Companies pour billions into AI, but where is the money? Numbers don't lie.

Mackenzie's latest state of AI survey reveals that only 11% of companies report significant tangible impact on company level earnings from their Genai investments.

S&P Global found that 42% of companies abandoned most of their AI projects in 2025, which is up from just 17% the year before.

And that's based on the survey of more than 2400 IT decision makers. Only onethird said that they broke even and 14% recorded losses.

The average company in the US abandoned about 46% of AI proof of concepts before reaching production.

We're witnessing the greatest disconnect between investment and financial returns in modern business history. Record AI funding meets systematically disappointing financial results.

We hear AI this, AI that, but show us the money.

This is episode five of my series AI hive versus reality. And in this episode, we will dissect the topic of how much AI is actually making for businesses in the US, if anything at all.

Let's dive in.

The measurement problem. First of all, let's talk about how AI impact gets measured. How do we know where the line is between nice to have AI and AI that's actually moving the business metrics?

I'm going to speak about it from the perspective of a product manager. Every time I put something in my product, it is my job to ensure that the feature that I'm putting in achieves one of the three things. It either makes money, saves money, or retains money.

In the B2B world, the overarching idea is pretty much the same. When a business signs a contract with a new vendor, let's say a new AI project management tool, they do that with one of the three goals in mind. The app will either bring money, retain money, or save money for the business.

Now, what is an ROI? ROI only exists when there is a dollar tag attached.

ROI is not when you can write 15 emails instead of 10 in the same time frame.

ROI isn't when your app summarizes last month's analytics so you can prep a deck faster to report results.

ROI is a finance ratio. Hard dollar benefit minus total dollar cost divided by total dollar cost.

Outcomes that stay in hours or NPS point never reach the numerator.

Productivity and ROI are never the same. And the metrics that measure the two are not the same. And the signals that you're getting from tracking both of them are not the same.

So how should ROI be measured when it comes to AI products? Revenue growth. Did the AI solution lead to actual new sales, upsells or market expansion?

Cost reduction. Did the AI automate anything fully or did it optimize processes to actually lower operating expenses and retention or savings?

Did AI help retain customers who might have otherwise turned?

Now, let's talk about the metrics that look impressive, but the ones that die on CFO's desk, the so-called vanity metrics.

The most obvious one is hours saved. It's a big round number, but unless headcount or overtime spent actually goes down, this is potential savings, not book dollars.

For engineering, PR is merged or lines of code generated. Development platforms service this metric by default, but it shows speed, and speed does not equal value.

Faster code only matters if it ships revenue producing features sooner and that time gain gets monetized.

Tickets resolved per agent. It does make support dashboards look heroic, but it needs translation into actual headcount reduction.

Here's a quick litmus test. Does the metric appear on the profit and loss statement? Revenue, cost of goods sold, the day-to-day cost of operating a business like salaries, rent, software, or forecasted cash flow.

If yes, it's a candidate for ROI. If not, it's a productivity or quality metric.

Now, why does traditional ROI not work for AI investments? It doesn't work because AI creates value slowly and methodically.

Unlike standard technical systems, AI improves over time as it learns from more data and feedback.

With AI apps, rapid ROI calculation is extremely unreliable because it often improves things that do not have that clear-cut cost savings or immediate revenue.

The benefits are harder to tie directly to financial metrics because they show up months or years later.

AI rarely works in isolation. It often coincides with other changes like new processes or software or business models.

But when you're being asked to quantify the impact, it's really tough to pinpoint what portion of value stems from AI specifically.

But the problem is that the companies often do deploy AI in isolation. So measuring ROI is really hard.

But productivity is a lot easier to throw in a dashboard and tell a story. So that became the new narrative of the modern tech. Look how productive we all are.

And productivity became the northstar, the productivity obsession.

There is evidence that starting 2022 when GBT came out, US companies started shifting their focus from traditional profitability metrics to productivity metrics.

The shift occurred during that AI hysteria when everybody thought that AI tools would significantly boost efficiency and speed and productivity even more so than the short-term profits.

Now why did companies move towards productivity metrics? A lot of leaders rebranded their corporate strategy around efficiency as seen notably at Meta in 2023 where the company labeled the year of the year of efficiency amplifying productivity.

KPIs across various industries were redefined and expanded by AI. Companies started measuring things like development time or speed of time to market. moving beyond just standard profit margins.

By mid to late 2024 and into 2025, a lot of companies realized that productivity gains alone were not translating to clear bottomline results such as profits.

Companies also learned that the cost of maintaining AI doesn't always get justified by productivity gains.

Productivity metrics grew and plateaued and companies started rebalancing towards traditional profitability such as net income or cash flow.

So the narrative and executive commentary around AI changed in 2025 saying that to measure AI ROI, you have to be measuring productivity and profitability together, not in isolation.

