Even Google Can Scrww Up (So can their AI!)

[Paul Kelly ICHYB]

I have a Google basic business starter account, which suits my needs (I actually have 4!!).

Today I received an email saying I should really use my features more, including storage of 5T. My account only has 30G per user, so I thought that they had boosted plan benefits, however, going onto my account, it still shows 30G.

Off to support I go and type my query into their all singing & dancing AI chat machine. In an instant, it comes back with ' Based on your account information, your organization has Google Workspace Business Plus, which includes 5 TB of storage per user. ' Ahh, I think, maybe they have upgraded me and I missed the email.... I then start questioning AI why I cannot see my additional storage, but start going around in circles...

I next ask for an Agent and in a few seconds I am connected to Kadapa who was very friendly, but kept telling me that I do not understand the email and AI messages, because they are not telling me about what I have, but what I could have (i.e. if I upgrade). Despite requesting several times, I could not be guided to the sales message on the email or suggestion of AI. I do not want to take the position of 'bloody Indian call centres', but they could not see that the messages being delivered to me were not what they were being told they were meant to be! The email was in plain English and not devious or misleading

On the email, I would expect that Google have some very smart, well paid, people managing their campaigns, who know the difference between a sales message, call to action and signposts. Also, they probably know how to segment and target messages. But I could be wrong.

I also expect their support AI to be able to see what type of account I have - this makes me worry even more about using it in my businesses!

Sorry for a bit of a rant, however, it shows how, if the biggest and best can get it wrong, we should not beat ourselves up if we have the odd hiccup!

 

[Optegris]

See this nips at me. Google has been throwing money into AI development and is now on the cusp of releasing agent-to-agent solutions through its new sparkly platform, but it still cannot get the basics right.

I had a similar rubbish experience with the Amazon chatbot recently until I eventually got an agent to help.

AI, chatbots, and intelligent solutions can and do work exceptionally well, but only if they are trained and implemented correctly. You can't just slap a bot on a site and then sack your support team. You will always need people who fully understand how this stuff works, how to make the bots smart, and how to provide a fantastic customer experience to the point where they don't realise it's a bot.

That's the key.

and I'm trying desperately hard not to self-promote

This post was brought to our courtesy of a human and not a LLM bot

 

[eteb3]

This post was brought to our courtesy of a human and not a LLM bot

How would we know, when you’re so good at training them?

 

[Gecko001]

Maybe a bit off-topic, but a few years ago, I would get people at my door wanting to speak to Mary, James, Fred, etc. I told them that they did not live here. They had got the right number of house, but the wrong street name, which is easily done, but some were so insistent that they had the right address that they became rude. I did a search on Google Maps of all the streets in my area with my number, but Google's arrow thing was not pointing at my house.

Eventually, I started to get mail from a newish development of flats nearby, so I put the name of that development into Google Maps and hey presto the big arrow thing pointed to my house. Google did not recognise the flats at all, even after a couple or three years after the flats had been built.

I just redirected callers to the block of flats and that solved the problem, but it took several years before the callers stopped coming.

 

[Optegris]

How would we know, when you’re so good at training them?

You will never know

 

[fisicx]

AI, chatbots, and intelligent solutions can and do work exceptionally well, but only if they are trained and implemented correctly.

Got an example of a chatbot that has been trained and implemented correctly?

 

[ctrlbrk]

Got an example of a chatbot that has been trained and implemented correctly?

I have had a positive experience with ChatPDF.

Apparently limited only to extract information from PDF files and then answer any questions, but it seems to do so more precisely than any other chatbot I've tried.

 

[Optegris]

Got an example of a chatbot that has been trained and implemented correctly?

Zapier.

Their bot has been around for a while now and keeps getting better and better. I've used it a tonne for automation support.

 

[Optegris]

Of course, the best is subjective, as the Zapier bot would be rubbish in an e-commerce environment as it hasn't been trained to work there.

As a side note, I use Claude an awful lot locally for coding and writing, and then ChatGPT for research. They all have their own strengths but being able to hook into these LLM models and then use services such as Zapiers MCP for extensibility, there is nothing that you cannot achieve.

Edit: Hope you are well btw Graham, been a loooooong time since the days of Openmind

 

[fisicx]

A very long time.

For those that don’t know Phil ran an excellent hosting company and really looked after his clients. But was then brought out as what usually happens.

He also had more hair back then.

 

[Optegris]

He also had more hair back then.

True. I blame the clients