What are bots good at? They’re a heavily hyped technology, and a lot of the hype is too good to be true. But chatbots (essentially computer programs designed to have conversations with humans) are very useful in particular situations, and if built in a specific way. The key thing is to keep conversation, and development of an ongoing relationship, at the heart of a bot’s design.
Here’s four tips for building better chatbots, which I've refined over my years working for The Bot Platform – and some tips about when not to use them, too.
1. Think of bots as a machine to persuade …
Unlike email or social marketing, bots are a two-way conversation: the user talks, the bot responds; and each continues to respond to the other until the user has had enough. That kind of back-and-forth feels more like a sales conversation than a blast of marketing, and so sales techniques are useful to build into bots: particularly, steering the user to ask the “right” questions.
When building new bots, think about:
- Why are existing users loyal to the product?
- What are the sales objections to buying the product?
- How do you structure the conversation with the user to cover off those objections?
- What reassurance does the user need to hear to be comfortable they’re buying the right product?
Given all this, you might think bots are overkill to sell a single novel. But what they can create is loyalty, not just sales. By letting users ask questions at their own pace, you give users the space to reflect, and convince themselves they should be loyal to a novel, its series, or its author. You’re looking to deepen the emotional connection to the work – and long-term that’s more commercially interesting than flogging a few extra copies. As an example, if you like Hendrick's Gin’s forays into Victoriana, you may enjoy this bot.
The Hendrick's Gin Cucumber Adoption bot
2. Plan updates to re-engage the user
Bots have three stages:
a) Initial contact. Think of this as a “training wheels” stage. The user may not have used a bot before, and prefers simple menus to navigate through a bot.
b) The bot “core. Once the user is comfortable with the bot, they should then be able to play with anything in the bot - all accessible questions and answers, content and navigation. Some parts of the core can (and probably should) be only available to users who’ve earned access – for instance, by sharing the bot with friends.
c) Updates. Once the user has finished playing with the bot that first time, they’ve hopefully subscribed to receive updates later. Here’s where things get interesting…
Updates provide more functions to bots than to they can within email or social updates. That’s because each time you send an update, the user returns to the bot to read it. If you use the update to ask the user more questions, the user is immediately interacting with the bot again. To see this in action, check out the BBC’s Election Bot later this week, when it will start sending more complex updates that offer multiple re-engagements for the user.
When building a bot, consider how to structure onboarding (“Would you like to be sent five messages in the next five days to tell you the secrets of Cersi and Jamie?”), “unlockable” sections, and updates to ensure that conversation continues past that first contact with the bot. Bots typically see a 60% subscription signup, and a 99% read rate. Take advantage.
3. Don't try to make bots sound like people.
Artificial intelligence is overhyped to an almost hilarious degree, but it performs some tasks well: looking for patterns in data, for instance. Recognising faces or understanding speech.
But one thing an AI can’t do is inspire emotion. If your bot is a journey through which you want users to change their understanding, then use human copywriters to structure and write the emotional arc you want users to work through. And don’t try to fool your audience into thinking a bot is a human – an audience of readers will spot that kind of unreliable narrator quickly. Robbie Williams’ bot is a great example of a bot that’s not trying to be a person, but has some fun with “in the style of”.
4. Get the analytics right.
It’s unlikely that your bot will be perfectly designed the first time, but it’s also hard to use traditional web metrics to evaluate and optimise its performance. They don’t show how onboarding, a core bot, and updates can create a journey, or measure the success of the goal - to convince the user of the value of an artistic work. They’re also not great at testing whether a bot has successfully automated much of the fan interaction that previously was a chore.
So when adopting bot technology, make sure it tracks:
- How many users complete its journey?
- What are favourite path(s) through?
- Where do the majority of users stop – and what could fix that?
- What action do users take at the end of the bot?
- Do particular kinds of updates cause users to unsubscribe?
Are updates encouraging user re-engagement, or are they mostly ignored?
So bots have their uses, and hopefully you can see there’s something in the hype. For more inspiring ideas in the publishing space, you might like to check out what's happening at the New York Times and Washington Post (not many people are doing interesting bots around books yet). But that doesn’t mean that bots are about to replace publisher marketing teams – only that they’ll prove increasingly useful working alongside.