When investors evaluate any capital deployment, whether it’s a new factory, a new hire, or technology like AI, they ask four questions:
What is this going to cost? When will I realize a payback? What do the long-term dividends look like once payback is realized? And how does this compare to other investments I could make?
For AI, it is increasingly evident that nobody seems to know how to answer these questions consistently. In fact, they do everything they possibly can to dance around them.
In July 2025, MIT released a study: 95% of AI pilots fail to deliver ROI. Three months later, Wharton released a study: 74% of companies see positive AI ROI. Both studies arrived at definitive conclusions. Both are also methodologically flawed in ways that prevent them from painting the entire picture. And they can’t. MIT measured P&L impact within six months, a timeframe too short for most enterprise technology. Wharton measured productivity gains and satisfaction scores, metrics that don’t always translate to profitability.
Nobody reads methodology sections. Nobody reads the fine print. They read headlines. And headlines are creating two diverging perceptions: “AI is failing” versus “AI is working.”
Both are meaningless blanket statements that miss the point entirely. The only thing that matters is the economics of your specific business. AI, as a growing cost center, just happens to now fit into that conversation.
What these studies do is intensify an already noisy conversation about how to measure AI ROI. From the wildly reliable and rational hill of independent astute observation, I can tell you the measurement discussion has become so freaking overcomplicated.
There are frameworks from the National Institute of Standards and Technology (NIST), approaches from IBM, methodologies from Wharton, analyses from MIT, and proprietary models from every major consultancy. Each claims to be comprehensive. Each requires expertise to implement.
The measurement industry replaced “does this work?” with “are we measuring correctly?” and became the primary barrier to AI ROI itself.
As such, I’ve been inspired to share my thoughts in hopes of reducing some noise. What follows is my take on why this phenomenon, The Measurement Tax, is happening, who truly profits from it, and what I, myself, and my intellectual capital firm, Orpical Technology Solutions, do regularly to recalibrate ROI discussions for financial clarity.
How Measurement Became the Barrier
Here’s what’s perplexing. When I am in the field and talk to clients, they ask fair, straightforward questions: Will this save us money? How much? How fast? Can we afford it?
Yet, when I see what colleagues in the industry are building and talking about, it’s largely theoretical. It’s complex technical jargon. It’s measurement frameworks. NIST’s AI Risk Management Framework. IBM’s Hard/Soft ROI model. Wharton’s 72-metric approach. MIT’s GenAI Divide analysis. McKinsey, BCG, Deloitte all have their versions.
I’m not here to hate. These aren’t stupid people. They’re smart. Really smart. Many of them are much smarter than I am. But smart people can sometimes build elaborate answers to the wrong questions.
In my experience, the noise is not coming from organizations looking to invest in AI. It’s coming from the industry itself. Publications, RFPs, positioning, thought leadership. And that noise becomes the barrier because it shifts the conversation. What should be “does this work for our business?” becomes “which measurement framework should we use?”
By the time that question gets answered, the deployment ship has sailed.
The Business Model Behind Measurement
All this noise doesn’t happen by accident. There’s a clear incentive structure. Like every other industry, and with every other modern marvel, money talks. And it talks loudly. The winners and losers are often defined by the players who can spend the most to secure authority and swing positioning in their favor for long-term prosperity.
Traditional consulting firms make more money from longer engagements. A six-week sprint that answers “does this save money?” generates less revenue than an 18-month framework implementation. When projects fail, the framework provides cover: “We followed best practices. The organization wasn’t ready.” The firm still gets paid. The framework still gets referenced in the next pitch.
Academic institutions need novel frameworks for publishable research and grant funding. “Use simple metrics” doesn’t get published. “A multi-dimensional approach to AI value realization” does. Papers lead to citations, speaking fees, and executive education programs at $15,000 per seat.
Executives get political cover from prestigious names. Asking the board for $10M works better with “we’re following MIT’s approach” than “we’re going to try things and iterate and go from there.” Whether it succeeds or fails, you followed a rigorous methodology.
Middle management justifies roles through measurement programs. AI governance analysts, framework coordinators, change management leads. These positions exist because complexity exists. Simplification eliminates the role.
Everyone has an incentive to make the forward investment and implementation of AI more complicated than it needs to be. No one has an incentive to make it simple.
The cost isn’t the consulting fees or the new headcount. It’s velocity. Companies spend 12-18 months evaluating frameworks and building governance structures while competitors and disruptors ship, learn, and iterate. By the time the measurement program launches, the technology has evolved and the competitive window has closed.
When Complexity Provides Real Value
To be fair, I want to acknowledge that some situations genuinely require comprehensive frameworks. Healthcare AI making diagnostic decisions. Financial services AI making lending decisions. Defense systems using AI for autonomous operations.
When failure means death, systemic risk, or catastrophic loss, measure everything. That’s maybe 5% of deployments. The other 95%? Automating workflows. Handling inquiries. Creating content. Processing documents. Analyzing data.
