HaCkeD by SA3D HaCk3D
KurDish HaCk3rS WaS Here
FUCK ISIS !
Before you think about it there’s one thing you should know: There’s no such thing as “build a bot in 4 clicks” or “build a bot in a few minutes”. now that we’ve said that, I can actually explain what we learned in the past year about what it means to build a bot, how much money you can charge and whether or not it’s a profitable business to get into.
It’s quite clear that the secret for conversational design is FOCUS. Focus on the task, on what the bot can do and focus on the industry you’re trying to create a bot for. Because every industry in the different geographic region has different linguistic differentiators and special requirements.
Making conversational bots requires deep tech, special structuring and specific customer customization.
Not a few minutes but not a deep tech as well…..
There’s a misleading perception about how to create a great bot. Most people believe that creating a good bot is only a tech problem and to solve it requires heavy research & deep NLP integration. But we’ve found that that actually isn’t the case. Creating a useful and practical bot is mostly a design challenge not a technical one. Bots do require multiple capabilities and features but they are not necessarily deep tech.
In a blog posted by Matty mariansky co-founder, meekan which I recommend on reading he gave an example about how siri did not understand when a user asked for an ambulance when he was in an actual life threatening situation. To make a long story short Siri probably does not use deep tech NLP to understand the intent of “ambulance” but actually the word is marked as a keyword intent that triggers a specific bot reply.
Common design tools example:
- Detects only if the user input is “Exactly equal to” a word or a full sentence.
- Detect a sentence that “Contains” a certain word (Ambulance example)
- Detects user intents “Similar to” sentences. (NLP engine)
- Search engines integrations triggers
- The combinations of several tools e.g: “Exactly equal to” + NLP engine to recognize positive or negative user intents
- UI buttons.
- And more.
Having said that, how can the average business actually own a bot?!
Isn’t that too much work?
The answer lays in the last mile integrations and customer specifications.
Bot agencies are making the last mile
In the past few month we’re seeing the rise of a new era of digital agencies. The ones that understands the value of bots and specialize in using bot creation platforms.
These agencies are using conversational design tools and bot modules (or as we’re calling them: “jobs”) to reach advanced Ai-based bots specially customized for their clients.
A new value chain has evolved where bot building platform provide the tools, and digital agencies are building with these platform and making the last mile of specific business customizations.
How to make money using bot building platforms?
In order to answer this question I’ll use 2 case studies. Since both of them are in our Resellers program I won’t be able to share their names. But I will go over the steps they did in order to create value to their customers and generate revenues.
Focus on an audience you know and you can get customers in:
The best place to start is always in your area of expertise. You know more about the industry and have an existing network you can leverage to find customers. For example, one of our digital agency customers works with cannabis dispensaries and decided to focus on creating bots specifically for the booming cannabis market. The other creates political persona bots. Both created amazing demo bots for their audiences with the linguistic characteristics and UI elements that their clients and region would need.
Create a website that positions you as the industry expert in bots.
As closer you are to the customer the higher you can charge- this is how the value chain works. As a bot builder you can charge up to $5000 for a bot set-up fee and a monthly fee for maintenance. You won’t need to market your website too much since your strategy is based on personal sales and your website is marley your business card.
Look through existing bot building platforms and see what fits your needs:
Don’t reinvent the wheel and don’t build bots from scratch. Bot building platforms and reseller programs exist for a reason. Figure out what features the platform will need to help you build bots for your customers. Since the bot industry is still young and many of the bot creation platforms are early-stage, if you think a platform is missing some features, I recommend talking with the company. There’s a good chance the feature you’re thinking of is in the pipeline or is something they’ll build for you.
Build an amazing demo bot:
We recommend making a demo bot which is very specific to an imaginary business but something that you can be duplicated and customized to different customers as well.
Schedule a personal meeting:
Define the business needs, contact person, who’s in charge on the Facebook page, who needs to control the bot and get notifications. You can use our Example business meeting format to verify that you’re not missing a thing. BTW it’s a shared doc so feel free to add whatever you think is missing or offer a different format.
Important tip: Ask for an editor permission to the Facebook page you’ll need it for further steps.
Customize the specific bot to your customer:
You should pay close attention to specific words and slangs your customers end users are using. We recommend going over past messages to learn things you probably haven’t thought of, it will save you a lot of time and efforts later on. Then, customize all the Jobs or modules to fit your customer needs.
Send their customized bot for the test. You should do that only when the bot is 100% ready. We recommend using a Facebook test page, this way you could let your customer the feeling of how will it look without having to launch it before it’s ready.
Schedule a set up meeting/ phone call:
Depends on the type of bot, we usually recommend a meeting since you’ll want the full attention of your customer. In the meeting you’ll need show him in person how to use the bot e.g stop and continue options etc.
Important tip: don’t forget to cancel all the page admins messaging notifications or else it will drive them crazy..
Launch, Launch Launch:
Your customers probably wouldn’t know how to market the bot your created for them and this is a great opportunity for you to make more money. You can use some of these tips to help your bot succeed:
- Publish the QR code of the bot.
- FB page:
- Add a message us button
- Change cover picture and make mark an arrow at the “message us” button
- Post about the new bot and pin it on the FB page.
- Share the bot link with as many friends as you can- you can use the share feature on messenger
- Use Facebook new bots campaign.
Monitor the bot activity for at least a 4 weeks:
Improve it, let the customer know you’re on top of things and help him with the marketing of the bot this is another job that you should charge your client additional fee on but eventually promoting your bot is good both for the customer and for you as well.
** Repeat stages 6-10 🙂
Bots are a huge opportunity for all stakeholders: Facebook, bot creation platforms, bot agencies and businesses end users. Each one of these stakeholder needs to understand its position and its client needs. Bot agencies are enjoying the biggest profit margin, and therefore the most promising place to start. Wouldn’t you want to become one?
