Let’s cut through the noise for a second.
We’ve all heard it—the breathless headlines, the CEO keynote proclamations, the LinkedIn posts declaring that “AI will transform everything by next quarter.” Billions are being poured into generative AI. Startups are minted weekly. Enterprise software vendors are slapping “AI-powered” on everything from CRM tools to lunch-ordering bots.
But here’s the thing no one wants to admit over morning coffee: Where’s the money?
Not the potential. Not the promise. Not the “hours saved” or “NPS points gained.” I mean actual, hard-dollar, bottom-line, show-up-on-the-P&L profit.
Turns out, it’s mostly missing.
According to McKinsey’s latest State of AI survey, only 11% of companies report significant, measurable impact on company-level earnings from their GenAI investments. Let that sink in. Nearly nine out of ten businesses are betting big on AI—and getting ghosted by returns.
S&P Global went further: in 2025, 42% of companies abandoned most of their AI projects, up from just 17% in 2024. That’s based on a survey of over 2,400 IT decision-makers—not armchair theorists, but the people actually signing off on budgets and timelines. Of those who stuck it out, only one-third broke even. Fourteen percent lost money.
Consider this: nearly half of the AI proof-of-concepts never get deployed.
Investing is at an all time high, yet we see the greatest disconnect between investment and financial return. Disappointment is systematic and profits are nowhere to be found.
So, what gives?
I've spent the last decade building and influencing products in enterprise SaaS and more recently, AI-driven platforms. I've learned that the productivity and profitability equation is the reason organizations feel “efficient” and are burning through cash.
Lets get the facts straight. The truth about AI ROI is clearer than you think.
The Measurement Problem: When “Impact” Isn’t Enough
You are the product manager for an AI feature that summarizes customer support tickets instantaneously. Completion time is 50% faster and dashboards are beaming with your success.
That is until the CFO walks in. “Did this reduce our support headcount? Did it decrease operational spend? Did it prevent customer churn?”
Long silence.
The truth that I've learned through trial and error is that every feature must always achieve one of these three goals: make, save, or retain money.
That’s the triad. Anything else is decoration.
In B2B, it’s the same. Companies aren’t buying AI powered project management tools because they “feel futuristic.” They expect these tools to:
- Increase revenue: win more deals, upsell faster
- Decrease costs: reduce manual labor, shrink cloud spend
- Retain customers: lower churn with better service
If your AI doesn’t address one of these with a dollar value, you aren’t in business, you are in wishful thinking.
ROI Isn’t a Buzzword—It’s a Math Problem
Let’s talk ROI. Real ROI. The kind not slapped on a slide deck with vague arrows marching upward.
ROI = (Hard Dollar Benefit – Total Dollar Cost) ÷ Total Dollar Cost
What’s missing? Hours saved. NPS scores. Lines of code generated.
These metrics are in the productivity realm, not profitability. And, while productivity is a positive metric, it doesn’t pay the rent.
I’ll admit, I fell into this trap, too. We built an AI assistant that helped sales reps draft emails 30% faster. Everyone cheered, but when we dug deeper, we learned: no one was closing more deals. There were no commissions gains.
Earlier in the day, the reps just... wrote fancier emails.
That isn't ROI. That is theater.
To a CFO, saved hours mean nothing. In the end, it translates to the same overtime, headcount, and budget. If the headcount remains the same, those 'saved hours' just disappear into more meetings, Slack threads, or, let's be realistic, happy hours.
How then, should we measure AI's impact in the real world?
Vanity metrics mean nothing. Consider these three questions:
Did the AI create new revenue streams or just new ones, like a new market or a premium pricing tier?
Did it change the cost of the business, like fully automating a process that previously required 10 FTEs, or reducing the headcount overall?
Did it curb revenue loss, like retaining customers who were about to churn?
If you cannot answer 'yes' to 1, with supporting data, you are measuring the wrong thing.
Take Meta's 'Year of Efficiency' in 2023 for example. They didn't just deploy AI tools and walk away. They paired it with aggressive restructuring and automation, to the tune of a 22% headcount reduction. The outcome? Operating margins doubled. That is ROI.
Now, consider the failed McDonald's AI drive-thru pilot of 2024, which ended with confused customers and robotic misorders, then went viral. Zero return.
For example, studies show GitHub Copilot allows developers to finish tasks 55% faster, yet Microsoft still has yet to disclose any financial return.
An independent study busted the notion that bug rates and cycle times in production improved when AI was implemented.
Three companies. Three AI stories. Only one with actual investment.
So, Why Does AI Initially Fail to Deliver Expected Returns?
