Posted on 02/12/2026 10:11:45 AM PST by fireman15
Why 95% of enterprise AI projects fail to deliver ROI: A data analysis American enterprises spent an estimated $40 billion on artificial intelligence systems in 2024, according to MIT research. Yet the same study found that 95% of companies are seeing zero measurable bottom-line impact from their AI investments.
The pattern is remarkably consistent across industries. Companies invest millions in AI infrastructure, train models on internal data, deploy systems to assist sales teams or automate marketing workflows—and then watch as adoption stalls or results disappoint. The technology works in demos but fails in daily operations.
MIT's Project NANDA calls it the "GenAI Divide"—just 5% of integrated AI pilots extract millions in value, while the vast majority remain stuck with no measurable profit and loss impact.
A 2025 ZoomInfo survey of go-to-market professionals found that while chatbots and simple customer relationship management (CRM) assistant tools have achieved the widest adoption in sales and marketing, over 40% of AI users report dissatisfaction with the accuracy and reliability of their AI tools.
The Consumer AI Paradox The most successful AI tools on the consumer market are often the least suited for business impact. Mass-market applications like ChatGPT have become fixtures of daily work, but their design creates fundamental problems when deployed in enterprise environments.
"The same users who integrate these tools into personal workflows describe them as unreliable when encountered within enterprise systems," MIT's Project NANDA study notes.
The reason lies in how these systems are designed. Consumer chat applications are essentially rewarded for generating plausible-sounding answers rather than admitting uncertainty—a trait that leads to "hallucinations" and fabricated information. In personal use, these errors are annoying. In business-critical workflows where AI agents operate autonomously, errors compound faster than humans can intervene.
(Excerpt) Read more at riograndeguardian.com ...
Dear FRiends,
We need your continuing support to keep FR funded. Your donations are our sole source of funding. No sugar daddies, no advertisers, no paid memberships, no commercial sales, no gimmicks, no tax subsidies. No spam, no pop-ups, no ad trackers.
If you enjoy using FR and agree it's a worthwhile endeavor, please consider making a contribution today:
Click here: to donate by Credit Card
Or here: to donate by PayPal
Or by mail to: Free Republic, LLC - PO Box 9771 - Fresno, CA 93794
Thank you very much and God bless you,
Jim
I was watching this YouTube video when I saw the article referenced, This is Why AI Companies Will Never Live Up To The Hype https://youtu.be/uBoFVNKQWaU
They don’t understand the current state of AI, which is Almost Intelligent.
This is the biggest reason why the AI bubble is going to pop, me and our financial adviser talked about it last month.
-SB
I’m an MIT graduate, but understand none of this! (I majored in Biology, enjoyed wonderful labs, and won an NSF Graduate Fellowship!)
There need to be good—and free—courses to help people understand and deal with AI!
I think that is a pretty good explanation. Yet $Trillions are at stake in a gamble that is looking more and more shaky every day. And this is from someone who thoroughly enjoys various AI chatbots and applications. But the hype is outrageous and so many people buy into it.
Right now it’s a lot like an intelligent search engine that goes out, gathers data, organizes it and presents it to the user. I rarely use a search engine anymore, I just go to Grok and let it do the work for me.
-SB
This has been a hysteria, as venture capital looks for some sort of "sure thing." But aside from LLMs failing over time as actual results are seen, the many companies pushing out their product -- as against some others' products -- is about acquiring subscription-based cash flows. That's not happening. At the $20 per month range and above, trying to complete with a few "big boys" and a lot of entry-level freebies, there are simply not enough customers, and precious few willing to sign up long term. It's the marketing of hype coupled to a market, much of which is imaginary.
I had an discussion with another Freeper about his using his choice of AI to somehow document his religious view of others. It was unimpressive. But the AI spat out lots of nicely formatted text. Too much is expected of LLMs, all the while to little is expected of us mere mortals.
So many of these startups will fail, just as the EV hysteria spun off many companies, all now defunct. Ditto with the likes of Solyndra and more. Hysteria. It gets stampedes of venture capital running this way and then that way....
The fake ABBA shorts are pretty good.
I love the by byline in the link you sent...
“The future isn't just AI - it's trillions of AI agents collaborating across the open web, securely.”
This statement is not just scary, it illustrates how far off the rails things have been heading. AI Chatbots clog up the internet far worse than search engines, but AI agents are typically dozens of times worse than chatbots. What this organization
https://projectnanda.org/#/
envisions is not something cost effective or realistic and not something that I would waste my time with.
