12 Dec Getting Ready For The Bots Of Tomorrow
Someone in my twitterverse recently commented how they hadn’t come across a bot in their normal daily life, and therefore we must still be in a hype cycle. This is in many ways true until you realise there are thousands upon thousands of new bots being deployed every month at the moment, so it’s just a matter of days or weeks before you come across one in the wild.
There is still a hype cycle around this, particularly where messaging platforms are concerned, but the world is marching forward at a significant pace towards this “4th Industrial Revolution”, Earlier this year Gartner had any form of bot or conversational UI as being 5-10 years away in their hype cycle, but are now suggesting “AI bots will power 85% of all customer service interactions by the year 2020”
The success for an individual bot still lies in content & context, and there is no silver bullet for this. Bots need rich and deep content for consumer satisfaction. Training your bot needs to be handled like maintaining a website – keep it relevant, keep it focussed, make it deep and rich, not broad and shallow. Personalisation is also key as consumers expect bots to be aware of who they are and why they’re engaging with the brand.
As with any online engagement strategy, whether it’s commerce transacting or customer service, real-time analytics and re-engagement methods are key to maximising customer interactions.
The current wave of bots are generally falling short of customer expectation, with poor experiences being driven by shallow content and a lack of context awareness. The rush to develop a bot has led to a misunderstanding of the limitations many of the bot platforms have in handling the complexity of conversation vs. the simplicity of answering a question.
The next wave of bot deployments will take content, context and engagement to a whole new level through true conversational ability – bot 2.0 is where to focus your energies to be ahead of the curve.
Content: – Focused & Personalised
Few if any bots are really deep and rich in content about a particular domain subject. This is partly due to limitations of understanding how to use the pools and lakes of data that sit idle on the side lines, as bots tend to get built from scratch. Another key failing in current bot deployments is understanding general conversational flow and, as importantly, how to steer conversation. Minimising deviation will maximum customer satisfaction before they drift into conversations that your bot isn’t capable of handling.
Many of the more widely used bot platforms are limited by their ability to maintain reasonably lengthy conversations. Amazon’s Alexa is a good example of this. The core technology [True Knowledge] was only ever intended for question/answer use, not conversational ability. Amazon is currently looking to fix this with their university prize fund, and their ambitions for Alexa to hold a twenty minute conversation.
I imagine Amazon has realised that by Alexa becoming conversational they can learn a lot more about about customers, and those conversations can be analysed to generate further commercial advantage. With the machine learning and deep learning teams they are growing across Europe & the US they’ll generate much more valuable insight with conversational data than they could with other data sets currently available.
Such ambitions will require in-context memory, multi-session memory, topic switching, and a host of other fundamental conversation handling techniques that some of the more mature platforms use. My previous post about the conversational tipping point provides further explanation.
I mention this, as it is vitally important to keep your bot focussed with deep, rich content to avoid pitfalls. Get truly conversational with your bot strategy though and you can unlock a treasure trove of information about individual users that would otherwise be impossible or costly to discover.
Context: – Relevant To Content & User
Whilst content should be deep and rich, context should be broad. By this I mean you need to understand who the customer is, what they are doing, what did they tap on just before initiating the bot conversation. What is the environmental context, like date, time, location, previous actions, etc, etc, etc. The context on which to base an interaction is key for successful engagement, it’s simply CRM for bots. This is where machine learning and deep learning can assist greatly but you can go a long way with some creative engineering and access to user data.
The fusion of data sets provides rich context by pulling in personalised data, live analytics, etc. to build as full a picture of context about the user and current situation as possible. Focus your attention on grabbing data that is holistically vital to the success of your engagement, and leave the rest for a later development sprint.
If you’re, say, handling customer service for the return of goods sold then focus on meaningful conversation around that subject, make sure your bot has access to your returns policy, user transactional history, etc., so you can provide context relevant to a conversation.
Engagement: – Timely & Convenient
When choosing your engagement strategy apply “lean thinking”. Super successful bots are time consuming to build and maintain at present. By narrowing your focus in terms of the function your bot is going to perform, and being rigorous by building iteratively you will be able to build a bot with higher NPS. It’s better to build out incrementally based on customer demand learnt through your chat analytics. A great way to think about this is ‘getting to product/market fit’ as quickly as possible.
Choose engagement channels based on a balance of demand and control. You need to be able to roll out through a large enough channel to prove useful to customers and gain feedback for improvement, but not so large that technical debt can’t be managed. There would be nothing worse than rolling out to your sizeable Facebook following only to find a small oversight causes significant embarrassment – like Microsoft discovered with Tay.
Timely and convenient are good hooks for engagement, ie. when is it most timely in the customer engagement cycle and where is it most convenient to provide access to your bot.
When you come to thinking about what your bot is going to do for your customers (eg. your bot roadmap), I highly recommend the Tom Conrad method of prioritisation – “What would be stupid for us not to do in the next 90 days?”, I use it all the time.
Whatever function you choose for your bot, you might find the following post of interest too:
If you’d like any help or advice, please feel free to get in touch.