Building Your First Bot: 5 Considerations

Facebook has just jumped into the bot space in a big way, joining Amazon, and a host of others who see bots as the new customer engagement strategy for the coming decade. I'm hoping Apple will follow suit shortly. 

The recent surge in bot numbers is largely due to Apple's launch of Siri, bringing the technology into the mainstream, and now we have all manner of companies using bots for commerce transactions like ordering Dominos  Pizza to more complex customer service agents.  There are also plenty of good uses cases for these 'conversational UIs' in Entertainment, Education, Healthcare and as a productivity tool. It's the Ubiquitous UI.

Language is the very foundation of human communication that every person knows, whatever their understanding of technology or access to education. That makes these NLP systems incredibly powerful for every enterprise and consumer. 

Before you make your bold start into the world of bots, here's five things worth keeping front and centre: 

1. Not All Systems Are Equal

You'll read a lot about how this technology has been around for years, but don't assume this means they are mature, all equal, easy to implement, or even using the same principles. Bot conversational engines are diverse in feature sets, processing models, and capabilities.

For example the two main processing models are rules based models, or statistical based models, and often a blend of the two. All have valid uses, but can have potentially disastrous consequences in a minority of instances, for example the Microsoft Tay incident that in part used statistical modelling.

One of the best ways I've seen of deciding which platform to use is to map out the user experience in detail for a particular use case without any knowledge of bot technology, and then ask the vendor to explain how each successful interaction is reached. This will highlight where shoe-horning is taking place. 

2. This is a content play - invest in content players

Whichever role you're employing your new bot to undertake, it's content is key. Bots provide probably the most engaging experience your customers will every experience from your company, save for real human interaction. It's therefore key to get the utmost from these interactions through carefully crafted content. 

The mostly widely used Bots (Siri, Cortana, Amazon Alexa, Google Now etc. ) are all driven by content. It's why they are so popular, Siri is processing 1 billion requests a week and is used by a third of millennial users on a daily basis . Microsoft clearly sees the importance of content as Cortana’s writing team includes a poet, a novelist, a playwright, and a former tv writer, rather than leaving it to a team of computational linguists. A few years ago I created a chatbot based game and used a freelance journalist to craft the majority of content to ensure language is user focussed rather than system focussed. 

Poor content is where I believe a number of bot implementations fail to impress customers. Whether it's for simple commerce or customer service, treating this like a content play will pay dividends through higher engagement rates. 

3. Keep it narrow, deep and rich

When Siri launched in 2011 the data set was reasonably broad for a first iteration. This led to a few cracks being exposed, eg. you could ask Siri to call you an Ambulance, and from then on she'd think you' wanted to be addressed as 'an ambulance'. Language is more complex than it appears at first glance. disambiguation, switching topics, maintaining context are computationally highly complex things to master. If you keep your bot focussed in a key area or on a key task you can train it with deep, rich knowledge to impress customers. From here you can then start building out around the edges, after pulling insight about user behaviour and their desires from your analytics.  The nature of these bots also means you can take a very modular approach, eg. one bot as a sales assistant and another as a customer service assistant, and then give them sight of one another. (see my example of a Virgin Atlantic agent hooked into Siri)

4. Build A Plan Around Analytics

Bot conversational analytics provide unparalleled user insight. Like analysing website searches you begin to understand what you didn't know, plus you get sentiment analysis and a host of other benefits too, particularly where user personalisation has been implemented in your bot.

Building a maintenance plan around your analytics will see your bot increase in popularity and engagement driven by your customers.  

5. Make it omnipresent

Bots are technically accessed via simple messaging whatever input method you are using (voice, keyboard, sensor, etc). They also run server-side in the cloud, predominantly. This means you can build one bot and deploy it N+ times across all your communication channels with almost zero incremental cost. So your bot can now communicate with your customers via twitter, facebook, SMS, Slack, website, or any other comms touch point they prefer. The smarter implementations will also be able to manage conversations and transactions across all of these channels simultaneously for any customer making it appear far more knowledgable and intelligent. 

If you'd like some help or advice on the subject of bots please feel free to get in touch.