" in the knowledge that advances in natural language processing and text-to-speech models make it harder for people to detect whether they're interacting with a bot or not "

More validly on the assumption that folks won'tt detect it . But just try asking the bot a question it hasn't encountered before - although I must admit this frequently flummoxes human call centre operatives as well, but at least one can sometimes argue with them to get it escalated.

On the other hand, 'customer service' is typically viewed as a PR loss leader, so nobody really cares whether the caller gets a resolution to their problem.

They are just using scripts to answer you. Getting to talk to someone knowledgeable it almost impossible. So CCI (Call Centre Intelligence) bots are perfectly fine, they would be mostly indistinguishable - probably you can identify them when they don't get angry when you try to explain your router is perfectly fine and there's no reason to reset it to factory settings...

I'm lucky with my ISP ( [1]'Free' is its name). When I call the support line, I start by saying I work in IT, so the support doesn't start with the usual check list. That's possible because the call centre is populated by competent technicians, not by underpaid staff in some third world country .

== Bring us Dabbsy back! ==

[1] https://en.wikipedia.org/wiki/Free_(ISP)

You're very lucky. On the few occasions I've had to call Free customer service I've ended up talking to someone with an almost incomprehensible N.African accent, who simply reads from a script, just like the India-offshored UK call centres do.

I'm a "drone" in an ISP call center, although I am not front line any more. If a caller demonstrates through their conversation that they're somewhat knowledgeable, I am glad to short circuit the checklist.

Basically these AI systems are designed to have customer give up. If there is a problem, customer must feel that it is not worth their time dealing with it and they should just buy the product again.

That being said, usually if you type something like "this is rubbish" in the AI chat (they keywords may vary), then they'll connect you to a human operator.

Last week I was on holiday abroad when my car displayed a fault message. I used the built-in "call support" button, and was soon talking to someone in the UK. When she heard that I was out of the country she transferred me to the European breakdown service.

That left me at the "press 1 if you want advice, 2 if you're stranded in the fast lane of the autoroute, etc" menu.

Luckily I wasn't in a dangerous location, because the car's built-in calling system had no way to bring up a keypad, nor did saying "one" work...

At least a bot that could hold a basic conversation might have got me to a real person.

Ridiculous predictions could be made by AI and be believed by the AI replacements for PHBs.

there is also value in partial containment, such as automating the identification of a customer's name, policy number and reason for calling. Capturing this information using AI could reduce up to a third of the interaction time that would typically be supported by a human agent,"

Most banks I use already have automated systems (without having to throw around a trendy term like AI - maybe once that's flogged to death they could use the "AI" to store the details in a "blockchain") to collect these details, which are then rendered utterly pointless by the agent asking me the exact same details again, so even if they could collect them by AI, I'd rather they spend the time in getting one system to pipe information to the next....

just NO! If you are in the UK, most people moan when we get pushed to a foreign call centre. Rightly or wrongly this is just the mentality of us. We want an English, Scottish, Welsh or Irish call handler just because we then feel they can relate. Having said that when I had issues with PayPal many years ago I got better service from the Malaysian call handler than the rude Irish guy that answered earlier in the day.

But anyway the point is EVERYONE hates the automated call handlers. So much so I always swear at it telling it to put me through to a human. Knowing the more I rant with garbage, the more it won't understand the quicker it is to put me through to a call handler.

I had my FTTH line snagged by the local farmer's tractor - don't ask me how but apparently it had been installed a bit low on the pole across the street.

The line was intact, but it was on the ground. So I still had my connection, but the line was on the ground.

I had to explain three times to the helpdesk drone before he got the message. The line is intact, yes, but it is physically on the ground . Please come and hang it back up again.

I even got a visit from an actualy supervisor (yes, in person) who came to check on the situation.

Didn't make any difference though, they left it on the ground until some vehicle passed and wrecked it a month later. Oh, then they were there in two days to redo the whole installation.

So how could I explain this to an AI bot when actual humans took an hour to understand and then still didn't do anything until the line was actually physically cut ?

Who knows, it might actually "listen" / work better.

Though I doubt it...

We'll know when AI is good enough for replacing call centres when they also use it for sales too

I assumed all those spam sales calls were already done that way.

I've never managed to get any sense out of whatever was at the other end when I told them my number was TPS registered and I wanted to speak to their data controller.

Vermin Media Indian call centre ISN'T AI bots?

Oh no, that would be able to detect you've phoned up half a dozen times before, had 4 engineers out and rebooting the fucking router for the 5000th time won't fix the problem..

I've had the dubious pleasure of having to interact with an AI-powered customer service. It works (barely) if what you want is within a small set of predetermined and well-defined functions. In this case, it is providing the exact same service as the old "press-3-to-do-this" robots, only it takes longer and is less reliable. For example, you might ask for how much credit you have left on your contract, and it might actually give you the amount. It might also provide you with your contract number, or ask for your credit card details so it can refill the contract, or ask you to update your contact informations, or whatever.

If what you want is not within said functions, then it's utterly useless. In that situation, after you describe your problem, it's pretty much going to pick one of the predefined functions at random.

In my case, eventually, I told it that I just wanted to speak with an operator, so it asked me to describe my problem anyway so it could connect me with the correct department. I did, and it connected me with a department at random, completely unrelated to my problem. The operator of that department forwarded me to the right department, and that was it.

The following time, I opened up with "I want to talk to an operator", and when prompted to describe the problem I followed by dictating the header of their website page dedicated to the service I was dealing with, verbatim. That seemed to do the trick. However, "press 0 to talk to an operator" then "press 5 for sales" would have been much easier and faster.

Human call centres and AI chatbots are both pretty crap. The AI's have too poor comprehension and only understand simple, common questions. Trying to phone a call centre can be a nightmare wading through deep menu systems only to wait on hold for an hour before the line drops. Communicating via emails or online forms with humans is also terrible. I had to report the same fault to Shell Energy multiple times over several months before they acknowledged there was actually a fault... but apparently it is beyond their ability to fix it (the smart meter display which they installed doesn't actually work).