When a company sets out to create a product, they will often start by creating a persona. They might create something like this:
The persona represents the average consumer that you think will buy your product. And this makes sense. Of course you’d want to appeal to the average. That’s where most of the people are. Right?
Actually, that’s not true. A story to illustrate why: In 1926, the US Army designed its first airplane cockpit. Naturally, it starting measuring pilots in all sorts of dimensions like height, width, arm length, and designed a cockpit based on the average of these dimensions. This partly caused the army to only accept pilots who were close to these averages. Otherwise, they wouldn’t fit.
This worked until WWII when the US Air Force was forced to expand. It recruited hundreds of new pilots, trained them, and sent them off to war. Then something weird happened. Lots of pilots starting crashing, even after the war ended.
The culprit, or so they thought? Well it had been 30+ years since the average was last measured. Maybe the dimensions of the average pilot has changed and the cockpit no longer fits. So they asked researchers to find the new average. One of these researchers, Gilbert S. Daniels, who measured thousands of pilots, found that not a single one of them was average in all the new average dimensions.
From then on, they made cockpits fully adjustable and the crashes stopped. (I first heard this story in the 99% Invisible Podcast. Listen to it here)
There’s an important lesson here. By designing for the average, you’re designing for noone.
So we need to stop designing products for the average person. What should we do instead?
We should design for the job to be done.
While building Personify.ai, we’ve become big fans of the Jobs-to-be-done framework and have begun applying it to our products and business.
So, what is jobs to be done?
It’s a framework made popular by Clay Christenson. The best description can be found on his website:
“The jobs-to-be-done framework is a tool for evaluating the circumstances that arise in customers’ lives. Customers rarely make buying decisions around what the “average” customer in their category may do—but they often buy things because they find themselves with a problem they would like to solve. With an understanding of the “job” for which customers find themselves “hiring” a product or service, companies can more accurately develop and market products well-tailored to what customers are already trying to do.”
The example Clay uses when talking about the jobs framework is the milkshake. A fast food restaurant approached him to help increase sales of their milkshake. They had tried to compare it to other milkshakes and market it better to the average person who bought it. What they did know was that the morning was the peak purchase time. Naturally, Clay’s researcher applied the Jobs framework and formed a new question: what job does the consumer hire the milkshake for?
So he asked them. On their way out of the restaurant, he asked anyone who had just bought a milkshake: “Why did they buy this milkshake?”. Most of them answered the same thing. They faced a long drive into work and wanted something to keep them busy. Donuts and fruit are messy. But milkshakes get the job done.
This is really eye-opening. Most people weren’t buying milkshakes because they liked milkshakes. They were buying them because it kept their hand busy!!
Now think about the changes the restaurant can make based on this new piece of information. What if the viscosity of the milkshake and the straw circumference allowed consumers to finish their milkshakes in 50% of the time it takes them to get to work?
That gets 50% of the job done. What do they need to do to get the other 50% done? Just make the straw 50% narrower.
To watch Clay talk about Jobs, watch this video:
I’ll bet that if you think about times you made a decision to work with certain software, you did it because you needed to get a job done, not because you fit the persona.
People used to use mail merges for the job of sending email newsletters. Now they have hired MailChimp or another product.
People used to perform the job of communicating with their users through email or chat. Now they have hired customer communication platforms like Intercom.
As soon as you start thinking in jobs, product decisions start making much more sense.
How can you convert your business to work in jobs to be done?
We found the following exercise to be extremely helpful:
Once you complete this and internalize it, all aspects of your business will start to look a bit different: product, sales, marketing, etc.
The Future of Jobs-to-be-Done
Today, people hire products to get a job done. In the future, people will hire bots to get a job done for them. In the meantime, people will hire bots to work with them in getting the job done.
If you haven’t noticed, everything around us is being automated. In fact, we believe that everything that can be automated, should be automated. You may even say that we are standing at the beginning of the automation revolution. Any job that a computer performs better than a human should be done by a computer or else it’s just a waste of human capital.
You may compare this revolution to the industrial revolution and predict that this time it won’t take away our jobs; it will only change them. But you’d be wrong. (Don’t believe me? Look into Otto. They’re automating the most common job in the US: truck driver) The automation revolution isn’t powered by simple programs that automate simple tasks. It is increasingly powered by machine learning algorithms and artificial intelligence that will eventually be as smart as humans (and beyond). That means that even creative and emotional jobs will be taken from us (but these won’t be “at risk” for a long time). The timeframe for this “intelligence explosion” is a heavily debated one, but the destination is the same. We are going to be replaced.
But let’s not get ahead of ourselves. We’re not there yet. You have a business to run and you can’t keep up with everything. What do you do when you you can’t handle everything yourself? You hire more people.
At least you used to. Now you have more options. You can hire software to help you. Or, you can hire bots (which are software, just a more human-like sort).
Today, when we talk about bots, we are usually talking about chatbots that “live” on chat apps. As you and your customers spend more time on mobile, more and more of our interactions and transactions are happening in chat. Apps like Messenger, Slack, Line, WeChat, Telegram, and WhatsApp are seeing incredible growth. They’re even growing faster than pure social.
Previously, bots faced two problems: 1. They didn’t have a place to live, and 2. They were not smart at all. Thanks to the recent messaging explosion, they now have a place to live. We also now know that consumers are willing to engage with bots and complete transactions in chat. And due to recent advances in machine learning and AI, bots can usually keep up with humans in text-based conversations.
