The Future AI-Based CX Has Nothing To Do With Chatbots

As a developer and co-founder at DataLingvo I’m frequently dumbfounded about a barrage of chatbot related paid articles, blogs, white papers, and “studies”. As someone who’s spent years in this space, built several chatbots along the way — it is very clear to me that the future of customer experience (CX) is certainly has nothing to do with chatbots as we know them today…

One of the first public sobering acknowledgement of this trend is the awkwardly publicized fact in the early 2017 that Facebook is scaling down the use of chatbot due to over 70% of failure (and hence f-AI-lure meme was born).

Let me repeat it again: Facebook with its literally unlimited $B budget couldn’t make chatbots work beyond trivial use cases.

Facebook’s reason was that over 70% of conversations require human curation to be properly understood. Human curation in of itself is not a problem (Google and Amazon employ hundreds of people in curation) but the self-learning capabilities of such systems is critical. I don’t know specifics of Facebook architecture but I’m pretty sure they have smart people working on it and it only underscores the fact that even domain-specific true autonomous NLU/P systems are extremely hard to build and frankly don’t yet exist beyond trivial support/notification/confirmation use cases.

But there’s more to the story here… There are more cracks in chatbots bubble in general:

Chatbots Is a Hobbled UX Platform

I can’t repeat this often enough. Chats as we know them are made to exchange mostly textual information with some use of other media: you type text and get text response. Yes, some platforms have rudimentary support for basic GUI components and various media types — but they are afterthoughts at best (e.g. Slack’s buttons).

In reality there is a very narrow application of text input and even more narrower application of text output. Think about a situation where today you could say something definitive and complete using your voice or text quicker and easier than using some special purpose UI on website or mobile app. If you think honestly and logically about it — you’ll be surprised how rare the voice or text input would have a clear win today.

When it comes to output — the situation is even more precarious. What was the last time when a simple text (with an image or a button or two thrown in) is the clearly better alternative to some purpose built UI? Apart from simple confirmation-like responses — rarely indeed…

In 20–30 years when NLU/P is much more advanced the applicability of voice input will widen significantly. But it will have absolutely nothing to do with chatbots — any and all systems and devices, consumer and enterprise, will natively support natural language interface.

TenCent WeChat Misread

One of the original justification for chatbots was the quick rise of WeChat from TenCent in China (launched in 2011 and has ~1B DAUs today). And to this day it still remains the main (and roughly the only one) example of successful messenger-based multi-discipline platform. Yet we are conveniently forgetting time after time the following facts:

  • WeChat has almost a monopoly in China for messaging
  • China is an Asian country with very specific culture and customs for technology adoption
  • WeChat was/is heavily promoted and controlled by the government of China

Last two points should really caution everyone looking to replicate WeChat success in EMEA or US. The first point, however, deserves its own special mention.

Fragmented Ecosystem

One of the true Achille’s heels of chatbots (and a major difference with China’s WeChat) is that US and EMEA markets are extremely fragmented when it comes to messaging apps. Slack, Skype, WhatsApp, Telegram, Google Hangouts, Zoom, etc. and the list goes on. I’m personally using 6 different messengers just because some of my contacts prefer different apps:

  • We use Slack for messaging at work
  • Skype for quick voice/text communications
  • Zoom for voice conference calls
  • Google Hangouts for auto-scheduled calls through Google Calendar
  • WhatsApp for some of my friends
  • Telegram for team mates in Russia + some interesting channels

At my apartment I also have Amazon Alexa and Apple HomePod (Siri). I’m absolutely certain that there’s not a single bot that works across all these platforms. And there’s probably a dozen or more other major messaging platforms outside of what I personally use.

That fragmentation is killing almost any benefit of having a chatbot UX. Even if I have a bot, for example, on Slack it means nothing for all other messengers I use. And if I have to go back to Slack app every time I want to use that bot — it defeats the original purpose of the chatbot to begin with. Why not go to the website or mobile app instead?!?

In The End…

Natural language interface is here to stay. It has an absolute intrinsic value — that’s how we, humans, interact and that is how computers should learn to interact with us as well. In years to come chatbots will look like no more than a temporary trend, while more and more applications and systems — consumer and enterprise — will natively adopt the free-form natural language interface as one of their main UX mechanisms.