Here's some tips we like to follow when planning a bot:
Set a goal
Remember it’s a journey
Have some personality
Provide training wheels
Don’t neglect broadcasts and onboarding
Focus on measuring achievement
1. Set a goal
Think of a bot as a machine to persuade.
It’s like having a conversation with a user: the bot asks questions to help the user define what they’re looking for. Unlike web or social, the user expects the bot to speak to it three to four times per conversation, to communicate with images and video as well as text.
And afterward, 60% of people subscribe to bot updates, so a majority of users expect the bot to keep talking to them.
It’s a historically new opportunity to persuade – to keep a user in conversation long enough to explain why your product fits their need.
Put bluntly, always be asking yourself: what deal do you want your bot to close?
2. Remember it’s a journey
Now you’re thinking of a bot as trying to close a deal, what emotional arc do you need the user to work through?
The user needs to understand there’s a problem. Then they need to be guided to ask the right questions so they understand why your solution is appropriate. And then they need the opportunity to buy.
We’re big believers in structuring the majority of bot conversation through menus, rather than free text input. 60% of users are on mobile and they want to press buttons, not type.
If you’re selling something complex, plot the journey before you begin:
How do you guide a customer to ask questions that play to the strengths of your product?
How do you order the conversation so the user is always comfortable, engaged and learning?
When do you close? Bots give you lots of space to form a relationship with a customer – don’t lose that by asking them to buy on the first message.
3. Have some personality
Bots work best when they are looser and spontaneous. Your marketing team will have strong opinions about the “voice of the brand”, and bots are where that voice actually speaks to you.
A serious brand doesn’t necessarily want banter from its bot. But the bot needs to sound like it’s actually speaking to you as a person, not reciting a menu.
And if you’re trying to convince, use images and video – not long sections of text.
Should you try to convince the user that a bot is human? Think of it this way: a smart robot, like a smart child, gets a lot of leeway. A weirdly unresponsive human does not.
Generally, we find it’s best to be clear the user is talking to a machine. It’s also less confusing if you later escalate a conversation to a person.
4. Provide training wheels
In 2017, the majority of customers have not yet used bots and the conventions aren’t familiar. In the opening sections of your bot, you need to guide users to understand how to interact.
We see two common problems. First, Facebook Messenger bots begin with a button saying “Get Started”. Users sometimes don’t realise they need to press it, so make sure the bot responds to messages like “hello?”, “?” and “Are you there?”. Say “yes”, and offer to display a menu.
Second, and this is true across the entire bot, every conversational section must have an obvious next step. Don’t have dead ends, where you display a message and – without warning – stop and wait for the user to type.
After a couple of dead ends, the user assumes the system is broken and gives up.
If you allow text input, make sure the user understands when they should type – or include a button to press to move to the next section.
Then the user stays comfortable they’re in control.
5. Don’t neglect updates and onboarding
An email update happens once – the user reads the email, and the update is “done”.
A bot update / broadcast, on the other hand, pulls the user right back into the bot.
Always end an update with a question or a button or a carousel. Then the user will start interacting again – even if they’d assumed your conversation was over.
Be ambitious about onboarding, too. Because Messenger is a fun medium to use, a bot’s first-time users are often happy to be sent seven messages over seven days. That’s a unique opportunity to personalise a sales pitch.
6. Focus on measuring acheivement
A bot is a journey, so it’s important to test if users make it to the end, and what conversational path they took there.
Ideally, a bot should allow you to tag related sequences of individual bot questions, text, images and video as a “block”.
Then analytics can test which blocks discourage or bore users. If you’re being hyper organised, you can attach lead scoring to blocks.
We could go on about analytics, but it really comes down to one issue. If you use bots to convince customers and automate upselling, make sure you’re tracking if, where and why the selling stalls.