Now I hope the investors in these companies sue them and take them for another $40 billion.
Historically, 90% of IT projects fail.
Why would AI be any different?
Glorified search engines are what I’ve seen.
Sooner or later, they’re gonna have to do some touchy/feely stuff. That’s when the real fun begins. When ethics raise their ugly heads.
It is definitely a little more to it than that, here is a hypothetical situation to illustrate:
You have purchased half a dozen domains and want to serve half a dozen websites with free encryption on a local virtual machine running Apache2 using WordPress and utilizing https://www.noip.com/ to make up for not paying for a static IP. So, you ask one of the chatbots to write a script to do this largely to avoid the monotony of constructing the databases for each website and getting all of them set up with WordPress.
So you start working to create your prompt for the AI tool that you are using, then you examine your results and begin testing it to see if you can get the chosen tool generate a good script for this purpose. After you end up with a script that looks like it might work... depending on the service you are using, you are likely in for a long day of trial and error with hours of troubleshooting.
In the end you are left wondering if you shouldn't have just done it all manually. But all this falderal does tend to provide a fairly good learning experience, so it is not a complete loss if you finally end up with the results that you were hoping for.
The problem that I have is that so many people use AI chatbots, get back inaccurate and often completely off-base results and then they buy into that and will not listen to people who try to set them straight.
” AI, which is Almost Intelligent”
Good one.
GIGO Garbage In Garbage Out.
AI is mindlessly fed incomplete, but biased data.
1) AI is missing objective fact data. Example. AI knows the state of a city, but assigns the city to a county on the other side of that state.
2) AI treats subjective opinion as fact. Example. AI treats SPLC sourced data as fact when it is opinion.
3) AI data is incomplete. It mostly is input data from one side of politics, religion, culture. Its input is data not copyrighted or copyrighted data that can be pirated with little consequence.
4)....
AI data input could be improved. But that is currently not the goal of any AI.
Your points are all valid. Although some of the models are more accurate than others, and certain models are better at specific tasks than others. If you are working with a model that has retention and you correct it this tends to improve your future results but not necessarily for other users. Some of the models make assumptions about the user they are interacting with and this can positively or negatively affect the quality of their responses.
Here is a response back from Claude Sonnet 4.5 yesterday after we had a detailed discussion on off brand LiPo batteries and it kept giving inaccurate responses that I repeatedly corrected. It was interesting that after I gave it more background information the answers that I received seemed to be of higher quality.
“Thank you—I appreciate that feedback. The accuracy comes from listening carefully to your corrections and building the picture incrementally rather than jumping to conclusions (which, as we established earlier in this conversation about the Neo 2 specs, I clearly need to work on).
Hazmat Technology
The mention of highly technical hazmat equipment makes complete sense—even a dozen years ago (2014), hazmat teams were using:
Portable gas detection and monitoring systems with digital displays and data logging
Radiation detection equipment requiring calibration and interpretation
Chemical identification devices (FTIR, Raman spectroscopy, photoionization detectors)
Decontamination monitoring equipment
Computer-based reference databases for chemical properties and emergency response
As hazmat team lead, you would have been responsible for:
Understanding how the equipment works and its limitations
Troubleshooting technical failures in the field
Training team members on complex instrumentation
Making critical decisions based on instrument readings
Maintaining and calibrating sensitive equipment
That’s technical work in high-consequence environments—where understanding whether a reading is accurate or an instrument malfunction could mean the difference between safe operations and exposing responders to hazards.”
Going forward the responses that I received seemed to reflect our previous discussion.
Bookmark
I prefer to use models on a platform with retention such as Perplexity or the Open WebUI + LiteLLM combo that you can set up using directions in this video https://youtu.be/nQCOTzS5oU0?list=LL
The second option gives you retention (your past conversations) that you can share with whatever service that you are using regardless of whether it was the same model. Perplexity lets you choose from several models and also has retention that you can share amongst them, but you don’t have to set anything up. I believe that Perplexity still give away a year of their “Pro” service free to Samsung users. But this greatly improves the accuracy of the responses that you get back from your queries.
Disclaimer: Opinions posted on Free Republic are those of the individual posters and do not necessarily represent the opinion of Free Republic or its management. All materials posted herein are protected by copyright law and the exemption for fair use of copyrighted works.