[threetonesun]

Same argument for removing customer service with chatbots or AI. It's entirely untrue, and creates a much worse customer experience, but because people drop out your KPIs / NPS is based off of people who were willing to put up with shit to get to a real human.

[DavidWoof]

Give me an AI chatbot over someone with poor English skills reading a script any day of the week. My problem probably isn't unique, it's probably something fairly obvious that was vague in the instructions.

Now, the important thing is offer a way to upgrade to a human. But I have no problem at all starting with AI, in fact I honestly prefer it.

[rcxdude]

That doesn't really match my experience. Usually if my problem is not unique it's already documented somewhere and I've solved it that way (And support generally puts some effort into documenting the non-unique problems to reduce their workload). If I'm calling support, it's because I've exhausted all other options and I've either concluded I need them to do something I can't do with an online form or the information is not at all accessible elsewhere, in which case first line support is nothing but an obstacle.

[nijave]

It's hit or miss. Sometimes screaming "give me a compotent human" at a chatbot is quicker than pleading with tier 1. Sometimes it's not

At least there's no hold time

[threetonesun]

Sure, because you've already lived with 10+ years of enshittification in the process. Customer support used to be an in-house team that was actually trained on providing relevant support, not an outsourced call-center that's as (or more) useless than a chatbot.

In some ways it's not that different with hiring. I used to work with HR teams that knew the roles they were hiring for extremely well and could make reliable calls on whether or not to pass a candidate to a hiring manager. More recently I've seen HR get outsourced entirely, or staffed with cheaper employees that just shuffle documents through systems.

[gowld]

The AI and the human are both programmed to avoid helping you.

[MattGaiser]

At this point I find the humans know so little that an LLM referencing documentation or past support answers is superior.

[AnimalMuppet]

Well... is a chatbot for customer service really all that much worse than a human who is not permitted to deviate from their script?

[eviks]

Certainly, because not deviating from the scripts also cuts off the infinite range of made up nonsense a bot can hallucinate. And it's not like the bot will have magic authority to fix the real issue it can't be bound by the script, so in this regard there is no upside.

[rafabulsing]

Chatbots != LLMs.

We've had chatbots for a long time before LLMs, and while they're of course much more limited as you have to explicitly program every thing it should be able to do, by that very virtue, hallucinating is problem they do not have.

For this kind of customer service chat scenario, I find them much better than just a free style LLM trained in some internal docs.

(Though really, probably the ultimate solution is a hybrid one, where you have an explicitly programmed conversation tree the user can go down, but with an LLM decoding what the user is saying into one of the constrained options. So that if one of the options is "shipping issues", "my order is late" should take me there. While other forms of NLP can do that, LLMS would certainly shine for that application)