Get best of the week

Speech bot in the bank - the worst UX ever

- Good evening, my name is Natalya, how can I help you?
- Hello, my card was blocked.
- Well, say your name.
- Vasily Morzhakov.
(drives on the keyboard)
- Your code word?
- Decobra through E.
- Sorry, you can not hear.
- De-co-braz, through EE
- Something with a connection, excuse me, can you repeat it? (and it seems they hang up)

And now I know a little about speech AI, I read about bots yesterday Habr, but still I can not understand what is happening.

Of course, I called back and said:

- Hello, I'm abroad, the connection is not very good, please do not hang up the phone. My card is blocked.
ā€œYes, tell me your name ...
I can hear something being driven on the keyboard.ā€ Surname and first name. The code word passed this time, did not load about the letter E (the connection is not very).
- Yes, I see that your card is blocked, I am switching you to another department. (pleasant music plays for 5 seconds)
Everything repeats with Ilya and again they tell me that the connection is not very good!

I thought about climbing into a tree and once again going through all the operators. Stress! You see, the money on the card is blocked, night, another country. Well, at least the operators pick up the phone instantly and ...

Do the operators pick up the phone instantly at night?

Stop! Here I am an idiot.

Is it that already voice chat bots in the bank are responsible for unlocking the card? Something new!

Iā€™m typing again ... for sure. all the same. What? Why does a chat bot drive something on the keyboard? Yeah, if I say something unexpected, they tell me about a bad connection. Fine. It ends with the fact that a certain very believable sounding (with a slight intonation of arrogance) Ilya explains that you need to call from your phone, despite the verification with a code word. But this is simply not possible for technical reasons. And here I was not able to find the right answer. The script probably expected ā€œthanks for the helpā€.

Frustration! During mild stress due to a blocked card. And if I had a wallet stolen and I urgently need to ask her to block it?

As a result: from 7 minutes of conversation with the bank:

  1. 30 seconds glad I quickly picked up the phone
  2. 2 minutes perplexed that with communication that?
  3. 30 seconds, I think that they are already mocking them! There are no communication problems.
  4. 10 seconds of strong emotions that played you
  5. and for the rest of the time you are already furious because the bots continue to type your answers on the keyboard, trying to guess the scripts and speak in the most legible way

Yes, as a result, you still canā€™t achieve the desired result.

Well, what did I want from the bot?


To not be deceived.

I would like a short warning. Additional instructions. The ability to start a conversation again or find out what the script expects from me. For example, like this:

ā€œHello, the automatic voice assistant Bot Elena speaks to you. May I ask you to speak easier and more clearly? I can almost certainly help you. But during the conversation, you can ask to connect with the operator, ask what I expect from you or start the conversation again. ā€

And when I call back during the day, you donā€™t have to repeat all this, just warn that itā€™s ā€œstill Bot Elenaā€.

This entry takes about 10 seconds, which is comparable to the waiting time for a response to any call.

Yes, the client will ask the operator, but after half a minute of waiting for an answer, you can clarify if you can switch to a voice assistant, who will almost certainly be able to solve your problem.

And a little about UX


I like the concept of voice assistants. By the way, I like the console / terminal instead of complex panels in the front-end for configuring complex systems or navigating complexly structured data.

In our school there was the subject of TRIZ - Theory of Inventive Problem Solving . I donā€™t remember anything from there, but I remember one method of solving the contradiction in the statement of the problem: try to get rid of the entity that performs the necessary function in the system, while maintaining the function.

An entity is a front-end, i.e. user interface through which we interact with web services (and a lot of things become web services anyway).

In other words, connect the console almost directly to the service API. Or connect the voice assistant directly to the service API:

Warn about saving words (what the bot can do, reset everything).

Ask how you can help.

In turn, ask the query arguments, if any, if necessary, voicing the type of expected data (by letter, only numbers, date, time)
Answer the request.

If you are working with a context, then warn about the context in which the bot understood the request.

Of course, restrict your rights and think about security.

The final part


Yes, I understand that itā€™s stupid to try to teach UX voice assistants how to do their job. And the bankā€™s business department calculated the profit from the introduction of voice assistants, the pilot tested it, and one frustrated user with a blocked card is just an outlier.

But I hope this story saves a little nerves by the reader of this article in the event of a hit on the voice bot. And he will definitely amuse all the participants of this implementation at the bank (I can imagine how they rolled around, recording all these speech messages).

And the strangest thing is that I canā€™t even find references in the media about the introduction of this type of assistant in this particular bank. And, even after hearing 40 times about ā€œcommunication problemsā€ to any non-standard answers with excellent communication, Iā€™m still only 99% sure that it was a robot, and not 2 living people in the bankā€™s call center who decided to mock.

P.S.


In fact, I was only a witness to this cool story, first-person pretense was used as an artistic device. And all the names of the bots have changed, because I just donā€™t remember them.

More articles:


All Articles