After Facebook announced its new chatbot platform in 2016, Series B funding quintupled and early-stage development funding rose 129% for conversational interfaces. Since then, hundreds of thousands of conversational experiences have emerged across platforms including but these aren’t just bots; they’re interfaces.
In this innovator guide we answer questions like how to build a chatbot? how to introduce a chatbot into your business? We cover marketing automation, customer service, artificial intelligence and machine learning with chatbots. We also take a look at how chatbots and conversational interface app development is fundamentally changing the way businesses interact, both with customers as well as each other.
Chatbots are simply computer programs that can hold a conversation. Once known as “chatterbots,” they’ve existed since the early sixties, thanks to industry pioneers,MIT scientist Joseph Weizenbaum, and psychiatrist Kenneth Colby, who identified psychological patterns in people’s discussions and programmed bots to maintain an illusion of natural conversations.
Since then, there have been many breakthroughs in natural language processing (NLP), a computer science field studying ways to extract structured data from human language. NLP and AI advances mean chatbots now have productive conversations. How productive? Conversational apps have come a long way since the ‘60s.
What once felt like entering techie commands into a chat window, now feels remarkably like a real discussion between people, especially when follow-up questions are possible and context is preserved across messages and sessions. Speech recognition adds another dimension: many bots verbally chat. Virtual assistants like the Google Home, Apple’s Siri, Microsoft’s Cortana and Amazon’s Alexa actually talk aloud , and listen when we talk back, whether it’s to order groceries, ask questions, or turn on the lights.
Today’s services have application programmer interfaces (APIs), software layers that any third-party developer can use to quickly build conversational experiences on top of popular platforms like Facebook. Entrepreneurs, enterprises, and startups have all been quick to grab this opportunity, finding creative ways to chat up new customers and assist current ones.
Chat has moved from the conversation space to the app space, one where bots help with many different tasks. Chatbots are now sophisticated interfaces where people click, tap, and swipe buttons, carousels, and visual cards, for quick responses. These conversational interfaces (CI) are changing the ways in which people and businesses transact. Some enthusiastic, early adopters even believe that chatbots and CI’s will soon surpass traditional apps.
Comparing Different Kinds of Chatbots
Stateless VS. Stateful
Some chatbots are stateless, while others are stateful. Much like the voice UIs (VUIs) discussed in our Voice Development Guide , this is all about how well chatbots handle context. One is neither better nor worse; it all comes down to the use case.
- Stateless bots typically operate on a one-question-one-answer basis. Ask a question and the bot replies, but immediately following, all context is forgotten. These bots may be forgetful, but that doesn’t mean they aren’t useful. For example, Facebook Messenger-based “Instant Translator” translates English messages into Arabic. In this instance, the bot doesn’t remember the previous text, but it doesn’t matter because every sentence is independently translated.
- Conversely, stateful bots are more engaged conversationalists. Stateful chatbots have varying memories. Some stay context-aware only in a single session, like a travel ticket booking system. However, some (like Google Assistant and Siri) are personalized to individual users, so they remember various discussion histories and preferences across all time.
Non-AI VS. AI
Some bots use simple rules, like a series of IF-THEN statements, while others chat using artificial intelligence (AI) like NLP and machine learning (ML). Both are powerful in their own right; again, it all comes down to the use case.
- Non-AI (rule-based) bots respond by noticing keywords and phrases. For example, if you browse a travel site and type: “book,” a date, and two cities, the travel site chatbot knows how to respond by scanning for keywords and phrases (i.e. book, city name). This behavior is not an example of AI, but rather a simple algorithm that relies on IF-THEN statements. Useful conversations might happen in a situation where the following occurs: IF phrase A, THEN do action B. For a travel site, this might look something like: “IF input.Contains(‘book’) THEN do customerBooking(input)”. The interface determines if your input text contains the word “book.” If it does, then the bot passes the rest of your input to a function “customerBooking(),” which will extract the dates and cities from that text.
- AI bots, on the other hand, interpret intent with advanced AI techniques, including:
- Natural language processing (NLP): Many programs demand deeper understanding of human language’s meaning and organization. They create a useful data structure from processed messages, and then use this for additional actions and reactions.
