Habit forming with conversation design: How we got from 0.5% to 60% retention
When we launched our Slack Chatbot, Meekan, in June 2015, we saw all our metrics going sharply up. New user registrations surpassed all of our other products (iOS, Android and Outlook apps) very quickly, and the incoming support requests were looking like fan mail.
Our decision to try a conversational interface seemed very risky at first, but was looking much better with every passing day. Users were coming in through the door to try talking to this new bot thing, and they were loving it. There was only one tiny problem.
Users were trying it once or twice, and then forgetting about it.
You see, our bot, our product, is very good at scheduling new meetings. But how often do you need to start a new meeting? Once a week? The frequency of our main use case was too low. After trying it once, people would forget about it, and go back to doing things the old way (switching tabs to peek in their good old Google Calendars). We needed to form a new habit, and it was time to go back to basics.
If you haven’t already, I really recommend reading Hooked: How to Build Habit-Forming Products by Nir Eyal (or even better, attend one of his fantastic workshops). Nir Eyal’s research says that every habit-forming loop begins with a Trigger: the actuator for a new behavior.
Looking at our own product, we could quickly see that the loop was broken. We were not creating any triggers, but rather waiting for our users to initiate the conversation on their own. No trigger, no action, no habit.
We stared out by the most simple of triggers, the reminder.
Find a good excuse to remind your user that you exist
In our case, it was quite simple. 10 minutes before a meeting, Meekan sends a quick alert:
Reminder (10 minutes from now)
10:00AM Sales team meeting
This was working really nice. The vast majority of our users chose to keep receiving these reminders (with only ~5% of users asking to unsubscribe). And actually, some of the users would go so far as to answering “Thank you, Meekan”. Notice also that in an Instant Messaging environment, initiating a conversation can push you to the top of the threads list, and mark your bot with the red badge of unread messages (This could be too much, but more on that later).
But a trigger alone is not enough. It has to be followed by an action if we want to crank up the entire habit loop. So we quickly followed up with suggested actions, to complement our reminders:
So now you could not only Snooze (and get reminded again), but also quickly tell everyone you’re late, or just cancel the meeting, or even have Meekan figure out a new time for everyone. Together, all of these were setting the full habit loop into motion: A trigger eliciting action, followed by a reward and sometimes an investment that loads the next trigger (eg Reschedule).
And you know what else? By offering these action items, we were also giving a nice incentive to use our way of scheduling over the old way. Meekan — like most of the other bots out there right now really — is not only trying to form a new habit, but also replace an existing habit and a mighty strong one at that (Google Calendar, no less!).
The chatbot advantage
Context switching is really, really hard. What does that mean? Let’s say you’re chatting with your friend over Slack, and you talk about finding a time for a breakfast next week (I’m obviously giving examples from the scheduling world, but it’s the same for calling an Uber or ordering pizza). You switch apps into your calendar to check how your next week is looking, and eventually come back to the conversation. This whole process might take half a minute, but in terms of cognitive load, you’ve lost much more. The return trip from one app “world” into another and back is a huge distraction. As you leave the conversation for a second, anything could happen: you’re looking at your calendar, oh wow. Your insurance is expiring next week! You really should do something about it. Who was that agent that got you the discount last year? Where is her number? What were we talking about right now? (Okay, you see what I mean)
But if you’re using a chatbot to accomplish your task, he’s right there with you in the conversation, and your flow is not interrupted. This could appear as a rather small win, but when you’re trying to replace an old habit with a new one, this uninterrupted access results in a shorter path to success and less friction for the new habit you’re trying to install.
Other trigger types
So far we looked at the reminder trigger, the simplest one to figure out and start with. But how about some other types?
This trigger is more effective than your plain reminder. It alerts you to a situation that requires some action to resolve. A good example is an RSVP request:
You’re describing the situation (someone is inviting you to a meeting), allow access to information that may assist your decision (“Your calendar shows you’re busy!”) and the ability to take decision right there and then. No context switching required.
Now this is the holy grail of triggers. It takes the “Action required” pattern one step forward. It’s a chain of events that can be described like:
- The bot is warning you that the sh*t is going to hit the fan very soon. Maybe you forgot the stove on, and your house is going to be in flames in 20 minutes.
- You’re glad you got the heads-up on this, but now all stressed out — how are you going to fix this? Your won’t make it back to the house on time. Your mind is racing to come up with possible ways out of this.
- The bot immediately offers a good solution (It turns out you can remotely shut off the valve). Shall we do that?
- Easy out. Life saved. Yes! Do it, bot!
- I better keep this bot around.
If you can come up with disaster+solution pairs like this, you’ll be saving the day every day. Here’s an example from Meekan:
So sure, your house is not on fire, but this is as close as it gets. Once again, you double booked yourself on your home and your work calendars, and you’re going to miss your kid’s soccer game. If not for this warning, you’d have noticed this only when it’s too late. But instead, Meekan is offering to reschedule the sales meeting for you right now.
Stress is on, stress is off.
It doesn’t happen too often, and not always easy to resolve with one rescheduling attempt. But when it does, it’s pure magic, and your bot gets a new friend forever.
More ways new bots can replace old habits
So, we saw that by being inside the conversation, bot are easier to use than other, outside-the-context solutions. We also looked at actionable triggers as a way to form habits. Nir Eyal’s model offers some more ways for your new habit to overcome the old way of doing things:
- Velocity and frequency. Be faster, and more often. If you’re giving the same trigger as your competition (eg, a reminder), make sure that your reminder arrives first to the user’s attention. This way, the user is more likely to act on your trigger, making the other one obsolete on arrival.
If your use-case is not frequent enough, come up with sub-modules that appear more often.
- Higher reward. If two notifications arrive at the same time, the user will likely choose to act on the more rewarding one (like mentioned above, if a problem comes with the solution attached, it’s the easy choice).
Studies show that you have to be 9 times more rewarding than an old habit if you’re to replace it, so think well about this one.
- Less friction. If it’s a no-install, no storage space needed, no-instructions, (in other words, a bot) VS a download, install, learn process — who do you think will win?
There is no spam like chatbot spam
People hate bot spam, and they are 100X more sensitive to bot spam. They will block your bot in a second and tell all their friends about it too.
When you initiate a trigger, you absolutely must make sure you’re sending valuable information (valuable to the user, not to your business goals!), and in the right timing and frequency. Otherwise it’s spam, and it will get blocked, habit or not.
Echo, we have a problem
I want to leave you with a final thought. If your device has no way to initiate a conversation (Amazon Echo, Google Home?), how are you going to get people into the habit of using it? It’s difficult. I’ll try to discuss that in a future post — if you have good ideas, by all means share them in the comments below.
Oh, one last thing, sorry.
So, is this stuff working?
Since implementing our different actionable triggers, we’re seeing a very steady 60% 3-month retention. Around 50% of our users engage with Meekan 7 times or more in a month.
These are still not Google-Calendar-level numbers. We still have work to do. We need to be so good at this, that our users will no longer want to ever go back to the old way of doing things.
Written with Eyal Yavor, Meekan’s CTO and co-founder