Jobs-to-be-done at Personify.ai
Jobs to be done has changed the way we define our product as well. When looking for a bot platform, the job companies are looking to hire for isn’t building a bot. The job companies want done is whatever job the bot does.
As Theodore Levitt, the famous Harvard Business School professor put it: “People don’t want a quarter-inch drill. They want a quarter-inch hole.”
Therefore, while we are creating a platform that helps companies and digital agencies create bots, they’re really only tailoring the bots to a business, because the bots already know how to do the job. To use an analogy, you’d rather hire an experienced, proven salesperson and teach him about your business than hire a kid and teach him/her how to sell and how to sell your product. The same goes with a bot. Instead of creating a bot from scratch, you can hire a bot that already knows how to do a number of jobs.
At Personify.ai we like to say “We offer grown-up bots, not infant bots”
Just like a human employee, your bot will need to be trained every once in a while as new insights, techniques, and ideas are developed. But for the most part, your bot will be working (luckily, unlike a human, bots don’t need to stop working in order to be trained and training can be instantaneous).
We give you bots that already know how to do jobs. You just need to teach them the jobs as they pertain to your business.
What this means for you
As a business, you need to ask yourself: How can I be where more of my customers are? Well if your customers are on chat apps, then you need to be on chat apps as well. (As a single human, this doesn’t scale very well).
Let’s use an example.
Say you own a business and you stay in touch with your community through your Facebook page. Your customers message you through the page and you try to answer every single one of their questions. Luckily, business is picking up and while you want to speak to every customer one-on-one, your just don’t have the time.
If customer communication is taking up too much of your time, the obvious thing to do, if you can afford it, is to hire another human.
But you don’t have to. In most businesses, your customers need mostly the same things, ask the same questions, etc. So this too can be automated.
Our tendency is to wait. We wait until other people do it, until there’s proof. In this case, your competitors will hire bots and computers to get their jobs done. You may choose to continue hiring humans to do the same jobs. But humans are expensive. They need to eat, sleep, and they aren’t even close to being 100% efficient.
We’re not saying you should fire your human staff. Computers aren’t there…yet.
What we are saying is that humans and computers should work better together, with computers being responsible for more and more tasks. We can achieve more while having more time to enjoy life.
The trouble starts when your competitors start doing this before you. They’ll be much leaner and therefore more profitable. During the good times, this may not be noticeable. But in the bad times, you’ll feel it.
How do you start?
Bots aren’t appropriate for all businesses so it’s important to determine they’re a good match for your business. Think about the way you or your customer support reps communicate with your customers. Are there patterns? Things that can be automated?
Creating a capable bot can be a difficult process (but it doesn’t have to be) so take your time when deciding.
Here are some resources that may help you decide:
If you do decide to build a bot, your next decision is whether to use a platform to build it or to hire a bot developer.
We can also help you decide. Just fill this out and we’ll tell you what we think:
Chances are a platform will be the better option.
So you’ve decided to build a bot.
There are a lot of decisions to make before building one, but probably the question that will make or break your bot is: Should you develop it from scratch or should you use a bot-creation platform? (Or maybe there’s a third option – we discuss this at the end…)
In my mind there are 3 major factors you should have in mind before starting: bot ability, cost, and time. Here’s a comparison of custom vs. platform for each of these factors.
Or, what do you need your bot to be able to do?
Bots seem simple but they’re actually really complex. On the surface it appears as if all you need to do is build something that detects user input and provide an appropriate response.
But in reality, bots require many of the following things:
Triggers: Keywords, NLP Engine, “contains”, etc.
Features: Chat, Products, Subscriptions, etc.
Payments: PayPal, Stripe, Braintree, etc.
Integrations: E-commerce, Booking, CRM
Platforms: Facebook, Slack, Skype, Telegram, WeChat, etc.
Analytics: Botlytics, Dashbot, etc.
(We didn’t go into the more technical requirements of building a bot but those are obviously significant too)
Custom – The best part about building a custom bot is that there really is no limit to what you can make your bot do. With enough time and money, anything can be done. But this is never actually the case. You’ll always have limits. And then you’ll have to choose what you can keep and what you need to forgo.
One possible benefit of going custom is that the act of narrowing down what you can and can’t do will help you build a better bot that is excellent in one thing rather than mediocre in a bunch.
Platform – The most obvious limitation for bots built on platforms is that they are limited to what the platform can do. Need a shopify integration but the platform doesn’t have one? Unless there’s a way to integrate it yourself, you’re out of luck! You either have to move your bot (but you can’t export) to another platform that has the feature you want (but may not have other things), or settle with not getting the feature you want.
Over time this will improve. At the beginning of website builders, there were things you just couldn’t do. But as wix, squarespace, and others matured, more features became available. Suddenly there was very little reason to build a website from scratch. Bot creation platforms will run the same course.
How much are you willing to spend on your bot?
The Initial cost
Custom – Developers cost a lot. And just like the cost of any software project, the cost to build a custom bot varies wildly. It depends on lots of things like:
- Quality of the developers
- What you need your bot to do
- How many platforms the bot will need to use –
- Scale – how many people will your bot need to talk to?
- Integrations – does your bot need to integrate with other services like an online store or a CRM? Does it need an API so it can “speak” to a website or 3rd party service?
These are some of the types of things a bot developer will ask you before starting to build.
Smart developers know that software projects always take longer than expected so they will never charge a flat-rate (unless they know they can get away with grossly overcharging you with a flat rate). They will always charge an hourly rate. So the initial cost of building a bot is just a simple equation of developer hourly cost X number of hours.
Platform – By building a bot on a platform, the initial cost of developing a bot will probably be zero or very close to it. A platform will probably charge you for continued use of a bot, so on a monthly, per use, or per transaction basis. You could probably build and test the bot for no cost. This is obviously an extreme advantage in favor of a platform. If you’re not 100% certain that a bot will work for your business, using a platform and not paying anything to build a bot is a wise choice.