- Machine learning (ML): Some AI bots are even smarter—they learn from every conversation and feed this data back into their system. This is what we call machine learning.
- Deep learning (DL): Deep learning is a type of machine learning, that’s typically used to describe neural networks. Chatbot neural networks connect what you say with your intentions (what they believe you want them to do) using special weighted models that adapt with time. Recent, influential deep learning techniques like sequence-to-sequence learning (Seq2Seq), are being used to improve chatbot accuracy.
To learn more about AI, ML, and DL, check out our Artificial Intelligence Development Guide .
Channels and platforms
A variety of different conversation chatbot channels and platforms are available, offering various benefits and tradeoffs:
- Social media and chat: Some CI experiences and chatbots use Facebook, Slack, Telegram, and Kik, while others use SMS. All of these offer simple integration and broadly reach millions of worldwide customers on their favorite platforms.
- Web: Some websites have integrated chatbots that provide “an always-on” presence, ensuring leads are captured and funneled.
- Standalone bots: Other CI experiences are standalone applications running on smartphones, desktops, or smart speakers like Amazon Echo, providing an opportunity for rich customization and full-fledged voice apps .
Conversational experiences run the gamut from purely text-based to user interface- or voice-based, with all the expected pros and cons:
- Text-only: Some chatbots, like SMS bots, function purely on text. These kinds of bots can vary in terms of their capabilities, but the downside is that you usually can’t tell how intelligent the bot is until you try using it. As we learned in a previous section, if you encounter one that is rule-based, your requests must be quite limited and utilize basic key words. Otherwise, there is a good chance that the bot won’t understand.
- Text- and UI: Bots, such as Facebook Messenger, use a combination of text and UI elements like buttons and cards, which provide actionable, pre-canned responses. These address some of the limitations of text-only bots.
- Voice-only chatbots: Smart speakers like Amazon Echo, which can be considered a type of chatbot, do not require screens. Instead, it relies solely on auditory cues. This is referred to as a voice UI (VUI). VUIs offer users the ability to place online orders, play music, check the weather, and even turn on the lights, just by using your voice. Voice assistants and chat experiences are quite capable (Alexa supports 20,000+ skills), but people can still get stuck if they don’t ask questions “properly.”
- Hybrid conversational interfaces: Web and smartphone-based chatbots can have CIs with all of the features listed above-text, UI, voice, and audio. These dynamic UIs can provide beautiful experiences with great flexibility. However, this can be problematic when they’re masters of none. An AI that excels at a few specific input types usually beats one that is ok at many.
We’ll soon dig deeper into some of the incredible conversational experiences and chatbots that exist today, including how you can build some of your very own.
Bot for the grace: Chatbots’ many benefits
Chatbots and conversational interfaces are heralding a new era of commerce. According to Harvard Business Review, “chatbots are changing how companies talk with customers.”
So, why are so many businesses turning to bots?
- Conversation is natural: The primary advantage of chatbots over other UIs is that people prefer talking to tapping multiple buttons. In a customer service benchmark, eDigital found that live chat had 73% satisfaction, compared to just 53% for apps and 48% for social media .Conversation works especially well for complex tasks like analytics or supply chain. Asking a digital agent straightforward questions means they can do the thinking and answering, freeing employees’ valuable time up for other things. Unlike the Microsoft Office Clippy of years past, well-designed chatbots ease frustration rather than creating more of it.
- 24/7/365 service: Even for complex use cases still requiring people, chatbots facilitate operational efficiency by gathering important information that can be used by customer service reps. This increases customer engagement, which translates to more savings and higher customer retention in high volume environments.Business Insider estimates that chatbots will save $65 billion in annual salary costs by automating insurance, financial, and customer sales representative roles. In banking and health care, Juniper Research expects chatbots to save an average of 4+ minutes and $0.50-$0.70 USD per conversation by 2022 , compared to traditional call centers.
- Largest-ever audience: It’s no surprise that 80% of brands will be using chatbots by 2020 . Chat services have created the largest human gathering in history, surpassing even social networks. Now the same bot talks to different people across many different channels, opening totally new marketing opportunities. Chat service, Kik, even says its newest platform is dedicated to “chatvertising.”
