Gen AI in enterprises - playtime is over

With GPT-3, Generative AI has taken us all by storm. I still remember where I was when I first played with the model.

But not everyone is riding the wave as smoothly as some others.

In the talks that I give about Machine Learning to educate enterprises, I notice that there are companies who are scrambling to adopt genAI, and that they often do that without a clear understanding of what they are actually dealing with.

They think they are getting a magical AI genie when, but in fact, they are just getting a really fancy calculator.

Generative AI, is just one piece of the AI pie.

You have Logistic Regression models, Support Vector Machines, K-Means Clustering models, K-Nearest Neighbors, Gradient Boosting, and so on...

And all those models have been around and used for a long, long time.

My first encounter with Machine Learning was in 2008, where I was asked to build a dynamic pricing model for a temping agency. The models that I used were a Time Series models (LSTM) to forecast future demand for different job types and also locations, a Gradient Boosting model (XGBoost) that I used to predict a base price for each job-candidate match, and with that going, I needed to apply a Reinforcement Learning model to continuously optimize prices based on real-time data on the booking rates of temping staff, changing market dynamics, and of course....client behavior.  

And last but not least, all these model outputs needed to be combined using an ensemble method to generate final price recommendations. And then some tweaking to make sure that we hit a balance between competitive pricing, profitability, and of course fairness.        

Yet, lots of folks in boardrooms and IT departments are mixing it up with every other AI type out there.

According to Gartner, this confusion comes partly from tech vendors who use jargon like "AI agents" and "AI models" wherever they can and for any situation.

Research from IDC shows that over 60% of companies have dipped their toes into generative AI, but most are still stuck in the kiddie pool. I see them building old generation rule-based chatbots, with an natural language AI kinda look and feel interaction on top of it.

They are afraid to dive in because they are unclear on what generative AI can realistically do and they are bogged down by issues like data governance and risk management.

Read the article: Build a data strategy before you commit to AI ! before you start with any ML/AI project.

Deloitte’s latest report adds to this confusion. They state that there is also a lot of head-scratching about where and how it can genuinely add value.

The boardroom boys just feel the rush to jump on the AI bandwagon, and that is leading some companies to forget that AI is not a COTS solution.

AI is more like a Swiss Army knife, and it is only useful if you know which tool to use.

Before we start!

If you like this topic and you want to support me:

1. Comment on the article; LinkedIn appreciates that and it will really help spread the word ??
2. Connect with me on Linkedin ??
3. Subscribe to TechTonic Shifts to get your daily dose of tech ??
4. TechTonic Shifts has a new website, with podcast, courses, and more ??

Generative AI used in only 5% of production use cases

Generative AI isn’t doing all that much heavy work in real-world applications.

Generative AI represents about 5% of AI used in production today.

In other words, it’s hogging 90% of the hype but only putting in 5% of the work.

McKinsey's 2024 survey backs this up.

It is showing that 65% of organizations are reporting that they are using generative AI in at least one business function, but that most are just fooling around with it.

They are stuck in a perpetual state of "pilot project purgatory", where proof-of-concept never quite turns into proof of profit.

??

(cannot help but laugh at my own jokes sometimes)

But the sad reality is that many companies are finding that the return on investment for generative AI is not what they expected when they ventured out.

Deloitte’s Q3 report shows that while some companies see some gains in productivity and cost savings, these are often overshadowed by challenges like scaling and integrating gen AI into existing systems.

And do not forget the strain on the current employees (AI burnout at work is a thing!)

The problem is that companies are bedazzled with flashy AI features but haven’t got a clue about how to make them work in their day-to-day operations.

AI Agents are being confused with static AI mdels

I see that there is a major mix-up happening in the (enterprise) AI world.

For instance, the confusion between AI agents and AI models.

This isn’t just a technical nitpick though...

This is leading to some big, and expensive mistakes.

The lure of Agentic AI is that of massive cost savings. That is, if you use them correctly.

In that case, companies like Fount (read Fount helped save 13.4M in nine months with AI based work optimization), Blue Prism (read AI based Intelligent Automation: [RPA + BPM] x AI = ROI) could save you tens or even hundreds of millions.

