Rule-based chatbots have their limitations, but you can overcome them.
My insurance company has always been truly excellent whenever I’ve called them. Extremely helpful customer service agents, regardless whether you called them to file a claim or ask about a bill.
Yesterday, the insurance company sent me an e-mail. They wrote that they’d updated some terms and conditions, and that I should go to their website and read some new documents and accept the new T&Cs.
So, I went to their website, logged in, and got a page with list of documents, two of which were indeed new and unread.
However, when I wanted to open the documents, they never opened. I then tried logging in with another browser, and I could then open the documents. That happens from time to time. Little glitch; no big deal.
Now, this is when the real story begins: Do you know the feeling when you want to kick yourself for having tried to be nice to people who didn’t appreciate it?
Being the kind person that I am, I thought that it might be useful for the insurance company to know that I’d been unable to open their documents in a particular browser. I wanted to quickly tell them, so that their IT guys could look into problem and get it fixed, in case I wasn’t the only affected user.
I looked for the ‘Contact us’ menu on their website, found it, and clicked it.
“What do you want to know? Welcome to our knowledge base!” the page said, and it had a big search field that told me to type in my question.
Well, I didn’t really have a question. It was more like I had a finding that I wanted to report. After 5-6 fruitless attempts at wording my finding as a question, I noticed a small link at the bottom of the page. It said “Talk with us!”
Just what I needed: One of the insurance company’s excellent agents to my rescue. So, I clicked the link.
Never should have … they’d implemented a chatbot:
- Nameless chatbot: “Hi. How can I help you?”
- Me: “After I log in to ‘My page’, documents will not open in Safari browser.”
- Nameless chatbot: “Are you looking for a particular document?”
- Me: “Yes, and it won’t open in Safari. Just thought you guys ought to know, so you can check up on it, and fix it if other users than me turn out to be affected.”
- Nameless chatbot: “Are you inquiring about a free yearly check-up of your insurance portfolio? Or. Do you have premium car insurance, and are you inquiring about a free check of your car’s windshield wipers and washer fluid levels?”
- Me: “No, but I think you should know that your documents don’t open in Safari.”
- Nameless chatbot: “Are you looking for a particular document?”
- Me: “Yes, one that’ll open when you need to read it. But it doesn’t. Your IT guys should look into the problem.”
- Nameless chatbot: “If you have a problem, you can at any time get advice, if you click the ‘Talk with us!’ link at the bottom of our web pages.”
- Me: “In case anyone goes through the log of this conversation, you can at any time contact me for feedback about the user experience that your chatbot solution provides. You appear to be using a rule-based engine that’s no better than the ones they used in the very first chatbot experiments at MIT in the mid-1960s. In 2019, you can achieve more with chatbots. Don’t be afraid to seek advice. You can get help, it’s not too late. Lots of hugs from … Well, I guess you know who I am, because I’m still logged in.”
- Nameless chatbot: “Are you having trouble logging in? If you have a problem, you can at any time get advice, if you click the ‘Talk with us!’ link at the bottom of our web pages.”
Really …
I believe that the insurance company could have done better – as in waaaaaaay better – with their chatbot service. Fortunately, many other organizations do much better.
In my case, I didn’t want advice about a problem I had. Instead, I wanted to tell the insurance company about a problem they had, but no one had expected the bot to have to handle such scenarios.
The bot kept going in circles. There was no way out for the poor thing, because it was only able to react with a limited set of replies based on words that it recognized in my input. That’s exactly how those early chatbot experiments from Massachusetts Institute of Technology in the 1960s worked. More than 50 years ago.
If you’re thinking about taking your customer service chatbots forward from the 1960s, but you don’t want to go totally overboard with AI, there’s an interesting article about overcoming the limitations of rule-based chatbots.
Morten Müller is Documentation & Localization Manager at Zylinc’s HQ in Denmark.