- Reaching teens: Gen Z is leaving Facebook in droves in favor of more chat-focused mediums. As a result, companies like Facebook, Slack, Skype, Telegram, Kik, and WeChat are opening up bot platforms targeting this young, highly-engaged audience. Teens from 13-17 years old are now inherently mobile-first shoppers . In fact, the annual purchasing power of teenagers who spend more than 3 hours a day on messaging apps is around $44 billion USD.
- Bigger discussions mean big data: With more bots than ever, there is a huge accumulation of conversation data streaming in. Big data analysis means understanding people far beyond demographics, and can help businesses everywhere craft better solutions.
Bot wisdom: Limitations & Lessons Learned
Early chatbot and conversational interface enthusiasm was tempered by some sobering lessons. If 2016 was the peak, 2017 was the “trough of disillusionment.” In January 2018, Facebook announced that it was killing M, its flagship virtual assistant. Tech and business magazine, Wired, subsequently declared, “chatbots are dead.”
But, like so many tech world proclamations that “_____ is dead,” this couldn’t be farther from the truth. In fact, conversational experiences are just getting started. As tech writer Kayla Matthews points out, chatbots are exactly what we need in an always-on society; they make our interactions and conversations with ever-present tech gadgets far more natural than before. Why search endlessly for a menu button or the proper light switch when you can just ask for what you need, instead?
Declaring chatbots dead is like declaring conversation dead; if the medium is more natural, people will definitely use it. The key lies in proper implementation, understanding limitations, and applying established lessons. Here’s what the tech world learned in its first few gung-ho chatbot years.
- Hidden costs: Facebook’s advanced M had unexpected behind-the-scenes costs. Although machines were learning from conversations, they still needed lots of training. Chatbots weren’t providing the promised cost savings; they just shifted the effort. Employees had simply been turned around in their chairs, from facing their customers to facing the AI.
- Embarrassment—and worse: Microsoft’s Tay had to be shut down after it made racist and misogynist Twitter comments . AI learns from society and social media. It begins its digital life naïvely, unable to tell right from wrong. It’s up to us and the “vanguard of computer scientists” to make sure it’s learning properly.
- Not quite human: Chatbots can’t yet replace our conversations and emotional connections. Today’s bots automate common mechanical and repetitive tasks, the kinds of scenarios where people aren’t needed. Aspect Research found that 44% of surveyed US consumers actually prefer chatbots , and this number is rising. Bots don’t display empathy, which is sometimes exactly what we need.
- Authentication: People recognize faces and voices, but conversational agents often don’t. XETV viewers in San Diego learned this the hard way when their Amazon Echos all ordered dollhouses because the TV news anchor mimicked a child ordering a dollhouse on air. With time, voice and facial recognition featuers will be added. In fact, many smart speakers and voice assistants already have these abilities. Unfortunately, biometrics are a greater challenge for text-based interfaces.
- Designing for the lowest-common denominator: Some channels like SMS offer basic text-only communication. Others, like Slackbots, exchange videos, images, emoticons, visual cards, selection menus, and carousels. Designing cross-channel conversional experiences sometimes means designing for the “least common denominator.” Solutions like Converse.ai try addressing this, but until universal standards gain broad acceptance, the marketplace will continue to be filled with disparate solutions.
- People vary, but can chatbots be both broad and narrow? Machine learning advances over time, but in some industries, this can be a problem. In the travel space, certain behaviors can actually skew machine learning models over time, even though AI needs to be more of a generalist. TripAdvisor discovered that people’s preferences vary depending on their travel purpose; business travelers have very different needs than tourists or those who want to live like locals. This is partly a limitation in AI’s learning capabilities; they’re great in narrow scenarios, but not so much when you need them to go broad. (We discuss how today’s AI is only “narrow AI” in our Innovator’s Guide to AI.)
The Bots Are Back In Town
A mad rush into hastily-developed bots might have caused some initial disenchantment. But, even during their dip, chatbots collected incredible amounts of data and advanced the AI, NLP, and machine learning fields. This is why, despite killing M, Facebook said, “we’re taking these useful insights to power other AI projects at Facebook.” Since then, it’s become very clear that bots are here to stay for businesses, entrepreneurs, and innovators everywhere.
