A chatbot is supposed to do what Alexa and Siri do – why don’t they?

A chatbot is only as good as the underlying digitization of the business it serves – that’s why a lot of them suck.

Some years ago I was looking at startups at Hong Kong Fintech Week and one of the biggest categories was startups that created chatbots, AI bots, or other intelligent bots (regardless of what they were called) for financial services. Lately I’ve been wondering: how many of those startups still exist? More importantly, how many bots of the ones you have interacted with did you actually like? Which of them offered good customer service?

Bots are one of many examples of the digitization of legacy organizations and processes. Companies think their services and efficiency become better with bots. However, the reality is that they become worse if operations and processes are not designed for the digital era and customer needs.

Once, I tried to chat with an airline chatbot just to to clarify some details for my flights due to some new COVID restrictions. But no matter how I tried to describe my needs with different sentences and words, the bot always answered with some fixed options: do I want to change my booking, do I want to make a new booking, do I want to cancel my flight, etc. It was more like a drop-down menu, not really a chat.

Another time, I had to contact my bank to ask for just one document. They also had a chatbot to filter customers and direct them to the right service or person. It also offered me fixed options: whether I want to discuss loans, investments, internet banking, cards, or something else. And when you choose one, you got more sub-options. In my case, I always had to choose the option ‘something else’. Finally, I got to chat with a real person. I explained to the person my needs, and I was asked to wait. After 15 minutes of waiting, I had to leave the chat. Much later, I was informed that I could get the document I needed and was asked if I wanted to have it sent to my home address. So, I had to go back into the chat, restart with the same chatbot and go through all of those options again until I finally got to chat with a real person, explain everything once again and finally confirm that, yes, they could send it to my home address.

Google Search with a drop-down menu

Imagine that Google’s search engine worked like a typical chatbot. First, you would have to choose whether you want to search about science, history, rumors, news, or geography. If you choose history, you would get more options: modern history, ancient history, medieval history, middle ages, or something else (note: I intentionally have middle ages and medieval history here to represent the times when a chatbot gives you overlapping and thus confusing options). Suppose you choose modern history. What if, at the moment when you can finally ask your questions, you get an answer that “sorry, but we must transfer you to another department to chat because your question is actually about the middle ages”?

This is how many enterprise chatbots work. It is like automating information search from an old encyclopedia.

These experiences aren’t what you expect from a chatbot. It is more like navigating drop-down menus or a call-center IVR system where you dial an option and wait a long time. It seems like these companies haven’t been able to include any smart chat functions in these chatbots. Some chatbots that try to understand natural language either don’t really work or don’t have enough data to act on natural language messages.

On the other hand, there are chatbots out there such as the one for Financial Times customer services, where they have enough real people to serve you, reply in a few seconds, and typically fix all issues in a couple of minutes. This is what you expect from a chat.

We could perhaps blame chatbot software and the companies that created them, but I don’t think it helps. For one thing, the problem is much more fundamental. In many cases, it’s impossible to make a good chatbot for such companies because they already have issues delivering good customer service in any channel whether they automate it or not.

Also, there are also very good bots out there, like Amazon Alexa, Apple Siri, and Google Assistant. True, we don’t usually call them bots, but they serve similar needs – humans can interact with a machine, get information, and control actions. Some people love these home devices, although others are not yet convinced about their utility. Still, we can say that typically, they work much better than the chatbot of a bank, an airline or a healthcare organization.

The problem is not the chatbot

All of this brings us back to the fundamental problem – many organizations are not truly digital. I have written earlier about how enterprise software is often sold with unrealistic promises, and automation and digitization not done properly is like putting lipstick on a pig. We encounter the same problem with customer service chatbots.

If a company’s services, customer operations, and processes are not designed for digital services, the customer experience will be bad if you just build a bot that implement those old legacy operations. Just think of my examples earlier. If I can ask all kinds of random things from Alexa or Siri and get quite good answers, why I can’t I just say to the bank’s chatbot that I need a certificate of my bank relationship, and it would immediately make the order and check my mailing details from me? Similarly, why can’t I ask the airline’s chatbot what are the required Covid documents if I fly from London thru Paris to Singapore?

The answer really amounts to a list of digitization issues at enterprises:

1. The information is in silos and different systems, and it is not possible to use all information effectively
2. The organization and systems are designed based on products and internal structures, not on customer experience and needs, and that’s why you must (for example) choose which bank’s department you want to talk to
3. The implementation is just an additional layer on the legacy IT systems
4. When the bots were built, no one thought about what the customer experience should really be with chatbots, or how it might differ from the experience of a call center or talking in person.

It was an illusion that an old-fashioned bank could compete with digital neo-bank services just by introducing chatbots and other customer service bots. Clearly there is more to it than that.

Probably the worst chatbot hype is already over, and companies will hopefully realize that a chatbot alone doesn’t solve issues and improve customer service. It is fundamental to start by thinking about how to implement things in ways that enable customers to get what they need. This is not only about chatbots – they are merely the tip of the iceberg. But they illustrate very well what goes wrong with automation and bots if a company is not able to transfer its organization, processes and systems to the digital era, and instead tries to bolt on an extra layer of technology on old operations. The results speak for themselves.