Chatbots have seen a huge growth in the past two years due to their amazing ability to increase customer service net promoter scores (NPS) and automate repetitive enquiries. However, you probably already know that. So the question is, what does the chatbot look like in 2020 and how can we capitalise on what they can do next?
The evolution of app/chatbot hybrids
By the end of 2016 the chatbot hype really took off and since every person and their bot jumped on the bandwagon, qualified and unqualified – a lot of bad bots were made. Which is why the hype cycle I would argue lasted short of 18 months. (Longer than Bitcoin at least!).
Part of the reason for the flop is because chatbot Natural Language Understanding (NLU) wasn’t to scratch to the expectations of the users. This was both due to the technology limitations at the time (which is making some amazing strides year on year through the likes of Google’s Dialogflow and Tensorflow) but also, chatbots were being used in new ways not explored before, so the training process and architecting took time to test to get to a good standard.
An upsetting but suitable recent example is the recent failure by woebot.io to spot sexual abuse. I’ll be going into more depth about this in a future article to evaluate why it failed and how we can do better moving forward.
For now, as chatbots become more capable, it’s clear we enter ever more difficult and ethical territory.
A second reason is many thought chatbots would be the next app and would replace them entirely with a messaging interface. It turns out apps are still better for most things. Which makes sense in hindsight, an image speaks a thousand words after all
Lastly, chatbots tasked with sophisticated, yet subtle objectives must be powered by equally powerful Natural Language Processing (NLP) models. The teams crafting the chatbots were not always the experts in NLP techniques. Much in the same way user experience app designers didn’t exist 5 years ago, new niché developer jobs and skills are emerging in this field.
All this being said, it’s evident chatbot have their place and they have enormous potential to improve the world. 2019 is an exciting time, where we’ll see the eruption of messaging app hybrids and start mastering the chatbot/app hybrid.
Chatbot platforms will start building and testing themselves.
Chatbots’ complexity is growing more and more each year.
It’s now common to build a project which has about 150 intents and the complexity of conversation itself is 100s of decisions and condition points in the dialogue.
A newfound pain for chatbot designers is that this complexity is getting too labour intensive and expensive to monitor and maintain.
Picture this: you have 1 Intent, which responds with 3 options to select, each one of those also has 3 options to select. Already we have 27 data points to monitor. Most conversations are far longer than this and we have 150 of them. It becomes an exponential problem!
So the key question is: how do you ensure that everything is working in tip-top shape and how can we reduce the intense amount of hours it takes to monitor, build and test these new intents?
I predict the rise of automated bots solutions. This you will enable you to automatically test a bots dialog and NLP.
Some of the big pains of building a chatbot is collect data and utterances, matching them to intents and building a customer journey to solve said utterance.
Platforms such as chatbase and our own enterprise bot manager (EBM) have the aim to reduce the analysis time by letting the chatbot architect design the flow and then the system automatically tests all known flows when intents change to check for undesirable changes in behaviour & flow.
Messaging and Livechat worlds collide
Fight!
2020 is the year I predict a lot of worlds collide:
It’s safe to say digital marketing and customer support is going to become much more omnichannel. Which makes our job more difficult to make sure the customer has a smooth transition through the sales funnel. For now, everyone is playing along nicely enough and keeping their API’s open to allow us to transfer data between platforms.
The question is when one of these monoliths makes a move to capture the other markets.
The winner of this is much harder to predict: my guess (or hope?) is that one of the messaging apps such as messenger wins and turns into the next WeChat of the west. Where we operate everything from one app in a synchronised, harmonious and stupendously smooth user journey.
This obviously has huge repercussions for the likes of Hubspot and Salesforce and is understandable why a lot of the CRM providers are investing heavily in their Livechat and chatbot capabilities.
This also applies to the likes of email campaigns, there are early reports of carrying out similar email campaigns via a messaging app such as messenger, with higher opening and conversion rates.
This demonstrates why it’s so important to keep hold of your own IP and not rely on one provider too much. The safe move is to ensure our systems are flexible and omnichannel, to adapt to wherever our customers decide to move next.
In my old small start-up, we suffered this pain with Facebook messenger when facebook has its data Cambridge Analytica scandal. We essentially closed shop for 2 months. Our whole sales funnel collapsed because we were unable to push or pull data. A lesson learnt the hard way!
Voice usage and adoption will only become mass usage if accuracy and utility continue to improve.
Google’s home hub
This one is obvious but does as some prerequisites: much like chatbots, voice will certainly become one of the main methods of interfacing with technology, but ONLY if it the accuracy moves beyond 95% and the functions moves beyond a cute gimmick.
Google has demonstrated moving beyond just voice and chat by recently releasing their “home hub” which backs up my prediction of our user journeys becoming a hybrid of app, chat and voice.
Automated telephonic interactions
My prediction is that voice accuracy does improve and towards the end of 2020, my crystal ball says all the tech giants will release their open beta of automated telephonic interactions. After the dazzling display of Google Duplex at the X expo, it blew the world away with it’s human-like “umms and errs”
Overall, I am highly sceptical of Duplex. My assumption is that the demo displayed was the best of the best examples they could get.
When someone says things like DeepMind AI reduces Google Data Centre Cooling Bill by 40%, I am amazed. That is crazy impressive!
While duplex is demonstrating progress, I am not ready to go wow over this voice call use case just yet.
HR goes to the next level
At Filament, we’re having a lot of fun testing out some basic bot integrations via Slack, such as automated team scheduling or a timesheets reminder bot. We have a lot on our to-do list to try out such as Workbot, Lunchtrain and OBIE.ai
2020 is the year I think these bots move from fun gimmicks and small value proposition add-ons to more impactful bots for recruiting, on-boarding and hr queries – All of it can be automated and handled by a chatbot to a point as is demonstrated by Jane.ai.
This prediction needs a whole article in itself which I plan in the future, so watch this space!
In 2019 AI will be a *net* job creator
My boss will probably clip my ear for going anywhere near AI ethics, but after Gartner’s spectacular 180 turn around, it’s kind of hard to ignore. If you’d like to learn more about this heated topic I highly recommend Kurzgesagt video on “why automation is different this time”.
Just be ready for a minor existential crisis afterwards.
I’ve also had anecdotal experience of this first hand: some of our enterprise clients which have automated some of their simpler but time-consuming customer enquiries have now either upgraded the customer service agents job to handle more complex and interesting enquiries or shifted them to start building and analysing the chatbot flows.. So the good news is no job lost yet, but none created either.
To wrap up,
I feel incredibly privileged to be living in such a fast-paced digital age with technological wonders appearing for our eyes month on month. I personally can’t wait to see how chatbots evolve to become more useful and prominent in our daily lives beyond just being great at customer service automation!