People are easing app fatigue with chatbots
People are tired of apps. A Gartner survey showed that 20% of people are downloading fewer apps than in previous years. Furthermore, an August 2015 Millard Brown Survey revealed that 43% of US smartphone owners only used four to six apps on an average day, even though most phones have approximately 40 to 70 apps downloaded on them.
Whether or not they want to use traditional apps, people will always need to accomplish smartphone- and web-based tasks. As Microsoft CEO Satya Nadella said in 2016, “Chatbots are the new app.”
Here are some outstanding, award-winning examples where they’re making waves.
This social voting chatbot helps people make decisions by sending questions around the world for votes. Pictures, questions, and opinions receive instant responses from others, facilitated by the bot. Swelly has a native mobile app as well as Kik, Facebook Messenger, and Telegram integrations. It won the 2017 Chatbottle Editor’s Choice Award.
Swelly helps people get opinions from others.
A weather cat-bot with a personality, Poncho runs on Facebook Messenger, Kik, Viber, Slack and a native mobile app. For being a one-trick feline, it’s actually quite sophisticated. It accurately guesses intent even when questions are vague and don’t have typical keywords. It’s also set apart by language: a writing and editorial team successfully made a simple weather bot fun, quirky and humorous. Add in cute, cat-themed graphics and it’s quickly become the internet’s favorite digital weather conversationalist and raised $2.4 million in seed funding.
Poncho is a unique and chatty cat-bot, proving weather doesn’t have to be boring.
Expense management compan, Expensify, has a built-in virtual assistant Concierge that helps employees properly file “expense reports that don’t suck.” This B2B chatbot is used by companies like Uber, Warby Parker, and Virgin Hotels. Besides resolving over 75% of problems, Concierge has seriously amplified Expensify’s brand, resulting in 500% more free trial subscriptions.
This experimental bot was created by ORF, the Austrian public broadcaster, during the 2016 Austrian presidential elections . It’s an innovative case where a social media-based chatbot was used to inform and educate the public without any partisan slant. Wahlbot answered questions about news and gave election projections and results. Wahlbot demonstrates how sophisticated and successful even non-AI bots can be.
This Facebook Messenger chatbot bridges the gap between storytelling experiences, chatbots, and augmented reality . It was created to build empathy for the African water crisis. Yeshi, a fictional Ethiopian girl, takes people on a simulated two and a half hour walk to fetch water. It’s an enriching, immersive experience that combines Q&A sessions with multimedia like images GIFs, audio, and videos. This bot is designed to inspire more empathy in people, with its inverted conversation strategy that quizzes the audience.
Wired has featured DoNotPay among its top Facebook Messenger bots. Originally, this simple bot helped people fight parking tickets in the UK, NYC and Seattle by asking questions to figure out if appeals are possible. Then it guides people through the appeals process. It effectively provides always-on assistance and assurance in a stressful situation. It’s also effective: DoNotPay overturned 160,000 parking tickets, a 64% success rate worth USD $4 million in penalties. DoNotPay has now launched over 1,000 new bots fighting all kinds of legal battles , including helping refugees file for asylum in the US and UK. Even more impressively, DoNotPay’s creator is just 19 years old.
Natural Language, Artificial Intelligence
Some chatbots and conversational interfaces are AI-less, while others have such complex NLP that they decipher a person’s intent. Some go a step beyond, learning from earlier interactions and perhaps even predicting what a person wants using machine learning techniques. Designing an effective chatbot means understanding their basic architecture and what kind of intelligence is needed.
Inside chatbot architecture
Chatbots have three main parts: the registered bot, the channel or platform (like Facebook Messenger or Slack) and millions to billions of users (if you pick a popular platform like Facebook, Slack or WhatsApp.)
First, a chatbot developer registers their chatbot in the channel’s approved bot list. Then, the bot and channel then connect with a “webhook.” This webhook is a promise between the chat service and the bot. This promise says that the service, run by someone like Facebook, will reply to people’s messages and pass what they’re saying to your bot’s “brain.”
This is true for all bot-to-bot platform interactions. As mentioned earlier (see the section about Non-AI VS. AI bots), each bot’s brain is wired differently. Some are rule-based, and some AI-based, while others follow a structure referred to as a conversation tree.