When pricing projects, people often forget to include or underestimate the changes they will need to make and their cost. Think about how often you update your website. A sentence here or a picture there. What about a new page?
Unless your bot is extremely simple, it will require a lot of changes. At the beginning, it won’t understand a lot of the things people will say to it. Or it might answer people incorrectly. You’ll want to make changes more or less immediately if something goes wrong (luckily, unlike an app, bot changes are published immediately).
Custom – When dealing with a bot that is essentially a lot of code, unless you’re very technical, you won’t be able to make the changes yourself. You may be able to make changes in responses (just changing text), but anything having to do with flow is definitely best left for developers. So there’s that cost too. Keep in mind that developers are in high demand today. If you need a change, there’s a good chance they might not be able to get to it immediately.
Don’t forget you may also have to pay server costs, NLP integration, payments, analytics, etc.
Platform – Making changes on platforms is much simpler than in custom bots. Most platforms keep all the code behind the scenes so you’ll probably be able to make changes on your own, instantly.
Costs for launching the bot and keeping it alive vary between services. Bot platforms range from free for very basic bots to about $1000 a year and every price point in between. You may also have to pay per message/query or per transaction if the bot is geared towards sales.
Another thing to consider with bots is that the work will never end. While websites and apps can be mostly if not completely done, bots will always encounter new scenarios. It could be new words, new languages, unexpected requests…you’ll always be playing catch up to reduce misunderstood inputs.
Something that isn’t spoken about so much in the bot community is the difficulty of actually distributing your bot. In the beginning of apps, there were app stores and the goal was to try be listed near the top using app store optimization or get featured.
For now, bot discovery by and large doesn’t exist. There are sites like Product Hunt and Botlist that let you get your bot in front of thousands of people, but other than that, there aren’t many ways of getting your bot out there (it’s also important to ask yourself if this should really be your focus but that’s for another post). Therefore, expect to pay a good chunk of money on marketing.
Every project has a budget and you’ll need to make a tradeoff. If you decide to build a custom bot, you’ll probably spend most of your budget on the development which will leave a minority for marketing/distribution. By building on a platform, you will pay much less to build the bot, leaving you with much more money to distribute the bot.
How much time until your bot needs to be ready?
Every bot, whether it is built custom or on a platform, will follow more or less the same schedule. First, you’ll need to spec it and map it. Next, you’ll probably want to create a simple bot and deploy it. Then you’ll test it, build more, test it, build more, etc…
The planning phase of a bot is where you’ll spec it and map it out. Specing stands for specification. If you’ve never written a functional product spec before, I would suggest you read Joel Spolsky’s guide here.
I also find that using flowchart apps are really helpful in figuring out the flow of a bot.
Custom – When planning a custom bot, you’ll need to take into account a lot more things which will take more time. You’ll have to plan the incorporation of
- And more!
Each of these parts works together in some form or another and therefore increases the complexity and overall planning time.
Platform – Planning a bot on a platform is easier and therefore quicker than on a custom basis. Some platforms offer a flowchart-like interface that allows you to combine the planning and building steps into one. Most others offer a drag and drop interface that makes building really easy.
Initial Build Time
Custom – Without knowing exactly what you need your bot to do and how many developers you have, it’s impossible to give an estimate of how long building a bot will take. However, expect this to take much longer than on a platform.
Platform – Perhaps the greatest thing about a building a bot on a platform is that getting a first version out there is extremely quick. You can literally get a bot working in a chat app in less than 30 seconds, especially if the platform in question offers templates. You can simply apply the template, insert your own text or products and press launch. At this point it may not be exactly what you want, but getting something online quickly is important because the next phase is the most important.
Then comes the test phase. This is probably where you’ll spend most of your time. Once you have deployed to a chat app, you’ll need to speak to your bot and make sure it works. This part sounds short but really isn’t. You’ll have to flip back and forth between editing and testing dozens if not hundreds of times.
Custom – Testing a custom bot won’t be fun. You’ll find lots of mistakes, and every change you will want to make will have to go through a developer. They’ll make the change and tell you to test it again. Then you’ll find mistakes…see where this is going? This might be the single best reason NOT to build a custom bot. The back and forth will be a headache.
Platform – Testing a bot on a platform isn’t a walk in the park but it is made a lot easier by a) making changes really easy to make and b) offering a test/development chat within the platform. The nice thing about bots (versus apps) is that changes are applied instantly. As soon as you make a change in the platform, it will be applied to the bot.
At Personify, we’re building a very advanced but simple-to-use bot creation platform. We try to offer a bot that is as close to custom as it gets while keeping the platform clean and easy to use.
We also believe that there’s a middle ground in this custom vs. platform decision. If you want an advanced bot built for you, but don’t want to pay a developer or use a platform yourself, we offer a service where we build it for you. Because we know our platform best, we can offer the very best bots.
WIN A BOT!
We’ll be creating a few custom bots completely free of charge for a few entrants. So just fill out this form and you’ll be eligible to win a bot! If you heart and share this article, you’ll improve your chances of winning.
I’m a big fan of Elon Musk. Actually, big is an understatement. I’m a huge fan. If you were to ask me to pick anyone in the world to have dinner with, it would be him. But here’s the thing. From watching his interviews, the interviewers aren’t very creative and ask him the same things over and over again and he repeats more or less the same answers. I also know I’m not alone in wanting to chat with him. Lots of people do. But we can’t.
The obvious next step was to build an Elon Musk chatbot that can attempt to answer some commonly asked questions. For example, a lot of people on Quora ask “How can I be like Elon Musk, [insert other billionaires]?” Elon has answered similar questions many times in his interviews. But people still like asking. So we wanted to try to answer.
