able to build into the bot.

Tony Lucas, Co-Founder of Converse AI (acquired by Smartsheet in January 2018), recommends companies deploy chatbots quickly and then refine them based on how they perform. “The best piece of advice I give customers is to build quickly and ship a first, minimal version to a small group of users, and then iterate, iterate, iterate,” he says.


Tony Lucas

“There is not enough data today for specific companies and use cases, about how their users will interact, talk, or engage with a bot. So, the key thing is to use the product development process to help this by shipping early and often and not spending months or years before deploying a first version to receive feedback from customers. Trust me, they will engage with bots in ways that you couldn’t possibly expect - both good and bad.” Lucas emphasizes.

Navigating Concerns and Ethical Problems with AI Chatbots

Using talk bots raises some tricky choices and ethical dilemmas, especially as chatbots become harder to distinguish from humans.

For example, is it ok to create machines that attempt to pass themselves off as human, so they can communicate with actual human beings? While some bots are straightforward about being machines, others are not. They’re good enough at hiding their nature to raise questions about how ethical it is to use chatbots without disclosing it.

Honesty is vital, especially when the bot is handling sensitive information, contends Pavliscak of Change Sciences. “A big ethical concern is transparency. Do people know they are talking to a chatbot? Or, do they think they are talking to a human? Some chatbots are a blend of both. Ethically, people have a right to know,” she says.

Data privacy and what a chatbot does with the information it collects are also hot topics. If data gathered by a chatbot is shared internally, customers need to know before they start talking to the bot. Given that users may be tempted into less-than-professional communications when they know they’re interacting with bots, how many users would be comfortable knowing that transcripts of conversations with chatbots might be read by real people?

Chatbots that are malicious by design add a potentially criminal dimension to these agents. A chatbot designed to misrepresent its identity, gather and use information without consent, or hurt the user is obviously unethical and possibly illegal. As they learn, conversation bots cannot afford to pick up vulgarities and obscenities or offer discriminatory or violent views. Remember Microsoft’s Tay, which had to be taken offline (within 24 hours) after it was taught to espouse virulent racism.

Another chatbot that currently falls into a privacy-related gray area is Woebot, a talk therapy chatbot created by Stanford University psychologists and AI experts. Using chat conversations, mood tracking, videos, and word games, Woebot is designed to help people manage mental health, and scientists who examined users’ interactions with Woebot found that the bot reduced interpersonal anxiety, perhaps since the users weren’t as afraid of being judged by a machine as by another human.

Woebot users self-reported reduced depression and anxiety symptoms after a short stint using the bot. But privacy activists have been critical of Woebot because it works through Facebook Messenger, and Facebook owns the content of all conversations with the bot and knows the human user’s real identity.

Other shortcomings hamper chatbots. For one, they can’t gauge the context of questions with any great degree of accuracy. Humans are capable of contextualizing conversation almost instantaneously, but chatbots can’t be relied upon to “remember” details provided even a few minutes ago. Their poor grasp of emotional cues also means they don’t know that it’s time to escalate an issue to a human agent.

Some bots aren’t intelligent at all in that they’re not fully responsive to interactions. Bots like these are basically dressed-up decision trees, and trying to find options outside of the tree results in the user quickly hitting a dead end.

Moreover, in certain fields, such as education and medicine, the spread of chatbots is limited by concerns about the “dehumanization” of interactions. Many people aren’t comfortable speaking to a bot that can’t evince real empathy or sympathy, and, in some cultures, having to deal with a machine instead of a human being might even be considered insulting. Even if bots gain greater sentiment-reading capacities, misgivings are likely to remain.

These concerns are magnified as artificial intelligence improves. In his 2013 book Our Final Invention: Artificial Intelligence and the End of the Human Era, author James Barrat explores the risks of human or superhuman artificial intelligence, saying that it would be difficult to control.

On a more mundane level, researchers are struggling to help bots display the semblance of a personality to improve their conversations, which lack a human touch.

The Future of Artificial Intelligence Chatbots

These issues and others are spurring technologists’ work to perfect AI chatbots. They face important milestones in development.

“The industry is in its very early stages, and there is a lot of maturing still to do,” says Lucas of Converse AI. “Solid use cases and ROI will get demonstrated, companies will become more comfortable with using chatbots, both internally and externally, and the technology underpinning bots will also mature, making it easier to build more powerful, flexible solutions,” he insists.

Lucas sees three main priorities:

  • Storytelling: He notes that “Chatbots deliberately have a limited, but structured interface. This makes it easy to build them, but means it’s more important to have someone who understands how to tell a brand story or build an interactive dialogue than it is to have a traditional UI/UX person. And it also means there is a risk that bots are built that are genuinely useful, but not intuitive or enjoyable to use.”

  • Suitable Use Cases: Lucas says that after a few years of experimentation in 2016 and 2017 in which bots were built with many unhelpful use cases, the industry’s focus is on “existing, known problems, how people work today, how they can work better, and the future of work. This means it’s significantly easier to actually address sensible issues and find value in chat bots.”

  • Discovery Versus Spread: How do users discover what a chat bot can do, and, just as importantly, what it can’t? “Narrowly scoped bots with clear capabilities and boundaries are where success is being seen today, and I believe that will continue. But, there is also a need to ensure that users can easily discover new capabilities that are available,” Lucas contends.

So where do chatbots go from here?

GoHire’s Duarte says that chatbots will become more accessible to smaller organizations as tools to build and maintain them become less expensive.

The user experience and decreasing user frustration are major areas of research for developers. They are seeking to combine voice with user-friendly graphical conversation interfaces. This pairing would integrate natural-language input, whether in text or voice format,  with graphical elements for a seamless and frictionless interaction. This advancement will make chatbots more natural to speak to, more accessible for the non-tech-savvy, and perhaps less frustrating to deal with.

Pavliscak anticipates progress on helping chatbots pick up on emotional cues. “Emotion artificial intelligence that detects physical traces of emotion can help identify when customers are getting stressed. Eventually that will help chatbots communicate with greater empathy,” she says.

When it comes to efforts to make chatbots seem a little bit more human, Facebook has been trying to teach chatbots how to make small talk that’s not filled with strange, illogical leaps; this, the experts believe, will help make them more personable.

Another hot area in chatbot development is the enhancement of NLU capabilities. This should better allow chatbots to contextualize and understand what users want.

The Future of Work with Automated Processes in Smartsheet

Empower your people to go above and beyond with a flexible platform designed to match the needs of your team — and adapt as those needs change. 

The Smartsheet platform makes it easy to plan, capture, manage, and report on work from anywhere, helping your team be more effective and get more done. Report on key metrics and get real-time visibility into work as it happens with roll-up reports, dashboards, and automated workflows built to keep your team connected and informed. 

When teams have clarity into the work getting done, there’s no telling how much more they can accomplish in the same amount of time. Try Smartsheet for free, today.



Discover why over 74,000 brands trust Smartsheet to get work done.

Try Smartsheet for Free Get a Free Smartsheet Demo