Many video games define their character dialogue options in conversation (or dialogue) trees. Rule-based chatbots do the same. Conversation trees map out all the potential directions the discussion may go, with many different branches for all the possibilities. If you drew it out, it would even look like a very intricate tree, with a potential input (like, “How are you doing today?”) at the top of each branch, and many different options branching off from there.
The conversation tree is responsible for predicting all possible message inputs as well as the chatbot’s reactions, and the discussion can take many different paths through the tree. A rich conversation tree ensures you’re successful and the bot seems smart, even though it’s not using AI. It should also help people quickly accomplish their goals, make information easily accessible, keep the bot in control of the conversation and prioritize a business’s needs.
AI as a service
Enterprises often custom-build NLP and machine-learning engines. However, NLP and ML are challenging problems requiring significant expertise. This is one reason why third-party service platforms are emerging.
Microsoft LUIS.ai, Facebook Wit.ai, Google Diagflow (formerly Api.ai) and IBM Watson are some of the many popular NLP service platforms. These live third-party services accept natural language and return structured data (data that’s broken up into a simple, organized format for easy processing), typically JSON. Many also provide their own machine learning components (called modules) which can be added to your platform or product. Instead of building your own NLP or AI, you can just plug-and-play theirs. These companies have turned years of experience into platforms giving entrepreneurs and small businesses powerful machine learning solutions. Thanks to them, businesses without significant AI expertise don’t have to reinvent the wheel.
Additionally, some people build custom AI-based conversational interfaces using Google’s deep learning engine TensorFlow, and it’s also possible to create custom deep learning programs from popular libraries like DL4J (Java), Theano (Python) and Torch (C/C++).
For more ideas on getting started with AI and machine learning, check out our Artificial Intelligence Development Guide.
Developing And Building Conversational Experiences
As you’re considering a chatbot for yourself or your business? Here’s how you can look before you leap.
First: Identify your goals
Before developing a bot it’s important to evaluate if it fits your business and problem.
Common reasons and scenarios for using chatbots include:
- Social strategy: Chatbots on social platforms are an extension of your social presence. They combine the advantages of personalized newsletters and social media. You can use an interaction to gather people’s expectations as well as broadcast information.
- Always-on presence: Chatbots are ideal when agents are needed for 24/7 support, whether that’s customer service, delivery, round-the-clock sales or collecting worldwide leads.
- High volume and queues: If your public services have high traffic, conversational experiences are an excellent solution.
- Repetitive tasks: Chatbots effectively handle simple and repetitive tasks, making employees happy and reducing human error.
Second: Choose your channel
A chatbot channel is the provider or platform the bot uses to talk with customers.
- Single- versus cross-channel: Cross-channel bots have the broadest reach, and many platforms like Converse.ai support cross-channel natively. However, others like Manychat are only Facebook-based.
- UI depends on platform: A bot’s UI and design are restricted by channel choice. If your customers use feature phones like Capital One’s do, SMS is a key channel. By contrast, Facebook has stacked buttons, multiple choice polls, visual cards, date-time pickers, and more.
Websites as channels
Social media and chatting platforms aren’t the only place for conversations. Some of the most visually advanced CIs run on websites like Jakt ’s to immediately greet customers as they land on the site.
Additionally, website chatbots are ideal navigational assistants. However, they don’t have the audience reach or access of Facebook Messenger and other social media experiences. Depending on your needs, they can be good on their own or in combination with other channels.
Third: Develop a design
UX and visual design are the keys to conversational experience development. Because they’re discussion-based, the CI is even more critical than a traditional UI, and an excellent design team is arguably more essential to success than the development team.
When it comes to design, we love Google’s common-sense guidelines for VUIs , which also apply very well to CIs.