These often don’t need NIST compliance. They need a handful of KPIs.
The problem is: the frameworks designed for medical devices got applied universally. Necessary rigor became mandatory process for sales tools. This is creating an unnecessary measurement tax where organizational overhead slows deployment without improving outcomes.
Don’t get me wrong. I’m not advocating for no measurement. I’m advocating to put a stop to measurement systems costing more than the thing being measured, thereby killing potential unrealized returns. It’s the classic conundrum of spending five dollars to chase down one.
Four Questions for Financial Clarity from an Unlikely Source
A Quick Backstory
Okay, before sharing my gross simplification of measurement, I think it’s now important to distinguish a few crucial characterizations about myself. I say characterizations as a formal disguise of “the things that make me weird.”
My first quirk, often unbeknownst to many people that meet me for the first time, is the fact that my background is in journalism, not computer science.
The misclassification of me as an engineer is grounded in fair assumptions. Most of my conversations with friends and family, my team, clients, and industry peers center on technology and digital transformation. This is largely because it is my livelihood, and there are many people who depend on me. I must be sharp. I must be on the frontlines. I must be learning something new every single day. I must be a subject matter expert.
As Co-Founder of Orpical, this isn’t just my responsibility. It’s everyone’s. Tech changes so fast that the definition of competence shifts monthly. What qualified as expertise six months ago might be table stakes today. My team understands this. We’re all learning together, staying sharp together, because that’s what survival in this space requires.
To use an analogy, imagine a chef eating out at dinner with their spouse and a couple of friends. An appetizer is served at the table. The couple, regular patrons of his restaurant, perk up. “Chef, that dish looks incredible! But we have aggressive peanut allergies. Do you know if it’s safe for us to eat? Can you tell us what’s in it?” The chef shrugs. “I don’t have the foggiest idea,” he says, reaching for a fork. “Let’s find out.”
Those are the caught-with-your-pants-down scenarios that keep me up at night.
To sleep a full eight hours uninterrupted, I take great pride in sharing a new Git repository with my engineers. I take great pride in collaborating with a client’s data scientist to present unconventional approaches to fine-tune their machine learning model. I get a rush and a great sense of relief from being able to sit in front of the C-Suite and effectively address what PHP stands for. Spoiler alert: it’s a recursive acronym. Hypertext Preprocessor is what it stands for. Yes, I learned the hard way—like a chef who couldn’t distinguish a dish with peanuts. But at least I know I put in the work to learn the tools of my trade.
With all that said, I’m not a programmer. I don’t know how to write code. I can’t query a database. I can’t configure a CI/CD pipeline. I mean, I’m sure I could figure it out using my background as a journalist, but that’s not my strength, and that’s not where I provide the most value.
Where I provide the most value is by leveraging my journalistic background and what I’m most passionate about. To ask the right questions. To get to the truth as quickly as possible. To then leverage that extracted information to write a story, or in the context of business, a comprehensive business case.
To craft this business case, I lean on another character quirk, one that I gained far earlier than my formal college education. It started when I was a baby.
According to sources, my mom’s best friend growing up, a woman named Sandra, who now doubles as my de facto Aunt Sandra, there was only one thing that could effectively stop me from crying when she babysat me during my teething stage. The Philadelphia Inquirer.
Not the comics. Not the sports section. The finance pages.
Aunt Sandra would open the stock listings and market summaries and read them aloud. Ticker symbols. Price movements. Earnings reports. Within minutes, I’d stop crying. She thought it was absurd. My mom thought it was hilarious. I apparently thought it was exactly what I needed to hear.
I don’t remember any of this, obviously. But it tracks. Because decades later, I still find clarity in numbers that show you what’s actually happening. Not theory. Not frameworks. Just signal.
That instinct, combined with a journalism background that taught me to ask questions that cut to the truth quickly and the financial discipline of bootstrapping Orpical with zero outside capital, shaped how I think about measurement. When your own livelihood is on the line, you stop asking “are we measuring correctly?” and start asking “does this make us money?”
The Four Questions (Again)
What I like about our business is we take the best of a few different worlds—traditional consulting and engineering, startups, and venture capital—and throw them in a blender.
From consulting, we learned to ask the right questions and solve problems systematically. From acting like a startup, we learned that speed matters and you can’t hide behind process when survival is on the line. You must finish the last mile. You must ship fast. From venture capital, we learned how investors evaluate whether to deploy capital.
That last one is where these four questions came from. Remember them from the intro?
- What is this going to cost?
- When will I realize a payback?
- What do the long-term dividends look like once payback is realized?
- How does this compare to other investments I could make?
These aren’t AI-specific questions. They’re capital deployment questions. And they work because they force you to think like an investor, not like someone implementing technology.