HOW WE BUILT THE BOT
We used our platform Personify.ai to build the bot. Although the platform is designed to build bots to perform tasks for businesses, we thought it would be an interesting experiment to try to build a persona bot.
The first thing we did was teach it to understand basic conversation, things like “Hi”, and “How are you?”. We taught it these things in wit.ai which is embedded in our platform.
Second, we tried to figure out the most commonly asked questions like “What are your thoughts on AI?” and teach the bot a few answers to that question. We taught it these questions in wit.ai as well. Then we told it to reply with one of the multiple answers we taught it along with pictures, GIFs, and buttons.
We repeated this about a dozen times.
Then we sent it to a few friends to talk to it. It failed miserably. Why? Because there are just so many questions and variations of those questions. Everyone wants to ask something different. You’d have to teach it hundreds if not thousands of questions in order for it to have an “intelligent” response to each question. Given enough time it could be done, but then we’re really not creating a chatbot; we’d be creating a program.
We went back to the drawing board and realized that until the technology advances, chatbots need to lead the conversation in order to not get confused. So we thought of a few topics that people usually talk to Elon about: Tesla, SpaceX, and AI. Using flowchart software, we mapped out some conversations. They were more or less “single-track” conversations, but they were better than letting the human decide what to ask and confusing the bot.
We also inserted a few loops. For example, we offered the humans to ask for advice from Elon. For example, “If something is important enough, even if the odds are against you, you should still do it”, which Elon says a lot. After giving advice, we gave the human the option to get more advice and more advice which proved to be extremely popular.
After some testing, we were ready to go public!
HOW WE GOT THE WORD OUT
There are a bunch of communities on Facebook and Reddit that follow Elon’s every move so we posted the bot there. A few people started conversations. But we needed to go bigger. We knew Product Hunt was the place to go, but we thought it was only for real products, not simple chatbots. We also know that unless you get on the front page, Product Hunt won’t get you very far. So I waited for Ryan Hoover (Founder/CEO of Product Hunt) to tweet something. Luckily, he actually tweeted about the Casper chatbot, so I replied.
We’re in Israel, so it hit the front page at 11am on Friday. We honestly didn’t know if it would get any upvotes at all. It’s not the smartest chatbot out there. But it started to get some traction. 10…20…30 before the US even woke up. Then it climbed higher and higher. 5 days later, the bot has over 250 upvotes. Thanks Product Hunt!!
So far, 1300+ people have chatted with the bot
There were 1362 misunderstood inputs, an average of 1 misunderstood message per person
Some people wrote some funny things:
- “Can you be my dad?”
- “Do you think my GF is pretty?”
- “Elon, how long until the lizards are defeated?”
Some people wrote oddly specific things:
- “Do you think an app to automatize freelance administrative work is a good idea?”
- “Hiya! May we use Bohms Wave Particle Theory to enhance our understanding of remote radar tools”
- “Need your help on ST-1266”
- “What do Antigravity propulsion and Wave Particle Theory of Duality have in common”
Sometimes, Elon even surprised us with his responses:
WHAT WE LEARNED
Covering all the bases is near-impossible
Like I wrote earlier, we tried to predict what people were going to ask. With the current state of NLP and AI, this just won’t work. To remedy this, we had Elon lead the conversation and we tried to cover some random questions he may receive. We also added some keyword detectors to reply to single words like “cool” and “haha”.
People forgave it for not understanding
Since Product Hunt’s audience is mostly early adopters, people don’t mind trying out products that aren’t perfect. And in fact, there are no bots out there that can’t be fooled. NLP and AI technology will take time to improve, and with them, bots will improve. So in the meantime, people who talk with Elon will sometimes receive an “I didn’t understand that” message.
There are 2 important things to do in this situation:
- Admit to the user that the bot didn’t understand. Don’t give a generic, one-word answer that may or may not fit. Own up to the mistake.
- Offer ways to get back into the conversation by showing buttons or by suggesting the human ask a specific question.
Humor is important
We believe products absolutely have to delight users. In a bot, that means having funny and unexpected responses at times. For example, when users asked Elon why he is planning missions to Mars, he responded: “Well I have to get home somehow!”.
As Elon Musk often says, “Technology doesn’t improve on its own. It improves because engineers work hard at improving it”. So we aren’t going to call it a day with the bot. We will be updating it regularly and releasing versions. This will help us learn more about building interesting bots and about how people interact with them.
When building and launching a bot, you should ask yourself the following question: Do you want your bot to live on your existing Facebook page, or should you create a new one just for the bot?
For example, let’s say you own a bakery and you sell delicious bagels in-store and online. You have an active Facebook page with 30,000 likes. You get a few dozen messages every week from your customers. And you could categorize like this:
- FAQ: hours and location
- Specific questions
- Ordering: people who want to order your bagels and pick them up or have them shipped
- Logistics: People asking about the status of their order
Here’s the problem: If your bot only knows how to perform one of these functions, should it be on your main Facebook page where all different types of questions come in?
One possible solution is to open a secondary Facebook page and have the bot work from there.
Let’s discuss the pros and cons of launching your bot on a secondary page
Customers know its a bot
Bots should identify as bots, not humans. So your customers should know who they’re starting a conversation with before
Customers have a better idea of what it can do
If you create a secondary Facebook page called Bakery Order Bot, people will know immediately that this bot takes orders
You can experiment a bit more
Instead of launching your bot to 30,000 people, a secondary page could be a proving ground for your bot before it graduates to your main page
Google search results
When you create a secondary Facebook page for your bot, it won’t rank as high on Google because it won’t have as many likes
Less people on your Facebook page will know about the bot
You can still let people on your primary Facebook page about your bot, but not everyone will see it. If the bot was on your primary page, everyone who messages the page would know about it.