This list isn’t exhaustive, but some of the best ones include:
- Follow everyday speech patterns
- Give enough information, but not too much
- Use acknowledgement cues to show the bot is listening
- Suggest intuitive conversational examples
- Find ways to make errors sound more natural and human
- Copy phone trees and other outdated models
- Use robot-like or mechanical speech patterns
- Talk down to or condescend your users
- Use “Go back” instructions
- Provide obvious statements
Fourth: Pick your implementation
So, you’re ready to build a bot. But how do you choose between a ready-made bot and something totally custom? Here are a few considerations to help you weigh the options:
- Problem scope: Simple rule-based bots work for general purposes. If your problem is common, chances are there’s a ready-made chatbot builder with a solution.
- AI needs: If you’re confident you can handle all the possible dialogue options, perhaps you don’t need sophisticated AI and can just make a rule-based bot. As your business grows, though, you might need AI in the future. If your bot’s conversation tree is getting too complex to maintain, it might be time for AI.
- Channels: If most of your traffic is on Facebook, Facebook Messenger is a natural chatbot starting place. If you have many popular channels, go cross-channel. Pick the best single- or cross-channel solution, then reach your audience where they already are.
- Language support: Any platform should support your customers’ major languages and locales. Not all languages are supported equally, especially not with VUIs. If you need many languages, a custom solution (possibly with outside development help) may be needed.
- Integration: Some platforms can be readily integrated with Zendesk, Salesforce and other CRMs, plus calendar schedulers, shopping carts and other third-party services. Others lack this slick integration. List out your important first- and third-party tools, then check if the bot you’re looking at will play nice.
- First- versus third-party: Customer privacy an important consideration with third-party services; self-hosting gives you control and ownership over chat data, which may include customers’ personally-identifiable information (PII). But you’re also then responsible for keeping it safe. Likewise, choosing between self-hosting and third-party hosting comes with trade-offs between cost, convenience and maintenance.
- Voice support: Amazon’s Alexa shows that voice has a definite place in the future of conversational UI. However, there are only some scenarios where it makes sense. Whether it works for you depends on your channels, customers’ hardware and the problem you’re solving. If you’re planning to add voice, definitely make sure your platform supports it.
- Performance: Bot analytics measure chatbot performance much like web analytics, including the number of users, conversations, and average time spent (engagement). A visual funnel shows conversation progress and where people are getting bottlenecked. Custom analytics tools include Dashbot Botanalytics, Google’s Chatbase, Manner, Botmetrics and others. Ready-made bot-builders also include out-of-box dashboards.
Ready-made chatbot builders like Manychat and Chatfuel work when you need a 24/7 presence: staying connected with customers, responding to questions, sending resources and scheduling messages. With these tools, you can easily configure sequences: discussion flows people move through based on their responses. Builders like Botsify also provide voice UI.
Ready-made bot builders are typically freemium, meaning they have free offerings with premium feature upgrades available; whether it’s worth upgrading depends on which features you need.
Some ready-made bots are generic, but others are more specific, including:
- Domain-Specific Bots: Marketing, sales and support need specialized tools to address sales problems like scaled lead conversations. As a business grows, the number of leads rapidly increases, requiring an automated response strategy. Bot builders like Drift and Blip provide a highly scalable system. Some also have specific analytics, like marketing bots that measure cost per lead.
- Industry-Specific Bots: Most restaurants have similar issues related to scheduling reservations, responding with menus and broadcasting promo offers. Guestfriend lets restaurants build their own bots, while Alterra.ai has a travel industry specific bot solutions. Others like Octane AI give general e-commerce solutions and integrate with popular platforms like Shopify and Magento.
Bot development frameworks provide bot-specific programming tools. Some are fairly basic, self-hosted software development kits (SDKs). Others are full web services whose massive language datasets have very accurate models thanks to years of enterprise data.
Microsoft Bot Framework, Google’s Diagflow (formerly api.ai), Botkit, Pandorabot, wit.ai and Conversable’s AQUA have all gained popularity in recent years. Some easily integrate with channels like Facebook Messenger, Slack, Kik, Telegram and SMS, but many require self-hosting.
If you want to build a custom experience, here are some key points to weigh beforehand:
Parallel software projects
Building custom CIs requires parallel software development streams:
- AI and NLP model development: Coding first happens for the chatbot’s natural language components. This team must understand NLP techniques like stemmers, part-of-speech tagging and naïve Bayes classification, as well as machine and deep learning techniques. Alternately, if you choose a third-party NLP module, your development team must understand how to integrate (and potentially train) it.
