You can’t throw a rock in Silicon Valley without hitting someone talking about whether machine learning, AI, and chatbots are truly “the next big thing.”
I get it. It’s fun to be ahead of the curve in fast-moving times.
The truth is, we’re still in artificial intelligence infancy. Everyone is just trying to improve their technology and understand what works for online audiences.
But you don’t have to wait for AI to reach its zenith before you deploy chatbots for your business. Chatbots definitely work today. Some Sumo-lings have even started building bots as a service for their clients.
(BaaS? — Trust me, it’ll catch on.)
Facebook has found that customers are 50% more likely to buy from a brand after chatting with the company—regardless of whether it’s a human or a bot.
Can chatbots automate your entire sales, customer support, and marketing team in 2019? Hell no. But can they help your business? The answer is yes in almost every case.
We brought in the big guns to help us understand the industry today. We asked Andrew Ganin from ActiveChat to explain some of the finer details of the chatbot industry. In this article, you will learn:
- Chatbot basics
- Decision tree vs. AI/NLP chatbots
- How to use chatbots to improve your business
- Two examples of Sumo-lings who build chatbots
- A brief description of three AppSumo chatbot partners: ActiveChat, Continually, and Quriobot
Without any further ado, let’s get started with the basics:
What is a chatbot?
Definition: A chatbot is an online instant messaging tool that is programmed to interact conversationally with humans and perform certain tasks.
Chatbots are a popular way to let customers interact with your company on their own terms without jumping through hoops, wading through online FAQs, or waiting for a human to reply to their message. They’re a powerful marketing channel.
How do chatbots work?
Chatbots today come in two main varieties: decision-tree vs. AI/natural language processing-based chatbots.
Here’s what those mean:
Decision-tree (also known as rule-based) chatbots use a predictive model to give users a list of preset options — in the form of buttons and “quick replies.” This type of bot doesn’t require machine learning because it doesn’t need to interpret anything. The bot simply locates keywords and steers the conversation at each juncture by offering a couple of pre-set response options.
Andrew suggests thinking about decision-tree chatbots like an IVR menu. You know, that narrowing process we all go through over the phone when a robot asks you to “press 1 for English,” or “press 0 at any time to speak to customer service.”
IVR systems and rule-based chatbots both slowly narrow each conversation with a user until that person receives the response or service they’re looking for.
Benefits of rule-based chatbots
Put simply, rule-based chatbots are easier — and faster — to set up. They’re better for straightforward, simple tasks like placing a standard order or booking a demo. They solve real customer problems without adding a bunch of unnecessary bells and whistles.
Here’s a simple example of decision-tree based restaurant bot conversation from our friends at ActiveChat:
Bot: Hi! I’m Pizza bot, and I’d be happy to help you with your order. Would you like to make a reservation or order delivery?
Buttons: <Reservation> <Delivery> <Other options>
User: clicks <Delivery>
Bot: Awesome, our delivery agents can be at your door in just 30 minutes. What would you like to order?
Buttons: <Pizzas> <Salads> <Desserts> <Back to main menu>
User: clicks <Pizzas>
Bot: Here are some of the pizzas to order:
Buttons: <Margherita> <Capricciosa> <Tonno> <Vegetarian>
User: clicks <Margherita> …
At every step the bot offers the user a set of options to choose from, thus creating a “tree.”
Chatbots created using decision-tree models use “if-this-then-that” logic to help website users get to where they want to go.
How rule-based chatbots work:
Every conversation has to begin somewhere. (Usually with a “Hello! 👋” amirite?)
Then with every reply i.e. action (or lack thereof) from the client, the decision-based chatbot branches out into different scenarios.
Based on the feedback the chatbot gets from the user, it can:
- give different kinds of written reply
- trigger a task (e.g. send an email, forward to another page, schedule a meeting, issue an invoice, etc.)
The downfall of this model is that it is limiting to the user.
If the user needs something that isn’t mentioned in the limited responses, then they might abandon your site. Or worse, if the chatbot receives an unscripted reply, it will send the user the oh-so-annoying default reply.
AI/natural language processing chatbots
Though not mutually exclusive, AI-based chatbots use machine learning to run a natural language processing (NLP) program. These kinds of chatbots rely on pattern interpretation and making inferences instead of just scanning for keywords. AI-based chatbots can perform higher-level work for your business such as processing an insane number of insurance claims.
