Phil online

8 years ago I was working for a company in the computer telephony business. I was the CT/Voice-guru-designate for the company, so I got to go to conferences and see what the dreamers were dreaming of in the future. The dreams were all over their demos and promos. They didn’t work all that well, and they were expensive as hell.

The holy grail of the industry was a sort of turing test for IVR (Interactive Voice Response) systems. This is where you could call a customer support line and talk to a computer without realizing it was a computer. It would almost be as good as talking to a real human. They’ve been doing this with 411 informatin systems for years, but there has always been a warm body behind the curtain turning dials and punching keys. It goes like this:

Customer Service Agent Larry’s pre-recorded, carefully enunciated greeting: Hello, and thank you for calling Phoneytech Information. This is Larry. What city and state, please?”

Me: Atlanta, Georgia

[1400 milliseconds of silence, indicating to the computer that I’m done.]

Customer Service Agent Larry’s pre-recorded, carefully enunciated second question: What listing?”

Me: Bill Campbell

[1400 milliseconds of silence, indicating to the computer that I’m done.]
[Sound of a phone ringing.]

Customer service agent Larry, now live and in person and sounding tired: One moment. [clickety-clickety-clickety] Thank you. *click*

Computer voice system: The number is: [pause] Seven Seven Oh Nine Two Three …

It’s a clever mix of humans and computers. The computers answer the phones, play the greeting and even record my answer. Then they deliver it to a human who has to be prepared to handle odd requests, re-request miunderstood queries, and so on. When the human has done his part, he hands it to another computer who finishes by reading the information out to me (and freeing up the human for the next call).

Today, however, the human has been fired (”reassigned”) and the computer is taking complete control. And here’s the shocker: it’s better at this job than the human was.

I moved recently, and this week I called Popular Science’s customer support line to change my address. I could have done this online, but I was in my car and I wanted the reassurances that a human could give me. Here’s how my conversation went:

Computer: Thank you for calling Popular Science Magazine customer service. [Bunch of stuff about how to do this all on the web.] To change your address, press 1.

Me: *presses 1*

[Sound of a phone ringing.]

Another computer: (in a perky, friendly voice) Thanks for calling our change of address line. If you need me repeat something, just say “repeat”. [instructions on finding my account number] Please say your account number. If you need more time, just say “hold on”.

Me: surprised Um… 1234567890

Computer: Did you say 1123456790?

Me: No.

Ok… The computer got the number wrong here. But it was really my fault. And I was impressed with how clearly and naturally “she” read the number back to me. Gone are the days of stuttering, differently-recorded numbers for playback, the kind where you can almost hear the cassette tape rewinding between digits.

Check out what the computer does next. It absolves me of any blame. The computer voice here actually sounds apologetic and self-deprecating. I could almost here her say “tsk” in there.

Computer: Oh. My mistake. Please say your account number again.

Me: more clearly 1234567890

Computer: Did you say 1234567890?

Me: Yes

Computer: Got it. Let me get your records. [pause] Ok. If you know the phone number at your new address, tell me that now and I’ll try to use it to find the new address.

Me: knowing this won’t work because I ported my old number 770-923-1234

Computer: Did you say 770-923-1234?

Me: Yes.

Computer: Ok, is this your new address? 123 Ronald Reagan Parkway, Atlanta, Georgia 30012

Me: No [but I wish I could tell you how impressed I was with your pronunciation]

Computer: Ok, tell me the zip code for your new address.

Me: 30047

Computer: Ok. I have the city and state. Now tell me your street address. Don’t spell anything. Include the apartment if there is one. For example, just say “eight forty-seven Oak Lane”, or “twenty-seven Oak Lane, Apartment forty-seven”.

Me: Three two seven nine Riverside Way.

Now, I should explain that my address is full of must-spell words. The city is “Lilburn, GA”, but I always have to give it as “Lilburn — that’s L - I - L - B as in ‘boy’ - U - R- N — Georgia”. Notice that this time I never even had to say the city name.

When I give the street name, the person on the other end always says, “Riverside — is that one word or two?” And now I’m dictating my address to a computer, a beast who is just as likely to send my mail to “River Sideway” as it is to “Reverse Hwy”. There’s no way this can go well.

But I was shocked.

Computer: Ok, I’ve got three two seven nine Riverside Way. Is that correct?

Me: stunned Uh… Yes.

Computer: Ok, I’ve updated your address. Do you want this change to take place immediately?

Me: Yes.

Computer: Ok, you’re all done. Your new issues will ship to the new address. Please allow some time for printing labels to be updated. Thanks for calling Popular Science [yadda yadda, platitudes and congenialities].

When I got home, I had an email waiting for me confirming my changes and thanking me (again) for calling them.

So, the key points for me were these:

1. The computer was personable (no pun intended) and friendly. “Her” voice was clear and enjoyable. She sounded happy. She sounded like she enjoys her job. And she sounds exactly like that on every single call, every day, 24 hours a day. And she can probably handle dozens of calls at once. That’s one model employee!
2. When a mistake was made, she took the blame. Her tone of voice changed to sound conciliatory. It sounded natural. It sounded human. The spookiest thing is how spooky it wasn’t.
3. The computer got my address right the first time. I didn’t have to spell anything. I didn’t have to help her find the right location. I didn’t wonder if she typed it in right. And if I did, she emailed me the address to make sure she got it right. (And she did.)
4. The computer thanked me in a fresh voice that sounded natural and new. It didn’t sound like this was the 127th time today that she had said “Please allow some time for…”, even though it was. It didn’t sound like she was reading a script. When she said “Thanks” it sounded sincere. Ok, that sounds stupid, but it’s true. If a human robotically said “thanks” in all the right places, it might bring to mind their training, like when the phone company always says goodbye the same way, every single time, no matter what the circumstances or resolution of your call: “Ok, Mr. Hord, ThankYouForChoosingBellsouthAndHaveANiceDay. *click*” But when a computer says it, it never loses its patience, it never sounds rehearsed, and it never sounds automatic — even though I know it is, by definition. That’s what saves it. I can forgive the computer for robotically repeating phrases, but when a human does it, even if they manage to do it well, I feel like I’m being lied to.

So we’ve now reached the point where computer based voice response telephone systems are not just cheaper than humans; they actually perform their jobs better than humans can.

Let’s just hope they don’t figure that out.

This entry was posted on Friday, April 29th, 2005 at 1:14 pm and is filed under CastleHouse, Technology. You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site.