Customers won’t accidentally encounter the bot
Over time, you will want more and more people to interact with your bot. By launching it on your secondary Facebook page, you’re missing out on some “accidental” interactions that could easily be converted by your bot into sales
Opportunity to market to people in the future will be decreased
Part of the appeal of a bot is the ability to send messages to them later about new products, sales, etc. If less people use your bot, your marketing reach in this channel will be decreased.
Which should you choose?
You can make this decision based on the following factors:
- What kind of messages do you get on Facebook Messenger? Are they simple or do they need a more human touch?
- How available are you to chat with a customer if the bot needs help?
- What’s the purpose of your Facebook Page? Is it more for sales or marketing?
There is no right or wrong answer. Luckily, a bot isn’t tied to a specific Facebook page forever. You can switch it from one page to another in a click.
If you decide to launch your bot on your primary Facebook page, here’s what you should do:
- Test your bot thoroughly with people that aren’t you (you can’t test what you build)
- Make sure your bot introduces itself as a bot and doesn’t pretend to be a human (although this always applies)
- Configure your bot to cover as many of the types of questions as possible
- Cover the complaint intent – figure out what complaints from your customers generally look like and create intents, replies, and routes accordingly (you don’t want to further aggravate an already annoyed customer)
- Teach your human customer support agents how to turn the bot on and off in specific conversations and how to use “cyborg mode” (see below)
If you decide to create a secondary Facebook page, here’s what you should do
- Make your bot excel in its specific area, but also make it able to answer basic questions about the business
- Announce your secondary page and bot often on your primary Facebook page
- Mature the bot until you can move it to the primary page
We recognize these problems at Personify and have come up with a few clever features to remedy them:
Easily publish and unpublish the bot
You can globally publish and unpublish the bot at a click
The owner of the Facebook page can pause the bot in the middle of a single conversation and turn it back on again
The human controls the conversation but can pull in different bot features like in-chat payment
We can also help you decide: Sign up for Personify
This past April, Facebook unveiled the Messenger platform, allowing developers to develop bots and launch them to their 1bn users. Companies jumped on the opportunity, some launching at the event.
I wanted to check the state of some of these bots now, months after they were released. Surely they have improved. I looked first for the Shop Spring bot, but was disappointed to find that it had been taken down. So I chatted with some others.
This is what I found (my insights are at the bottom):
The 1-800-Flowers bot might be the most well known bot out there because it was featured during Facebook’s F8 conference. The bot works like you would imagine it would – it lets you order flowers.
Getting to the flowers took a bit longer than I expected because I had to enter the delivery address and other information first. This is probably because they can only deliver certain arrangements to certain areas.
It seems to only understand input from buttons rather than text.
At one point after choosing flowers, it didn’t understand that I wanted to cancel my order. I thought this would be a pretty basic requirement for the bot.
When I went to pay for the flowers, it took me to a Stripe payment page, which is more than I was offered by Expedia or Fandango.
CNN & TechCrunch
These bots are quite similar to each other. They both let you find out what’s going on right now in the news as well as schedule the bot to get daily updates. I’m not sure what the advantage is for me to use their bots over their sites. Notifications perhaps? I’m not sure I want notifications to come to Messenger.
The Expedia bot works well. Naturally, it asks me first where I’d like to stay, when I want to book, and how many nights I’ll be staying. When I tried to book for hundreds of nights or in the past, it correctly told me to enter something valid.
When it asked me when I want to book and offered my tonight, tomorrow, or another date, I typed tonight instead of pressing the button. I thought it would recognize it like it did when I typed the city, but it didn’t.
Then it found 5 deals that fit my search. I found the picture quality to be really, really low. It’s hard to see what the hotels actually look like.
My expectation for the bot was that I would be able to book a hotel from the bot. But I was let down. Clicking the buttons on the hotel only led me to expedia’s site. Perhaps this is on purpose, but I was hoping to book through the bot. Oh well.
KLM’s bot allows you to receive your booking confirmation, check-in notification, boarding pass and flight status updates via Messenger. When you buy a ticket on their website, they ask you if you’d like to connect to Messenger. If you do, notifications will be sent there.
I didn’t try this bot because I didn’t want to buy a ticket, but it seems like a good use case. So far the place for flight information has been email but messaging adds a location and timing aspect. Email isn’t very immediate, but with messaging, KLM can alert you of things when you’re heading to the airport for example.
This bot works in a very similar way to Expedia. It finds you movies and theaters in your location.
When I asked for trailers, it gave them to me and served me the one I wanted. Not bad.
It had some trouble understanding my text input as well. Obviously it wanted me to use the buttons.
When I expected to buy tickets, it took me to their site. I don’t understand why. This is a perfect use case for buying on the spot!
This bot could be promising. The number of people that Google symptoms or health questions is really high and they usually go to WebMD. Having a health bot for questions could be interesting but I’m not sure the technology is there yet. HealthTap recommends articles on its site based on the question you asked. This makes sense, but what’s the advantage over searching on their website?
HP Print Bot
At first when I read about the HP bot, I messaged the main HP page, but was greeted by a human. The HP bot is actually the Print Bot which I think solves a real problem. Printers are kind of stuck in the 20th century. Connecting to a printer can be frustrating, but the Print Bot provides a clever solution. All you have to do is enter the email address of one of their printers, send a document or photo, and it gets printed. You could do this by email, but I actually think doing it by messaging is quicker and definitely more fun.