AI-based chatbots take more time to set up because they have to learn how to respond to your audience. Fortunately, they get smarter as they go through both supervised and unsupervised learning.
An NLP program employs algorithms to identify patterns from past data and determine outputs. An AI chatbot may use support vectors, statistical regressions, and decision trees (or a combination of all three).
Benefits of AI/NLP chatbots:
AI-based chatbots are best for large companies that anticipate multi-lingual traffic and need the bot to respond to higher-level inquiries in multiple languages. These higher-level chatbots are more volatile in some ways because they use human inputs combined with machine learning to get better at responding to inquiries.
In 2019, we’re still waiting for AI and machine learning to fully catch up to our 2001: A Space Odyssey expectations. (I guess that’s kinda a good thing?)
Chatbots that boast using machine learning are definitely still under development, but they do work for many businesses.
Modern AI is quite good at understanding users’ intents expressed in natural language, but it requires:
- Preparation — collecting possible intents from existing conversations or coming up with your own assumptions about what your users might say
- Investigation — building a database of different wordings/expressions for the same intent
- Implementation — triggering specific bot skills or flows for each specific intent
When these intents are combined to form a complete conversation, your chatbot can become smart enough to be able to close up to 80% of customer interactions without human agent intervention.
But all this basically means you’ll need to be pretty hands-on (particularly at the outset) to get your AI-based chatbot to a satisfactory level of communication. It will require a lot of testing and experimentation to make sure your bot doesn’t infuriate, insult, bore, or confuse users. After all, humans will be humans.
Some AI-rule based chatbot platforms provide their own NLP engines, while others rely on integrations from NLP industry leaders such as Dialogflow by Google, Wit.ai by Facebook, Watson by IBM, and LUIS by Microsoft, etc.
How AI/NLP chatbots work:
Natural language processing uses machine learning to understand human speech. To do this, the chatbot doesn’t focus on keywords. Instead, it dissects phrases into logical elements:
- Intents: what the client wanted to achieve in the line of conversation
- Entities: variables that add details and clarify the intent
- Context: when the phase doesn’t contain a vital piece of information — intent or entity — the chatbot can recognize them in previous frames or ask clarifying questions
E.g. in the phrase “What time is it in London?”
> Intent is to know the time
> Entity is London
> And if I were to ask “What about Paris?” after that, the context would be, “What time is it?”
Yes, this type of chatbot can create entertaining chit-chat and answer some questions. But to provide optimum value, Andrew recommends combining this kind of chatbot with a decision-tree that will allow the chatbot to trigger tasks to lead the user to a desired outcome.
As a web-based messaging tool, basic chatbots usually respond to just two events:
- First interaction: A chatbot is used to greet new visitors by sending an initial message. The bot can frame what it does for the visitor: offer to answer questions, point visitors in a particular direction, tell them what the bot (or business) can do, etc.
- User message: A visitor can also send messages to the organization using the chatbot. The visitor’s message is usually processed with NLP or keyword detection to understand the user’s intent and branch the conversation accordingly.
But chatbots can also offer more advanced (AKA fun) functionality through behavioral tracking like:
- Shopping cart: used to trigger cart abandonment sequences
- Order processing: used to send order updates
- Website visits: used to send custom messages based on which website pages users are visiting
- CRM updates: used to send messages when user status is changed in CRM (for example, when membership plan is changed)
- Payments: to help your users with payments and invoices
How to know if my business will benefit from chatbots
How do you know if chatbots are the right tool for your business?
“If it’s 2019 already,” says Andrew. Or if you:
- receive a lot of inbound messaging
- want to become friends with your customers
- are looking to provide instant 24/7 two-way communication with your audience (lead generation and qualification, marketing, proactive sales, customer support, etc.)
- need reduce support costs and increase conversion rates, and customer engagement
What are my chatbot options?
1. Live chat
Live chat uses the same format as a chatbot — but puts a human operator at the other end.
Though live chat can be extremely gratifying and helpful for website visitors, the downside is that it can get overwhelming at peak times if too many visitors require attention at the same time. (Such as right after you launch a product on AppSumo. Looking at you, LTD partners!)
Instead of interacting with a human operator, the user interacts with an AI or rule-based decision-tree bot. Many chatbot services have an option of switching the conversation to a live chat when human assistance is needed.