Now let's look at three realworld AI rollouts and three wildly different bank statements.

GitHub Copilot helped developers finish tasks 55% faster in controlled experiments, but Microsoft has not reported any corresponding value.

And an independent study showed no cycle time improvement and a higher bug rate.

Meta's year of efficiency paired AI tooling with mass layoffs. They cut their headcount by 22% and ultimately doubled their operating margin.

There's your example where AI delivered the ROI because cost structures were aggressively re-engineered.

McDonald's AI drive-thru pilot built on IBM's LLM promised labor savings, but ended in viral ordering failures and was shut down in 2024 with no returns whatsoever.

Three companies, three productivity stories, and only one outcome.

Now, why didn't productivity obsession work? Productivity jumps are clear for individuals, for example, faster report writing or customer service resolutions.

But aggregating them across the company often reveals less impact. Time saved does not immediately result in more output or revenue because that time can be reallocated to less measurable or lower value tasks.

For example, meetings. Can we all just admit that all the time we're able to save by not writing updates or documentation results in leaving the office an hour earlier or using that time to book more unnecessary meetings.

Most companies have implemented AI in isolated teams or business units rather than transforming entire workflows end to end and lots of resources were spent on training and updating those workflows, AI outputs and change management.

These overhead costs often offset productivity improvements.

The only thing that AI has done for sure is that it has contributed to disproportionate difficulty getting into entry-level roles. I can see it in comments under my own videos.

Only companies that deeply change their processes and invested in AI talent consistently see bottomline results. Most don't.

Let's talk about some non-obvious trends. There is a clear evidence that American VC investment in AI startups including the so-called GBT rappers continues to be very strong and even growing in 2025 despite a broader shift towards focus on profitability.

And that is happening because the sector is benefiting from the maturation of generative AI.

Before 2023, VC investment in AI startups was largely driven by innovation hype and rapid growth ambitions.

Investors put a huge focus on cuttingedge technology and groundbreaking applications. And funds were often thrown at all kinds of ideas without clear paths to profitability.

This is what VC scene looked like two years ago. Large amounts of capital poured into early stage startups focusing primarily on novel AI capabilities rather than business models or financial returns.

Valuations shot through the roof. Everyone was throwing money at anything that sound even remotely generative AI.

Emphasis was placed on capturing market share and technological leadership quickly over immediate or near-term profitability.

The deal cycles were very fast and many investments were speculative bets on the transformative potential of AI expecting returns in longer term.

And lastly, Soft Bank's early big bets on OpenAI and other visionary startups exemplified that growth at all cost mentality.

But starting 2024, there has been a clear shift towards a different kind of evaluation. Investors are way more careful now.

They're looking closely at unit economics, real traction, customer retention, and all the boring stuff that actually matters.

The investment dollars have concentrated more on mature or enterprise ready AI.

So, you might be thinking, well, that's it for GBT rappers, then. Nope, not dead. Welcome to the rapper wars.

The AI rapper concept emerged as startups began building specialized applications on top of powerful existing AI models like GBT or Claude or Llama.

Instead of developing foundational AI from scratch, which requires immense resources, companies started wrapping AI capabilities into domain specific products.

The AI rapper economy matured beyond simple interfaces to more complex and more functional applications.

Rappers evolved into multi-layered applications that solve specific industry problems like legal or healthcare or finance or software engineering.

They embed AI into existing business workflows and they often offer very advanced features.

For example, legal AI rapper Harvey grew to $5 billion valuation and 75 million annual recurring revenue.

Coding AI rapper Nisphere reached 2.5 billion valuation with 100 million ARR rapidly.

The market for AI rappers expanded drastically with the generative AI industry projected to hit 38 billion by late 2025. And the biggest part of this prediction is driven by rappers.

So in summary, the AI rapper economy has evolved from rapid hypedriven wave of simple API based applications to advanced and arguably the biggest industry segment in AI.

It has given a chance to founders big and small to commercialize AI by wrapping sophisticated models into practical and revenue generating applications.

Conclusion. So despite the hype, AI doesn't automatically generate returns. And that's not pessimism. It's pattern recognition.

So when you hear headlines like AI is revolutionizing everything, ask where is the money.

Productivity is not ROI. Automation is not profitability. Time saved is not money made unless you can show it on your P&L.

Media loves bold narratives because they drive clicks. But business doesn't reward narrative. It rewards profits.

The US is still the land of capitalism. And underneath the noise, the AI ROI curve is quite slow.

So, stop the panic. You're not going to be replaced overnight. AI is not moving fast enough to wipe your entire career in a single quarter, but it is moving fast enough that you cannot afford to relax.

You've got time to upskill, but use it well.

As always, I hope this was helpful. Let me know what you guys think in the comments. We'll see you next time.