Sorry, but the ROI on enterprise AI is abysmal The deployment of AI projects and associated return on investment (ROI) have declined, according to a large survey of IT decision-makers. Appen, an AI data services company, working with The Harris Poll queried 500 IT decision-makers across various US industries for its report, The 2024 State of AI. The results indicate that …
No, but in this case the product is being marketed to prey on the FOMO of the executives of large corporations. Ain't their money that they are spending, and these are the same demographic that have fallen for the Emperor's New Clothes since the days before time.
I suspect all of us working in or close to large organisations will see this: Organisational decision makers are showing precisely ZERO prudent scepticism towards the claims, costs and possible benefits of AI.
Ha! Come look at my large, sclerotic enterprise. We're not moving into AI at all because we need to study it first, like cows staring at a cell phone tower, and then shuffle a few glacial POCs forward in about three years. Meanwhile the services are all blocked at the firewall and devs have their personal iPads at their desks for conducting "research" thru the guest network that can be emailed back to the office machine 2 feet away. Progress!
El Reg» The firm in July said about a third of senior leaders report that their organizations are overseeing broad AI initiatives. Among those organizations, "investments are delivering positive returns, especially in areas like operational efficiencies (77 percent), employee productivity (74 percent) and customer satisfaction (72 percent)."
Which middle-manager wants to give mediocre (or even bad) news to their C-Suite boss about the roll-out of the AI-doodah in their department? So, after the rollout of the AI project, customers are happier, employees are more productive and operations are more efficient. Indeed, the project was so successful that senior management celebrated by firing 2,500 staff.
There are many aspects of business where people would struggle to find evidence that what they are doing actually works. I foresee a "lack of high-quality training data" being a chronic problem with most of the things that the snake oil merchants want to sell.
Fortunately (for them, not us) the target market is full of people who regard their own opinions as high-quality training data. It appears likely, then, that we're going to build artificial yes-men and they'll sell like hot cakes.
I don't think that is their point.
Your definition of success is long term. If you redefine success as 'someone give a start up lots of money that they spend on huge salaries and nice expenses whilst not actually delivering much and just enjoying the ride' then this could be very successful.
... is unlikely to be a good strategy. Given a choice between chatting with an AI customer-service system and chatting with a human, what percentage of customers would choose AI? When you deliberately do something that your customers don't want, you have no right to expect a positive ROI.
But that choice is never given. I can't say that I've ever seen a company with two helplines: one with AI and one with real people. Instead, the first the customer hears about it is when they bounce off the AI some time _after_ they've bought the product about which they're enquiring.
Aren't they basically saying that the AIs have the same issue as people: if they're not properly educated they can't do the job you want them to?
It's a business decision. Instead of employing people in a call centre who can't grasp your problem or help with it, you get some AI thing that can't grasp your problem or help with it.
A while back, I tried to get some help from a large internet retailer. They had delivered my package and sent me a photo of it to prove that. There was no option to communicate that, while that may or may not be my package, that is definitely not my house. No way to get it resolved. Luckily the good sport who had received the thing brought it round the next day.
Unfortunately this all too true. I buy a lot of stuff from amazon because it's easy to do so and there is a wide range of items available. I'd prefer in a lot of ways to but from other companies but often their range is nowhere near as good, the delivery options just aren't there and that's before one gets to actually trust. As bad as it is, I trust amazon to deliver more than I trust many other websites, I know what I can do and what to expect when a delivery or product fails but I don't have the same with other websites. Which is annoying to admit because many of these other websites likely offer very good service.
Its a basic law of knowledge that all books come from the books written before them.
If you have enough books, you can infer the books that came before them.
An more importantly, you can exfer the books that come after them.
Weezencake's Unreliable Algorithm
Mentioned in The Last Continent, this ancient spell could be used to search L-space for unwritten books. However, it was very slow, taking years to put together even the ghost of a page. Placed in the hands of Hex, however, the spell can be made to run thousands of times a second, resulting in the compilation of very large fragments of books like How to Dynamically Manage People for Dynamic Results in a Caring Empowering Way in Quite a Short Time Dynamically. Sometimes.
Do NOT give a manager this book.
AI Models can still work, you just need to avoid feeding them on all the free crap you can find on the open internet and instead feed them a steady diet of high quality information.
The crap that AI's produce is not high quality information unless it is refined and tested for quality before being fed back in.
It is not hard so long as you remember thermodynamics. You cannot get something from nothing.
Couldn't agree with you more on this point.
I spend more time reviewing output of AI text for accuracy and applicability to the actual original query.
I'm just not seeing the value at the moment either, it's a novelty that's becoming a distraction with a weighting beyond the value it brings.
May of those low-paid humans who check that data are chasing late payment of their earnings this month after a poorly planned systems implementation by Appen.
https://www.reddit.com/r/CrowdGen/comments/1g7sy5j/the_new_crowdgen_from_appen_is_a_failure_we_need/
The IT story that Appen has been trying to bury with these inane press releases. (By the way El Reg was tipped off about this on Sunday)
https://www.afr.com/technology/appen-ceo-apologies-as-tech-failure-leaves-workers-unpaid-20240814-p5k2ex
It's not too bad for a pension fund if whoever manages it is aware of the fads and bubbles that come with such things. For instance, invest early and get out at what is determined to be a peak before it all collapses. Although the pension fund manager has probably now been replaced by a machine learning tool... which if it has gathered the knowledge base to operate in this way should be OK, but a complete pending disaster if not.
This is a moment at which you can start a business that can do things at a larger scale, or offer the customer more, for the same cost. Don't be a Luddite like most of the people who comment on El Reg's AI articles: make use of the new capabilities and leave them behind in your dust trail!
It's highly likely to be very messy. Too many investors have thrown too much money into the scheme that they cannot afford for it to fail. As a result, they either lose a lot of money and face now or continue pouring money in on the hope that they might get their money back. At some point such a scheme will either crash, and take a lot of investor money down with it (it's remarkable how billions of funding can disappear with nothing tangible to show for it), or somehow, possibly, it will slowly recoup the investment. Slowly. Any investor wanting instant profit will bail and that then crashes things.
As for how long it will take LLM tools to actually be useful to an extent where the return on investment happens? Who knows! But the results do not scale linearly with the investment in resources and infrastructure. Double the resources applied and we may see a 10% improvement. That kind of ratio cannot persist for too long but it's the current way things are going.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