As mentioned above, chatbots come in two basic flavors: decision-tree (these bots are controlled by buttons or quick replies and are actually small “websites” using messaging interface) and AI/NLP chatbots. The latter are most complex to build, but at the same time provide much more flexibility and look like a real person in conversation.
To get started with a chatbot, you’ll need to decide where the chatbot is going to live online. Various chatbot options on the market today live exclusively in FB Messenger or on your website as a widget (single-channel). Others, like ActiveChat.ai and Quriobot, can exist across several different channels (muliti- or omni-channel).
e.g. FB Messenger only or website widget only
e.g. SMS + FB Messenger
Online messenger (FB, WhatsApp, Telegram, etc.) + SMS + phone calls + website widget
Chatbots can’t do everything. But they can do a lot. Here’s a sampling of chatbot applications:
- Support chatbots
- Qualification chatbots
- Marketing and sales chatbots
- Engagement chatbots (including product tours and campaigns)
- Full-cycle chatbots
Depending on the needs of your business, chatbots are a great way to collect data about your audience.
Here the main purpose is storing and sorting the information about clients. Most of the time, such platforms do not have advanced chatbot tools.
A chatbot is a two-way data exchange. By adding a software integration to your chatbot, you can seamlessly collect user data (like email addresses or even more advanced user data) and give yourself all the benefits that CRMs provide your sales team.
Of course, there are still some basic chat-only solutions which will give you the ability to live chat with a client. However, getting the data in and out of them could be a hassle.
Are chatbots easy to install and manage?
If you’re using one of the “low-code” or “zero-code” bot platforms like Activechat.ai, Quriobot, Continual.ly, Chatfuel, or Manychat, there’s no need to get too technical. You can build a complete basic chatbot from ready-made blocks or even use a pre-made template that can be customized.
More advanced solutions like Dialogflow, FlowXO, or Liveperson will require some level of tech knowledge, particularly when it comes to installing various backend integrations.
As a rule of thumb: if you are using artificial intelligence, it will require more work — so make sure the business objectives align before making that investment.
How long does it take to get a chatbot working?
With a high-level platform, you can get a basic bot up and running in just an hour or so. If you’re using an AI chatbot, then comes the fun part — watching your bot have actual conversations with users and updating the flow to provide the best customer experience. This can take anywhere from a couple of days to two weeks.
It’s important to have a visual flow builder to be able to make changes to bot’s conversation even if you know nothing about tech. You should remember that a good chatbot is a living creature, and there’s always something to improve or new skills to add.
According to Andrew, “When building from scratch, you can have a basic FAQ and navigation bot within 2 hours and a full eCommerce bot within 2 days.”
What are chatbot campaigns and how can I use them to engage customers?
There are three basic types of chatbot campaigns:
- Ad-based – these are triggered when a user clicks on your ad on Facebook, for example. ActiveChat reports that conversion rates from bot conversations can be 4-5 times higher than for those ads that require performing website sign-ups or purchases.
- Follow-up campaigns – these are triggered automatically at specified intervals, and help build engagement and keep close contact with your audience. Each campaign can consist of multiple messages sent at intervals or on specific dates. These messages usually combine some useful content with promos and purchase reminders.
- Segmented broadcasts – these can be used to send manual messages to specific segments of your audience. Chatbots are known for great open rates/click rates, and smart conversational design can provide great ROI and conversion rate.
How to use chatbots to improve your business
Once you’re ready to try chatbots, it helps to have several different focused use cases in mind that you can test on your site. And that’s the key: A/B testing. Just putting a bot on your site doesn’t guarantee magical results. Make sure you’re measuring outcomes as you enter the world of chatbots.
But on the flip side, you should think about the opportunity cost of not having a chatbot on your site or Facebook page. If you want to use chatbots, you need to determine whether your audience responds positively to chatbots, and if so, how to optimize that bot for conversions and customer satisfaction.
For business, the goal of bots is 1) getting paid and 2) creating happy customers. How do you do this? Stay focused on the objective of your chatbot sequence. Or as Nielsen puts it, “Don’t be overly ambitious: create bots for simple tasks. Complexity is not well handled in the limited bot interface.”
Chatbots for SaaS: Test and measure
SaaS Option #1: Make sure your chatbot is working for you—not against you
To do so, run an A/B test that might look something like the following:
Install the bot. Then create a cohort of website visitors (a cohort is everyone who visited the website during a certain time period). When compared to a control group before the chatbot was installed, you can measure to see whether the cohort was more or less likely to become paying customers.