Having no ROI from AI in the initial phases is not surprising.
Getting returns from AI is like growing a plant. You need to water it, provide nutrients, ensure it gets sunlight, and it will grow in the long run.
AI improves through iterations. It becomes smarter through feedback, accuracy improves with frequent usage, and value increases with deeper workflow integration. This is the reason for the lag in returns; in some cases, it could be up to 24 months.
The sad reality is that most companies deploy AI in a silo. For example, they implement an AI chatbot for customer support, but the entire support process is not reengineered. They utilize an AI tool for content marketing, but it’s not linked to campaign performance data.
The AI becomes an expensive sidekick to the workflow; these sidekicks won’t impact the lines in the P&L statement.
It's a recurring theme in multiple companies: teams track metrics such as “tasks completed” and “emails drafted” but no one is asking, “Did this influence customer behavior? Did it shift revenue?
The Productivity Obsession—and Why It Backfired.
Let’s rewind to 2022.
ChatGPT is released. The world goes crazy.
Every executive starts to preach “AI-driven efficiency.” “Productivity” becomes the new buzzword. Meta claims 2023 the “Year of Efficiency.” KPIs shift: Profit margins are replaced by “time-to-market,” and “customer acquisitions” now focus on “lines of code per hour.”
For a month, we’re all pumped.
But by late 2024, the reality is clear. Speed ≠ value. Faster code doesn’t matter if no one wants the feature. Quicker reports are worthless if the decision is poor.
Worse, the hidden costs of speed trapped value:
- AI model fine-tuning,
- Prompt engineering.
- Consulting.
- Data pipelines.
- Legal reviews of hallucinations.
All that unproductive overhead.
Let’s not forget the human cost. Entry level roles are disappearing. Not because AI replaced them, but because the unproductive focus on AI gave companies the perfect excuse to freeze hiring and expect the same output from fewer people.
Even I see it in the comments under my own videos. “I can’t get an internship. Every job now says ‘AI experience required.’”
That’s not progress.
It’s displacement.
It’s riskless displacement.
The hidden winners are the AI “wrappers” that make real profit.
But it’s not all doom and gloom.
The AI wrapper economy is revolutionizing how money is made, and it is doing so quietly.
There is no need to spend billions on building foundational models. Savvy business owners are taking LLMs like GPT-4, Claude, or Llama and wrapping them into domain-specific applications.
They are not doing it for fun. These applications are tightly woven into workflows and are sold for considerable amounts of money.
Take for example Harvey, the legal artificial intelligence co-pilot. It does not only summarize case laws. It drafts motions, predicts how judges will rule, and integrates with firm billing systems. It is worth $5 billion and has an ARR of $75 million.
Or consider Nisphere. This artificial intelligence coding assistant is aimed at enterprise engineering teams. It does not only autocomplete code. It understands legacy systems, tracks feature impact, enforces compliance, and tracks feature impact. It is worth $2.5 billion and has an ARR of $100 million that is rapidly increasing.
People made money because they solved specific, and costly, problems and every feature was tied to an amount of money.
They sold profit protection, not productivity. Chase after value, not hype.
AI will not magically improve your business. It will only improve your business by compounding value. Treat AI like the strategic asset it is, not like a shiny toy.
There is no “AI transformation.” Focus on AI integration. It should be woven into core workflows, not placed on the sidelines.
Assign every use case to a P&L line.
Evaluate the profitability of a function, and not the profitability of the function.
Hire those who have knowledge of AI and the business domain.
Finally, be patient.
You will not be replaced by AI next quarter, but you will let people who know how to use AI to amplify their productivity outpace you. They see AI not as a cost of operation but as an opportunity to derive economic value.
Consider a situation where your CEO declares, "We're all in on AI!" Can you see the ROI on AI?
If you answer no to the question, it might be time to have a challenging conversation.
Capitalism cares about the bottom line and not your AI strategy.
The money is still hiding.
The money is not gone. It is just hiding from those who are unwilling to do the challenging and mundane tasks of integrating AI to real business results.
This is where the revolution starts.
Memorable moment: I remember pitching an AI feature for the first time to our CFO. My pitch started with "faster reporting by 50%." He stared and asked, "faster for who? What cost?" I had no answer, and it was a powerful moment for product strategy.
Also Read: Building a Sustainable AI Product: Beyond Professional Replacement
Initial productivity hype.
Even so, I witnessed teams burn out trying to do “more with AI” as leadership kept ignoring the underlying structural issues. AI did not cause that, but it did sustain the illusion that we were “solving” problems without making real changes.
AI truly just amplifies what’s already there, whether it be a good strategy or a bad one.
0 Comments