How To Build A Charming (And Useful) Chatbot
I’ve been working on my chatbot, GrowthBot, since April 2016. Today it has 95,000 users and answers 15,000–20,000 questions every week. Yes, I’m a little proud 🤓
This has been a really fun project for me. It started because I was curious about conversational interfaces and natural language processing, but it’s turned into more than that. You might say I’ve become a bit obsessed with chatbots.
I’ve been writing software for well over two decades, but building a chatbot is unlike any software I’ve ever built before.
After several decades and platform shifts in software (client-server, Internet, mobile, etc.) we are now finally able to build something that approaches an “intuitive” interface. As an industry we’ve been saying we’ve created intuitive software for years, but at the end of the day most software isn’t intuitive at all. It requires taking your goal (what you’re trying to get done) and intersecting it with your knowledge of how the specific software you’re trying to use works — and translating that into a series of clicks, touches and keystrokes.
What you want is an answer to the question: “How many customers did we sign up last month vs. a year ago?”. What you have to do is understand the software and figure out the magical sequence of things you need to do to try and get to that answer.
With conversational, NLP-powered interfaces, we can build software that allows users to just state their need and we do our best to provide an answer.
That’s what I’ve tried to do with GrowthBot. It supports a multitude of different questions that marketers and sales people often have.
I think this shift to conversational interfaces is bigger than mobile. Mobile was interesting but it wasn’t that big of a shift. We were still basically touching and swiping instead of clicking.
Mobile applications were often simpler than their desktop counterparts — but they weren’t quite magical.
We added some sensors to mobile devices (most notable being the camera) which changed the scope of what they could do — but for business professionals, the software still pretty-much worked the same. You had menus and buttons and notifications.
Chatbots are going to make information exponentially more accessible in a way that is both friendly and effortless. (And I’m putting my money where my mouth is on this one, HubSpot just announced our acquisition of one of the top, visual, bot-building tools).
Chatbots will allow businesses to instantly respond to customer questions, regardless of the platform the customer is on, or time of day. Customers will grow to expect this level of service — the ability to find answers not through a navigation menu, but just by typing out words (or asking Alexa). And it’s going to change the way we sell, market, and serve customers.
Excited? I am.
But like with search engines, the early years are going to come with some important lessons. Here are a few I’ve picked up from my first year with Growthbot:
Building a great chatbot is equal parts about technology, psychology, and copywriting-ology.
Technology
All chatbots are powered by natural language processing (NLP), a type of artificial intelligence. This gets a lot of attention and it is, admittedly, cool stuff. But NLP only addresses a chatbot’s ability to understand inputs. To get good outputs from chatbots (or any AI really), you need tons of data.
Remember the year of the app? The end result was millions of apps that solved extremely one-off problems. No one wants to download an app for that corner pizza store they go to, or one that gives you directions to your nearest Starbucks and a camera app (with filters!) provided by your favorite brand of yogurt. We just don’t have space in our heads or on our home screens for a thousand different apps. It’s ridiculous. If we’re not careful, chatbots are extremely prone to this same failure of imagination.
If you want people to use your chatbot, it needs to provide something uniquely valuable, not just a different way to do something.
Part of what makes most chatbots today so useless is that they don’t have enough data feeding them. Your average chatbot is less intelligent than your average email service provider. Why? Because your ESP can connect to your CRM, it has all kinds of data about customer lifecycle stage, purchase history, click data, etc. that you can use to create personalized emails. But most chatbots don’t have access to that kind of data.
Every time I give GrowthBot access to a new data source it becomes much more valuable. Today GrowthBot provides data from HubSpot (of course), Google Analytics, twitter, MailChimp, Instagram, and over a dozen other proprietary data vendors. Some so obscure, you might not even know they exist — but that answer very relevant questions.
- Who are the top Instagram influencers on SaaS?
- What PPC keywords is freshdesk.com buying?