You can also track churn: how does the cohort’s unsubscribe rate compare three months down the road? Perhaps they are less likely to become paying customers. If that’s the case, then you need to rethink your bot immediately.
SaaS Option #2: Support chatbots to bridge the service gap
Support chatbots can help customers feel seen and heard between the time they arrive on a Support page and the time a human can address their question. Chatbots can also be used to ask preliminary questions, suggest docs from your knowledge base, schedule a support call at a different time, or simply get the customer ready to talk to someone in customer support.
Again, you should test whether support bots help or hinder your business objectives. Run tests to see what kind of support your customers are expecting on your help page. Does a quicker popup response time on a support page help boost customer satisfaction? Track that cohort. Does having a support chatbot that talks with users before a human can get on the line help with retention? Test it out.
SaaS Option #3: Nudge potential customers toward the right package
If you have a three-tiered pricing structure, chances are you’re trying to get the average website visitor to choose the middle tier. Regardless of which package you want them to purchase, a simple “Did you know our customers who choose X package save an average of Y dollars a year?” could nudge potential customers in the right direction.
Again, definitely test something like this to see if people respond.
Chatbots for Digital Commerce: Bot as Seller
eCommerce Option #1: Upsell during checkout
In traditional retail or selling over the phone, upselling during checkout is a completely normal way to anticipate your customer’s potential needs—and conveniently increase the size of the transaction—all in one go.
So for eCommerce sites, consider using your chatbot to upsell once the user seems like they’re ready to purchase. A simple “Customers who bought X also bought Y” or “Hey, do you also need grip tape to go with your new skateboard?”
Many eCom sites give customers the option to easily add additional items at checkout, but this approach with a chatbot might be worth testing.
eCommerce Option #2: Collecting customer feedback
Another application for eCommerce chatbots is to collect user preferences. For example, a chatbot on a product page might ask a simple question like, “What’s your favorite color skateboard?” This is an easy and efficient method for customer research. If everyone loves pink and hates blue, you can hone your product lines to meet customer demand.
eCommerce Option #3
Upsell to the monthly subscription: “Would you be interested in a monthly subscription of this product?”* or “Save X% with autoship!”
*Hint: This simple hack could be a game-changer for some ecommerce businesses.
How Sumo-lings use chatbots in the wild
For solopreneurs and freelancers, maybe a simple live chat widget with a lead capture form (or staying responsive to FB Messenger) is all that we need to stay on top of our game.
For Doc Williams of Brand Factory, Inc., building goal-oriented chatbot sequences for his clients is where he sees the biggest impact. His bot sequences grow webinar signups, distribute coupons, schedule gamers, or promote viral giveaways.
What Doc has noticed is that when you have enough data to know what your customers want, chatbots provide some of the highest engagement numbers in the business. For instance, a recent open sequence with a thank-you gift (a coupon) resulted in 75% of website visitors clicking the coupon and 52% of visitors clicking through to the site. He knew these customers loved coupons, so he gave the people what they wanted. Not bad.
Another project involved creating a FB Messenger bot that allows fans to book paid gaming sessions with expert gamers. The bot collects basic info about what gaming system the fan prefers and allows them to directly book a time to play, all within Messenger. Money in the bank.
(Look, we can’t divulge all Doc’s secrets, ok? So don’t try to decode his masterful viral giveaway sequence above. Also, yes all the text is scrambled, so don’t even waste your time.)
The numbers Doc sees from building focused, task-oriented bots for his clients are notable, such as an opening sequence with 3 messages with a one day delay that he built for a client that resulted in 21 sales at ~$47 each with $0.32 CPL (cost per lead).
For others, like Sumo-ling Anina Net, CEO of 360Fashion Network, having a FB Messenger chatbot on her modeling page allows her to build a following, promote events, and gather leads while she sleeps. “People can book me. It fields out clients from passersby.”
Let’s call this the PewDiePie approach—but with a more value-add. “My company builds chatbots for other fashion models and influencers. I think a chatbot is great for celebrities to interact and share their message with their fans.” Anina uses AppSumo chatbot partners Quriobot, MobileMonkey, and Rocketbots for her various businesses.