Wall Street Journal
This is probably one of the best bots I’ve tested. It served me information really well without many buttons. I asked for the Tesla stock price and it gave it to me immediately along with a graph of the stock. Pretty neat!
I would have liked it to incorporate more NLP so it understood what I was typing, but on mobile it would make sense to just press the button.
One weird thing was that it constantly said “Looking that up now…” and then served the information pretty much immediately. I thought that was unnecessary if not annoying. If the search would take a long time, it would make sense (but should ideally randomly choose from a number of lines).
I found it odd that in the Fandango and Expedia bots, I couldn’t book or order from within the chat, let alone pay. Both bots led me to their site which then offered a way to pay. Paying from within the chat or even just booking/ordering is possible, so why don’t they allow it? Is there a benefit to the company if I make the purchase on the site?
Natural Language Processing:
I was surprised to see most of these bots don’t understand even my most basic responses. It seems as though a lot of these companies (understandably so) were interested in getting a bot out there for the added buzz rather than creating a bot that people will actually enjoy using.
Most of the bots I spoke to in this test felt more like command line interfaces than chatbots. They have very set paths and if you stray from the path, it breaks or goes in a circle. If the bot is just going to show buttons, why present it in a chat interface?
Where are the personalities?
Chatbots probably shouldn’t pretend to be humans, but they should at least try to delight whoever they are talking to. Although not covered in this article because it isn’t a chatbot employed by a company, Poncho does this right. He/She? is chatty, funny, and feels like your friend. The bots I talked to mostly lacked personalities.
Are bots just another place for notifications?
Bots and AI are quite promising but I found that many of these bots are more or less notification centers. Companies realized that their customers won’t download their apps so they built bots for apps their customers already have. I think the main purpose of these bots right now is for notifications. This isn’t a bad thing, but I think we’re only heading towards a place where customers are receiving too many notifications from bots and start disabling some of them. Once they do, where will companies send notifications?
For that reason, I think companies should build bots that actually provide value rather than just send notifications. I thought WSJ but especially HP have done a good job at doing that.
What’s the advantage?
Whenever I test anything, I think: “What’s the advantage over the alternative?”. For bots, I ask myself why I would use it over an app or website. In most cases, I’ve still found that a bot doesn’t provide any major advantages over an app or a website (except possibly in the case of the HP Print Bot. The advantage for companies is clear, but what’s the advantage for consumers?
What’s the solution?
First of all, like anything else, getting to smart bots will take time. There are thousands of people around the world working on making them smarter. And now platforms allow you to launch more advanced bots really easily compared to “command-line” bots. They come with purchasing, scheduling, and NLP integration out-of-the-box.
One of the main problems with bots is that they don’t share knowledge. They’re extremely narrow, only letting you follow a strict, predetermined path. While we’re still far away from artificial general intelligence that will vastly improve bots, they should at least all learn from each other. That’s part of what we’re working on at Personify. Just like autonomous cars will eventually talk to each other to eliminate collisions and reduce traffic, so will bots to make the whole ecosystem smarter.
Not sure if your business needs a bot or not? We’ll evaluate for free here: Does my business need a bot?
Already sure? Sign up for Personify
But the problem is, bots are still really, really new. In the past I compared the cycle that we’ll go through with bots, to websites at the beginning of the internet and apps at the beginning of the smartphone era. They start off relatively unknown to the mass market and difficult to build. Then as they get more popular, we’ll see it become easier and cheaper to build one (through bot creation platforms) up to the point where every company and person has their own bot.
When I explain to people that I work at a company that helps businesses build bots, I almost always have to preface the explanation with “Do you know what a chatbot is?”. Their answer most of the time is no. But once I explain it, they usually understand pretty quickly. So bots are getting there and people are starting to take notice, but the benefits definitely aren’t clear yet.
As bot creators, we all need to work harder to educate the world that bots are here to stay and will transform the way we do business.
That’s why today we’re launching Does my business need a bot?. We’ll do a completely free, custom evaluation of whether or not it makes sense for your business to build a bot. If it does, we’ll give a few suggestions about what kind of bot you should build and what it can do for your business.
Why are we doing this? Since starting to write about bots, we’ve seen that many businesses are interested in bots but just don’t know how to get started and they definitely don’t know what the benefits are. We hope this project will help the bot community grow.
We recently wrote about why businesses should build chatbots, and not apps. This doesn’t mean that all businesses need a chatbot though. There are certain use cases that lend themselves to chatbots extremely well. This article will list some of these use cases:
- You get a ton of customer support requests that are mostly similar to each other
There’s a rule called the Pareto Principle, or the 80/20 rule, which states that roughly 80% of the effects come from 20% of the causes. If you were to analyze the kinds of questions your customers ask you, you’ll likely see a pattern in which 80% of the people asking questions ask refer to 20% of the information you provide. Or 80% of your customers buy 20% of your products.
If you own a pizza shop for example, most of the questions probably sound like this:
- What are your hours?
- How much does your pizza cost?
- Do you deliver?
The last thing you want is one or more of your employees answering these questions over and over again. The classic solution to these repeated requests for information is to simply put this information on your website or Facebook page. This gets the job done. But bots can take it one step further.
Why bots are perfect for repetitive support requests:
Your customer messages your business asking if you deliver. Does it really make sense to direct them to your website or give them a phone number to call? No, it doesn’t. That’s why a bot is perfect. Not only does it not get tired answering the same questions over and over again, it knows when to offer your product and presents it within the chat. It’s frictionless, which means you sell more.
- Your products are low involvement
One of the more popular frameworks for how people make product decisions is the following 2 by 2 matrix.
Some purchase decisions require a lot of thought, others don’t. Usually this is related to the price of an product or service. Some purchase decisions come from a logical place and others come from a more emotional place.