You have to know that an AI model is like a cookbook.

It is full of potential, but it does not make dinner unless someone does the cooking.

Some models are built by combining several underlying models (ensembles), and other models function as standalone algorithms. So the cook needs to know about the ingredients, and how to cook them all together.

On the other hand there is the AI agent. That is more like a personal chef that actively does stuff for you. And often does that without needing your constant input. One of the approaches is inactive and requires (some) manual intervention to operate, and the other one is capable of performing tasks autonomously without constant input.

I would like to stress that confusing these two concepts is a significant error.

Research from McKinsey shows that some organizations are getting better at distinguishing between these AI concepts. But that there are still many which are still falling into the trap of treating them as equal to one another.

The result is that these companies are spend big bucks on what they think are intelligent, autonomous systems and that they find out that they have bought a pricey toy that needs lots of manual labor to be useful.

IDC’s findings also shows that organizations are getting frustrated because they expect these AI tools to be more "plug-and-play" than they actually are.

Well, wake up and smell the coffee....

The AI magic wand is more of a manual.

AI confusion is causing costly mistakes for organizations

This confusion around AI is quite expensive.

I have seen companies make some monumental blunders by misunderstanding what AI can and cannot do. Some have tried to use a static AI model where a more dynamic AI agent would have been better, and that was leading to production delays, and lots of wasted budgets, and even more facepalming in strategy meetings.

Others have over-engineered AI solutions for problems that could have been solved with simpler, and cheaper methods.

They are using a flamethrower to light a candle.

It's just overkill and costly.

And the reason is that because of the hype, all of a sudden, people are bombarded to head of AI without even knowing the difference between a Regression model and a Convolutional Neural Network.

And because they lack this fundamental knowledge, they rely heavily on their suppliers, who of course want to showcase every nifty feature they have.

This is just a recipe for a disappointment.

AI does not revolve around GenAI:

McKinsey's research shows that nearly 44% of organizations have had a negative experience with generative AI because they have usesd it wrong. The reason is because of inaccuracy and poor model management. Deloitte also tells us that companies need to invest more in understanding AI's nuances and applications before just diving in headfirst.

The developments in AI are moving so quickly, that it is easy to get swept up in the excitement and that you forget the basics.

And this is a costly mistake. But you will hopefully make it only once.

AI ‘playing time’ is over

The focus should be on operationalizing AI

The good thing is that some companies are coming back down to Earth.

It is like they are back from a "recess", a period where all kids are playing around with generative AI because it is the new cool toy in town.

And now, the school bell has rung, and it is time to get back to class.

The focus needs to shift from "Wow, look what this can do!" to "Okay, how do we actually make this work for us?"...

Remember Dee Dee from Dexters's Lab?

I see some organizations who are now investing more in data management, model training, and governance to make sure that their AI initiatives don’t just look good on paper but also deliver results.

McKinsey’s 2024 survey confirms that organizations are starting to develop better frameworks for evaluating AI performance and ROI, and that they are moving from hype-driven spending to more value-based investments.

So people, the time of AI experimentation is gone, and now it is all about making those experiments count.

Signing-off Marco

Well, that's a wrap for today. Tomorrow, I'll have a fresh episode of TechTonic Shifts for you. If you enjoy my writing and want to support my work, feel free to buy me a coffee ??

Think a friend would enjoy this too? Share the newsletter and let them join the conversation. LinkedIn appreciates your likes by making my articles available to more readers.

Top-rated articles

• Anticipating AI's next move ? article ① ? | LinkedIn
• Anticipating AI's next move ? article ② ? | LinkedIn
• Anticipating AI's next move ? article ③ ? | LinkedIn
• Anticipating AI's next move ? article ④ ? | LinkedIn
• AI/ML news summary: week 36 | LinkedIn
• We need to develop a vision for Machine Customers | LinkedIn
• How I used AI to interpret my MRI | LinkedIn
• A peek into my AI bookshelf | LinkedIn
• This company uses AI to help people optimize their work, and save millions. | LinkedIn
• AI based Intelligent Automation: [RPA + BPM] x AI = ROI | LinkedIn