Let's be honest: the companies that jumped off the AI Agent cliff were ones who already put customer service on the back burner. To them, customers and servicing them was a money sink, so throwing a little more after a hastily-trained ChatGPT model was made without a second thought. Companies which pride themselves on customer care either skipped AI or used AI to create tools to be used by customer service, not replace them.

That is the model which has a better chance of working, especially if the humans are empowered to supplement their skills with AI rather than become subservient to them.

My online customer support experience usually goes something like this:

Site: blah blah I'm Cretin, your customer support bot, how can I hinder you blah blah

Me: Connect me to a human

Site: blah blah is your problem with this particular thing blah blah

Me: Connect me to a human now

Site: blah blah or are you having a problem with this thing blah blah

Me: Connect me to a FUCKING HUMAN RIGHT NOW

And it usually goes downhill from this admittedly not very promising start.

Some of us would prefer not to speak to a human: just give us a comprehensive knowledge base and a good search.

That's fine for all the things that you've come across before, as long as you have a good and concise search tool to try to find what you want.

But keep people for the exception processing, where it's a new issue that falls outside of the knowledge base, or there is a customer who can't cope with the process of driving the search.

I think that the underlying thought in this article is that both AI and real people are necessary for good customer support.

That's great when you want instructions or information. When you want to actually DO something that isn't in the web portal or the app you need a human. And that is lots of things, most commonly cancelling.

i.e. cancel my subscription

Literally the only time I'm calling customer services is because I can't find the answer on their "customer forums", AKA - free work by the customers the companies should be doing. I NEED a human by that point.

Had a problem with HMRC and a refund. 10 days became a month, which became 2 months. My accountant couldn't see an issue, I couldn't see an issue so had to call. Wait on the line and finally get through to a human. Turned out there was. block on my account, which wasn't on the website and the person on the phone couldn't fix, but knew the department who could. Without actually speaking to someone, I'd still be waiting for that a year later.

You have 2 types of people who call these lines. 1) The people who REALLY can't use forums and NEED to speak to someone for even the most basic issues (ie your grandfather and his mobile phone account) or 2) people like commenters on the Register, who've already looked through the forums and done the searching online etc & their particular issue is an edge case or hasn't been seen before.

If human agents are chained to a script (a big hello to Virgin Media, as one example), have limited autonomy and can't make decisions that usefully help customers, or simply don't speak English well enough to interact with customers, then it is eminently feasible to use AI chatbots to replace them. The companies who are regularly at the arse of the various UK customer satisfaction tables already have low customer satisfaction, poor first call resolution, bad reputations, so why would they NOT use AI to replace meatsacks?

Companies like Barclays, Virgin Media and O2, Scottish Power, British Gas, Carphone Warehouse, Santander, HMRC, AON, just about any pension provider, many local councils. These organisations just don't care, and AI represents another valuable opportunity to reduce their costs and make their service still worse.

Support. From a competent human, capable of pushing it up a level if necessary. Ideally, one who speaks the same language natively, to simplify understanding. And most of all, one who is applying all their attention and intellect to _your_ issue.

What they don't want: Robots.

I'm all for robots, provided they do help me and solve the issues.

Since the LLMs are trained on the way current unsupport hinders customers I don't have too many hopes.

Not essential to be a native speaker, just a competent one. I needed email ports enabled on a VPS recently and the hosting provider's representative in the Philippines had no problem sorting it out. (I assume they make you talk to a person for this as an anti spammer measure)

I understand having a bot to filter out those who wouldn't need help if they just RTFM, leaving the humans to talk to those with actual issues. But when the bot seems to actively interfere with the aim of stopping you from being able to speak to somebody... that's the sort of tone-deaf nonsense that leads to people taking their business elsewhere.

That is why I ditched my energy supplier. Their customer support was dreadful and it was virtually impossible to communicate with a human; then they had the cheek to spam me with their latest offering... they were branching out into offering broadband too! I nearly fell off my chair laughing. No way, no chance, not a hope in hell.

The trouble is those customers who won't RTFM are always adamant they need to speak to a human and only a human can solve their problem. But I wonder if effective filtering is beyond LLMs? Recognising they don't know and need to escalate is beyond the kind of "reasoning" they are capable of. They'd more likely hallucinate an answer.

it is usually because it is a tricky problem I've not been able to find a solution to on their help pages, FAQ etc.

Through experience with chat bots and speaking bots, they are utterly useless and either try to direct me to the FAQ which didn't resolve my query, a non-existent web page, or they fail to understand what I am asking and keep asking me to repeat the query using fewer words, or go around in circles trying different tactics none of which resolve my query or simply crash and give me an error message. I can't think of a single occasion where an automated support bot has actually resolved my query.

To be fair, the humans providing customer support for Vodafone are only marginally better than bots. Sigh.

Next step: hire competent people.

I know, I'm dreaming.

having to either navigate through systems so restricted that they might as well be navigating a choose-your-own-adventure book, or come up against the brick wall of an AI unable to understand or resolve their query.

Yes VM has both. Gets you from both sides. Endless non-relevant options all of which lead to other options that are just as irrelevant, which eventually end up with an option that just contains the recorded message that they will send a link to the self-help page.

I wonder how much I could have made in consultancy fees to tell them this when it first reared its ugly head?

None, I guess, as all they could see was "shareholder value". A pity they got the direction of movement wrong...

Well, if you were not connected to the right people, you would have made nothing.

But you knew that, right? :)