Log in

No account? Create an account
Previous Entry Share Next Entry
My attempt to elucidate the problems with call centre technologies
Somewhere, lying on a beach in the Bahamas, is the NLSR sales rep who managed to convince some misguided change guru at AAPT to implement Natural Language Speech Recognition on its phone queues.

Pressing buttons is standard practice. "If you're calling about a fault, press 2." I'm fine with that. Enough questions and I should be able to get through to the right team, with my customer details already on the screen.

But speech recognition? Ugh.

"I can understand what you say."

Stress levels rising. I've just been pressing buttons. Why can't I keep doing that? If the question's simple enough, buttons are just as good. If the question's complex, how will I know what to say and how will you know what I've said?

"Are you calling about a sales enquiry, a billing or payment enquiry, or support for your AAPT service?"

Do I really have to embarrass myself by talking to a machine? What if I don't say it right and you tell me to repeat it, or put me through to the wrong queue? What if your test group had Australian, Kiwi, Indian and Japanese accents, but I have a broad Scottish accent?

"Do you need to report a fault, are you moving, do you need additional lines, or do you have a general enquiry?"

Do I say "fault" in a monotone voice and hope I don't get transferred to Billing like the first time I tried this? Or do I say "report a fault", or "I need to report a fault"?

"Please say your AAPT phone number."

With or without area code? How fast or slow do I need to speak for the speech recognition to work? If it doesn't work, will I have to keep saying it, or is an operator going to listen to a recording of what I've said and type it in, all the while laughing at my stilted speech and broad Scottish accent?

In summary: I've never met anyone who has liked an NLSR system, and I highly doubt it worth the money that AAPT spent on it (over and above the cost of building an equivalent keypad based IVR).

Yes, I've probably written about this before. Either this or my other phone queue gripe: interrupting music with messages. A customised hold track that seamlessly integrates music and company messages is good. But playing music and interrupting it with messages is bad. Here's why:

It's stressful: The music stops, there's a click, a human voice... I've got to be alert and get ready to talk to the operator. Oh, it's just a message. Repeat every 30 seconds.

It's negative: Your words tell me that my call is important and that you're sorry for the delay, but all you're really doing is reminding me that I'm still on hold. Recorded messages, especially when they're clichéd or use weasel words, can easily come across as insincere.

Combined with the regular disappointment of thinking I've got through when I haven't (see above), that's a whole lot of negative. Just let me chill out and listen to the music, and wake me from the trance when you can actually help me.

That is all. I wonder if I can get this into a business or call centre trade mag somewhere...
Tags: , ,

  • 1
Christ. The standard question everyone that thinks about integrating a NLSR system should ask is: why?

All it does is make me feel like a tool. It's more difficult than pressing buttons by a long shot because NLSR isn't perfect.

There ARE some situations where NLSR is useful. Moviephone style deals where you have something like 20 movies? Perfectly acceptable, I don't want to sit through someone saying every single movie that's on. Getting through a menu where you have less than 6 options per point on the menu? Waste of time and money.

Even then NLSR can generally only be used for simple tasks, anything else should be directed at an operator.

Christ, I mean these things cost money. Big money! Don't these people have to justify their bloody purchases?

It's not so much that NLSR is inherently bad - just that it's difficult to get right and has some drawbacks that should be considered before deployment.

The Moviephone example is a good one. That is, provided they handle errors well. That means not asking the user to keep repeating the movie title, but instead to offer a fallback (buttons or a human).

An NLSR implementation I liked was when booking a taxi in Sydney. Combined with CLI, it was a series of quick and painless questions:

- Is your address 30 Ross St Glebe? Yes.
- How many passengers? One.
- Are you ready now? Yes.

- I don't know how well it handled variations like different address or time. Probably well, given the volume of taxi calls.
- The simple responses I gave could have been done on a keypad almost as easily.

The thing I hate about these systems too is because they're voice triggred, I've been on hold saying to someone in the room "oh hey, its one of those shitty IVR systems!" only to have it go "you said "faults", is that correct?" A^%%$#$@#$@# piss.

"You insulted our SR system, thugs have been dispatched. >:("

The one that bugs me is when I think I'm saying something clearly, but all I'm told is "I'm sorry, I didn't understand what you said, please try again".

Note also that tv ads and comedians have played up these fears of poor recognition, adding yet another reason to steer clear of NLSR systems.

What I hate about NLSR is that if you are having no luck with the voice thing because you don't know what to say or it's not recognising your voice, you can't just mash the keypad till you get put through to an operator. If you keep pressing 0 or something it just terminates the call.

And as for on hold music. I remember being on hold once and they had interupted the music with a message. Said message was then interupted with ANOTHER MESSAGE!!!!!

  • 1