- Communication channel development: While one team is developing the natural language components, another team is developing the bot for its final home: the channel where it will be deployed (Skype, Facebook or similar). This team may include writers, designers and UX experts who can translate business needs into beautiful, useful and potentially even clever CIs.
Custom chatbots should be “loosely coupled,” meaning their components are easily replaceable when and if better solutions arise. For example, if your botkit.ai module isn’t working out, it should be easily swappable with meya.ai. The same goes for speech recognition and other components. Because many modules are open-source, they could be discontinued or acquired at any time. Loosely-coupled architecture is a way of managing this risk, especially with open-source components.
Fifth: Take it from us
We have our own experiences implementing chatbots and conversational apps at Jakt, from Intercom’s Operator Bot to Facebook messenger bots to completely custom bots. Here’s what we recommend for partners considering their own conversational apps using out-of-box solutions:
- Don’t be afraid to customize: Out-of-box bots like Drift often promise to completely replace other third-party tools like Typeform. However, it may be very expensive or unwieldy to replace what you’re already using. Don’t be afraid to customize your bot, and don’t worry if you’re not using every single feature to its full potential.
- Understand your bottlenecks: Although it’s very easy to create conversation sequences, pre-made bots still have limitations that can create user frustration. We noticed that many people got stuck when they were prompted to enter their email address; if they submitted an invalid email, they’d get trapped and frustrated in an endless loop. Analytics tools let us see these bottlenecks, and we were able to redesign the conversation flow for a better UX.
- Use your authentic voice: Writing conversation flows required sitting down and really thinking about how we’d talk to partners in-person. We wanted our website presence to reflect our company’s voice and avoid sounding robotic. The same should go for you; make sure you’re writing authentically. If your company has voice guidelines, they should be applied to this conversational interface.
- Make it dead simple: Many out-of-box chatbots come with tons of bells and whistles, including meeting schedulers, sales emails, account-based marketing and CRM integration. However, adding meeting scheduling isn’t a great idea if none of your customers need it. Just as Google recommends VUIs give just enough info, we think conversational apps should have just enough features.
- Out-of-box doesn’t always work: Sometimes out-of-box solutions aren’t a great fit. When there is domain specific or industry specific use case and a bot platform doesn’t already exist for that niche, you may need to build a custom bot.
A Bright Bot Future
Chatbots have come a long way. So, what does the future hold for conversational interfaces? We see a few key trends on the horizon.
Raising your voice
Gartner predicts that by 2020, 30% of web browsing sessions will be screenless . By 2021, brands with website voice and visual search will increase revenue by 30% . Most smart speaker owners are avidly using their devices, and with 20 million Amazon Echo devices sold, 20,000 Alexa tasks and more Alexa-enabled devices on the way, voice is only getting bigger.
For more on this key trend, check out our Innovator’s Guide to Voice .
A new age of understanding
As many conversational apps feed language data into shared services like Google Diagflow, CIs are getting better at understanding discussions of all kinds, even when they span many different industries. Bots are also getting better at making sense out of mistyped, mispronounced or poorly-phrased requests. This shift from natural language processing to natural language understanding will seriously disrupt the field.
Bring on the big data
Just as conversational interfaces will soon better understand our requests, they’re also helping us better understand customers . In this brave new digital world, people demand products and services on their terms. Meanwhile, nearly one-third of industry leaders confess they don’t understand their customers.
Luckily for these retailers, bots are social creatures that don’t often confine themselves to one channel. As millions of people have billions of bot chats, these interactions form a massive data collection operation that enables huge dataset analysis.
This big data analysis isn’t happening yet, but it’s inevitable. In the future, big data will allow us to deeply understand customers based on virtual discussions. Then we’ll turn right back around, asking our chatbots for customer insights and analytics.
Chatbots are the new app
We know conversational interfaces are more than just a passing trend. That’s why we’ve got some of the best chatbot-focused AI/ML, CI, UX and writing experts around. So, whether you’re looking for a unique bot from scratch, or you’re integrating a rich third-party experience across social channels, Jakt can help. Because, where we once said “there’s an app for that,” we’re already saying “there’s a bot for that,” instead.