- How’s the open rate on my MailChimp campaigns? (and how does that compare to other companies?)
- What is the top marketing automation software?
- How much organic search traffic does dropbox.com get?
- Give me a personality profile for dharmesh@hubspot.com
GrowthBot can solve hundreds of these little research questions because it’s a data scanning machine, and it can answer these questions for hundreds of people at a time.
Data sources are one technology component needed to provide useful outputs. The other is conversational interfaces.
One of the problems with the abundance of communication channels today is that they have a tendency to shift around.
— Send someone an email and they respond in Slack.
— Chat with a company on their website and a different salesperson will follow up on email.
— Facebook message a friend and they’ll send you a text.
This behavior isn’t going to change, so it’s something companies need to adapt to.
Today GrowthBot can be used on Slack, Facebook, and Twitter. What it can’t do (yet) is port those conversations. Imagine a chatbot you deploy to help your customers that exists on your website, Facebook, Slack, even Alexa and Echo. That’s powerful stuff, right? And if all these conversations are recorded in your CRM, a customer can resume a conversation at the right time and place for them.
How freaking delightful is that?! Oh, so very.
Psychology
The psychology of chatbots is largely about changing behavior. It’s not impossible for humans to do what chatbots can do, it’s just that when we learn to work with them, they make our lives so much easier. You can spend an hour searching Instagram and you’ll find the top influencers on SaaS, or you can ask GrowthBot and find out in 2 seconds.
But first, you have to change the “search” habit to “ask a chatbot” — that’s where psychology comes in.
I have 3 hard-won lessons on this front:
1.The onboarding has to be easy and example based. Show the user examples of what the chatbot can do. If they get lost, make it straightforward for them to see that list again. Let them run those examples easily for themselves so they can see the output. This builds familiarity and trust.
You want your bot to feel charming and familiar, like a helpful friend. Never miss an opportunity to show people what they can accomplish with your chatbot.
2. Create good reasons for the user to come back — either because you’re solving a frequently needed use case or because you have permission from the user to send them a message when something of interest comes up. This helps you pull back users that have become dormant.
Remember, your chat logs are basically the focus group to end all focus groups. When your users try to ask your bot to do things it can’t do, they’re telling you what they expect from your bot. Read your chat log every night and use that data to build a more useful bot.
3. Where appropriate, try to minimize the typing the user has to do by using the bot platform’s simple UI widgets, but don’t get carried away. The UI widgets shouldn’t be the only way to do everything (that loses a lot of the potential for chatbots), you want to use it to supplement the standard chat interface.
Bottom line: The best way to create a habit around your chatbot is to be useful again, and again, and again, and again…
Don’t expect adoption to be quick and easy. Provide something of value and then kindly show up and remind your users that you provided them with something valuable and it just became more valuable.
Copywriting-ology
Chatbots exist in a weird world where we know they’re not human, but we kind of wish they were. If you want people to use your chatbot, it can’t sound like a robot.
There’s actually an easy trick to give your chatbot a personality and unique voice: base your chatbot on an actual human.
For GrowthBot, I’ve decided to base it loosely on my personality. So, GrowthBot tends to respond in ways that *I* would respond. This means it’s light-hearted, always respectful (never snarky to the user), and occasionally a bit quirky and surprising.
You don’t have to base the bot on yourself, you can base it around some human you know pretty well, even better if it’s someone with a great sense of humor or a distinct writing style. The important thing is to have a person in mind when building the bot, it will make copywriting so much easier and up the charm factor of your chatbot.
Your thoughts?
This is just my advice based on what I’ve learned, but I’d love to hear from you.
- Specifically, are there any chatbots you currently use that you love and think are doing really well on technology, psychology, and copywriting-ology?
- Is your company currently using or considering rolling out a chatbots? What’s the use case?
Learned something? Click the 👏 to say “thanks!” and help others find this article.
Keep reading about chatbots…