Now, financial professionals might object that the above questions oversimplify capital budgeting. Fair enough. As I stated, I’m not a CFO. I’m an undercover journalist and business efficiency unicorn. And research consistently shows a gap between what capital budgeting theory prescribes and what actually drives decisions in practice. Even when companies use NPV calculations, IRR analysis, discount rates, and sensitivity models, managers fundamentally want to know the same things: how much cash goes out, how much comes back, and when.
The sophistication isn’t in inventing new questions to ask. It’s in how you calculate the answers. NPV tells you present value. IRR tells you rate of return. Discount rates adjust for time value and risk. Sensitivity analysis stress-tests assumptions. All of these methods exist to answer those four fundamental questions with greater precision.
Recalibrating Conversations Back to The Big Four
Most AI deployments don’t generate revenue directly. They improve productivity, increase customer satisfaction, reduce risk, or accelerate workflows. These benefits are real. They’re just not immediately obvious in financial terms.
The translation is straightforward: soft metrics create hard outcomes. Productivity gains either reduce labor costs or increase capacity to produce more output, generating more revenue. If your sales team spends 10 hours per week on proposal generation and AI reduces that to 2 hours, you’ve freed 8 hours per person per week. That’s either cost savings—you need fewer people to maintain output—or capacity increase—the same people can handle more deals. Calculate it: 8 hours × 50 weeks × fully-loaded hourly cost × number of people. That’s your annual dividend simplified. Not even factoring in new deal flow. Not bad.
The productivity gains mean some roles will change. But employees who learn to work with AI, who understand how to apply it to new problems, become exponentially more valuable. They command higher compensation and better job security. AI creates value and those who increase their value proposition through AI fluency capture it.
That’s the internal story. Externally, the focus must always be on the end user. Customer satisfaction improvements reduce churn or increase conversion. If AI-powered service reduces response time and your churn rate drops 2%, calculate annual revenue at risk from churn and apply the reduction. Boom. You have a dollar figure. Risk reduction prevents losses. If AI-powered fraud detection catches more fraudulent transactions, calculate prevented losses per year. That’s your dividend.
The pattern is consistent across every metric. Productivity? Cost reduction or revenue increase. Customer satisfaction? Revenue increase. Decision quality? Cost reduction or revenue increase. Risk mitigation? Cost reduction. Workflow acceleration? Cost reduction or revenue increase.
There are no other categories. Every business metric ultimately affects your P&L through cost or revenue. If it doesn’t, it’s not a business metric. It’s a vanity metric. This is why framework complexity is unnecessary for most deployments. You’re not measuring dozens of independent dimensions. You’re measuring variations of two things: does this reduce what we spend or increase what we make?
Aligning Incentives
Earlier, I wrote:
“Everyone has an incentive to make the forward investment and implementation of AI more complicated than it needs to be. No one has an incentive to make it simple.”
That was a setup for the real truth.
“Everyone has an incentive to make the forward investment and implementation of AI more complicated than it needs to be. No one has an incentive to make it simple. Except the people who are actually vested in the outcome.”
When your compensation depends on actual business results rather than engagement duration, everything changes. You can’t spend six months evaluating measurement approaches. You can’t build governance infrastructures that cost more than the returns they track. You can’t hide behind best practices when projects fail. You ask the four questions, calculate the answers quickly, build, and ship.
The AI economy will force this realignment whether the industry wants it or not.
Consulting firms will need to shift from billable hours to outcome-based pricing. Get paid when the AI delivers returns, not when the framework gets implemented. Academic institutions will need to publish research that helps companies deploy faster, not measure more comprehensively. Executive education programs should teach speed and financial discipline, not governance theater. Executives will need to justify AI investments with projected payback timelines, not prestigious methodology names. And middle management will need to facilitate deployment velocity rather than building roles around measurement complexity.
Everyone’s in business to make a profit. Consulting firms. Academic institutions. The companies deploying AI. Even the executives and middle managers. Their compensation and career progression depend on business performance. When incentives align with actual outcomes instead of process adherence, measurement simplifies naturally. You track what drives cost reduction or revenue increase. You correlate outcomes to the metrics that actually matter: cash out, cash in, and time to payback.
I know the measurement frameworks won’t go away by writing this article. The discussion will continue. But I firmly believe the companies that win will be the ones that treat AI like any other capital investment and move fast enough to capture returns while competitors are still evaluating which measurement approach to use.
Decades ago, Aunt Sandra would read me stock prices and earnings reports to stop my crying. In hindsight, I think I stopped crying because what she was reading was stupidly simple. It was black and white. These companies are doing good. These companies are doing bad.
Five years from now, the companies reading in that finance section will be the ones who shipped while their competitors tried to figure out what to measure. The ones who asked four questions instead of implementing seventy-two metrics. The ones who moved fast enough to capture returns while the measurement tax killed everyone else’s.
When it comes to AI, you can think like an investor and end up in the finance section. Or you can put your business in the back with the obituaries.
Truly, it’s up to you.