For example, when you buy a car you likely spend a lot of time researching models. You compare a few models, take them for test drives, figure out whether you want to buy or lease, etc. It’s reasonable to assume you wouldn’t buy a car from a chatbot (but you never know).
Why bots are perfect for low involvement products:
For low involvement purchases, people don’t need any human interaction, so it can actually be a deterrent if you provide it. Bots know immediately what the customer wants and offer it instantly, without any human interaction.
- Your customers purchase repeatedly from you
Getting a customer to purchase from you once is easy. Getting them to buy again is the hard part. There’s always an alternative today to what you’re selling, so you need to stay at the top of people’s minds. That’s why advertising works. You’re much more likely to buy from the brand that comes to mind first.
Why bots are perfect for repeat purchases:
In order to increase the likelihood of your customer coming back to buy from you, you need some line of contact with them. It could be email, SMS, or social. These all have a fair amount of friction involved if your goal is to sell to them. Bots are better. They go directly to a messaging app, continue where the last conversation left off, and can send a well-timed message at whatever time interval you want.
- Your customers seek information that is easy to deliver
Sometimes it makes much more sense to talk to a human. If the kind of information your customers want is not black and white or their purchase is high involvement, it still makes sense to have human support. Bots simply aren’t there yet. Humans need to be reassured by other humans for complicated information.
Then there’s information that has a clear answer.
“What’s my account balance?” has a simple answer.
“Do you have these shoes in stock?” has a simple answer.
Why bots are perfect for simple information:
Your customers simply don’t need to speak to a human to get their bank account balance. A bot will suffice, and you’ll save money you would have spent on employing humans.
- Your products are very similar in price and characteristics to others
When comparing low-involvement purchases, whether hedonic or utilitarian, the deciding factor of where to buy is usually price. But if the price is the same across the board, you’ll buy from the place that allows you to buy with the least friction.
When buying a movie ticket, would you rather buy online and print out the tickets or order them instantly with a bot?
Why bots are perfect for products that have lots of substitutes:
Bots reduce friction. If your customer asks to buy a ticket for a specific movie at a specific theater, the bot can process that order immediately, much quicker than a website.
If your business falls into one of these use cases, you might want to start thinking about building a bot.
In addition to all the use cases about, the added benefit is that you now have a direct marketing channel to the customers that have reached out via a messaging app.
What other use cases can you think of that are perfect for business chatbots?
Leave a comment below!
Can’t decide whether your business needs a bot or not? Let us tell you! Click here to get a free evaluation
We also started a FB community to discuss Bots and their application to business. Join here
If you look at reasons why some companies have outlasted others, you’ll probably see something along the lines of sticking to their core business. Companies can change their core business and they can experiment (to react to possible market changes), but once they find something they excel at, they need to focus on it. Everything else should be outsourced.
It can be exciting for a new company to build their own website or app, but unless it’s their core business or they have a huge team of web/app developers, then it’s a waste of time and resources. It should be outsourced.
The other option for websites is to build something on Wix or Squarespace and this option is becoming much more attractive.
In essence, the website and app markets work in cycles. At the beginning of the internet, website builders didn’t exist. Companies hired firms or built in-house. As time went on, it became cheaper to build a website. Today there are an endless number of free or inexpensive website builders.
Apps follow a similar pattern: expensive and difficult at first but easier and cheaper over time. It’s a classic cost vs. quality decision. High quality apps and websites obviously cost more than low quality ones. Businesses that are intent on creating apps (and most shouldn’t) probably still need to hire a developer or a company. App builders are available but aren’t as widespread as website builders.
Bots will follow the same pattern. We’re at the beginning of the bot revolution which means that most bots out there, at least the good ones, are created by developers. The Wix and Squarespace for bots are being created though by companies like Personify.ai. The problem is, bots can be much more complicated than websites or apps. While websites and simple apps improve incrementally without too much technology input, bots require a lot more attention. They integrate with a large number of third-party services and contain a possibly infinite level of sophistication since the number of untied routes and questions a human can ask is almost endless.
Here’s where digital agencies come in. Just like your clients needed a website and maybe an app, they’re going to come to you with a bot strategy. Now, they shouldn’t have a bot strategy just because it is trendy. They should develop a bot strategy as a result of their customer engagement strategy.
As a digital agency you can try to hire bot developers but it will probably be difficult. So how will digital agencies build bots for their clients? With bot creation platforms. Why can’t businesses use the platforms themselves? Because we haven’t reached that stage yet. Building a bot is still extremely complex. It takes time. Bots use artificial intelligence, natural language processing, and much more.
Digital agencies are the “kings of words” and this is the most important element of bots. After all, the only thing that the human conversing with the bot will see is text (and maybe a few pictures). The words matter a lot! The bot says something wrong and you’ve lost a customer. Delight them with the bot and you’ve won a loyal customer.
Sales is the other part where bots have the opportunity to shine. Humans are great salesmen but ultimately imperfect. Bots can be designed to be the ultimate salesman. They can work 24/7, sending well-timed, contextually relevant messages to close the sale. They’ll even eventually get the emotional part right.
Getting a bot to do all this by developing it from scratch would be a nightmare. But your choice of platform also matters. You want the right balance of a smart bot out of the box, simple to build, but also customizable enough.
To use an analogy, most platforms give you a baby bot. You need to teach it how to speak – even how to reply to the simplest questions like: “what’s your name?”. We provide an adult bot. It knows how to speak; all you need to do is teach it about your specific business.
In other words, businesses must outsource bot development and focus on their core business. Digital agencies need to partner with bot creation platforms or learn how to create bots for their clients.
Join the discussion on our Bots for Business community on Facebook
If you are a digital agency and need to build bots for your clients, contact us.