What is the future of chatbots? Where are chatbots going next?
Instead of being trained, they would rely more on self-training by analyzing unsorted data and their previous experience.
Internet of chatbots
The internet of things is already here. The next big thing may be the internet of chatbots — where they can learn and exchange information with each other.
Voice-driven AI-based assistants enter our everyday lives
Let’s face it. The existing technology doesn’t solve all our needs just yet. Still, businesses all around the world already use it to do the most amazing things.
In the market for a chatbot? Consider using one of our AppSumo partners:
(Note that all three LTDs are currently sold out as of the publishing of this article. But we want to hear from you: which bot should we bring back?)
ActiveChat is an omnichannel (website, FB Messenger, WhatsApp, Telegram, Twilio SMS, etc.) chatbot with advanced behavioral tracking that allows you to connect third-party NLP services like Dialogflow.
This may sound complex, but ActiveChat makes it simple with their Visual Bot Architect and catalog of premade chatbot templates. (Mmm, everybody loves a good template.)
After just 15 minutes spent on bot development, ActiveChat allows you to build product galleries, accept orders, and send order updates and cart abandonment messages.
With a focus on e-commerce chatbots, ActiveChat has native Shopify and WooCommerce integrations. You can also exchange data with CRMs and other business tools through JSON API and use Zapier integration to connect your existing CRM.
Advanced tools like Activechat.ai also do event tracking and store behavioral data about your users (like visiting specific pages on the website, making orders or asking specific questions).
This data can be used in conversational design to provide each user with personalized messaging, tailored to their profile, behavior, or personal preferences. It can be quite simple (like sending promo campaigns for similar products on your website after visiting specific product page) or as complex as completely changing tone of voice based on user’s age or location.
Google Home, Amazon Alexa, Twilio Voice + more messenger channels to come in 2019. Their next version will feature a white label solution that will allow bot owners to share access to bot dashboards with their customers with their own branding. Ok, just check out their public roadmap. It’s pretty cool.
Continually is a single-channel chatbot that uses decision trees to emphasize button choices, images, videos and links over free text input.
Continually’s canvas-based bot builder has caused a stir for how simple and easy to use it is. Get a bot up and running on a site in a very short amount of time.
Continually is focused on website widgets + live chat hand-off via Slack. (They also give you a free bot landing page. See example here.) You can “power-up” the conversation flow and make it more intelligent with features like targeting, variables, and conditional logic.
Their automated scheduling assistant syncs with Office 365 and Google Calendar to allow leads to book a demo time seamlessly in the chat. Your bot can adapt how it responds to each customer based on where they came from, their on-site behavior, any previous visits or interactions, the device they’re using, or any other data you might have about them.
Continually integrates with MailChimp, Slack, Hubspot, WordPress, etc.
Continually ships new features and improvements each week, so you can expect that to continue. Their next big updates are to expand their live chat features, give bot makers more control over their bot’s appearance, and more integrations with the other marketing tools.
Quriobot is a multichannel rule-based chatbot that is extremely customizable.
When it’s on your website, Quriobot is a microapp which can handle anything from a simple contact form to calculating your insurance premium using the profile data of the user and an API from your backend.
Quriobot channels include a website widget and Facebook Messenger. They also have a FB Customer Chat Plugin that you can add to a website as well as a dedicated landing page you can create.
When using the dedicated landing page or website widget, you can switch to another service when live chat is needed using either a fallback mechanism or by using JS API from a chosen external chat service.
With Facebook Messenger, you can use the live chat fallback to Facebook Inbox App at any point in the conversation. This can be initiated by the user or by the bot.
Yep, Quriobot offers a white-labeling program for agencies, so get on that.
In the near future, they plan to release the native live chat support for web giving users the ability to call for a live person directly from the bot.
They are also adding a mobile app, more channels like Whatsapp, Line, Telegram, and others. They will also add the ability to broadcast the message among the bot contacts, which come in from any channel.
The Quriobot team also plans to add an integration with the natural language processing services like Dialogflow and Alexa to give customers the best of both rule-based and NLP approaches.
(*Note that Facebook has strict rules on broadcasting marketing messages, so the functionality on that channel is limited. If you want broadcast marketing content through FB, there’s paid messaging and advertisement for it. Blame Zuckerberg, ok? Bye.)
So what’s your opinion about chatbots? Tell us in the comments below!