A field guide for CEOs who actually sign the checks
You're being sold AI like it's oxygen. Every vendor, every consultant, every LinkedIn thought leader swears you'll be obsolete by Tuesday if you don't deploy their magical models immediately.
Meanwhile, MIT just confirmed what you already suspected: 95% of enterprise AI pilots deliver exactly zero measurable business impact. Not "less than expected." Zero.
So here's the question nobody wants to answer honestly: When should you actually trust this stuff, and when should you politely escort the vendor to the door?
I'm not going to give you "it depends" nonsense. Here are the actual decision criteria, based on what separates the 5% who succeed from the 95% who light money on fire.
5 REASONS TO TRUST AI (And Write the Check)
1. The Problem Already Costs You Real Money
Not "inefficiency." Not "suboptimal workflow." Real, measurable cash leaving your business every month.
Lumen Technologies didn't start with "let's do AI." They started with "our sales team spends 4 hours researching each prospect, and that's costing us $50 million annually."
They didn't build a model. They bought a solution that compressed research time to 15 minutes. Saved the $50M. Done.
The test: If you can't show me a P&L line item this will improve by at least 20%, you're not solving a problem - you're buying a science experiment.
2. You're Buying from Vendors, Not Building from Scratch
MIT's data is brutal: vendor partnerships succeed twice as often as internal builds. Twice.
Your internal team is talented. They're also optimistic about timelines, confident about their ability to solve problems they've never seen, and completely unaware of the integration nightmares they're about to encounter.
Vendors have already failed with 47 other companies. They know where the bodies are buried.
The test: If your CTO says "we can build this in 6 months," add 18 months and triple the budget. If that number still makes sense, proceed. Otherwise, buy it.
3. You're Starting in the Back Office, Not Customer-Facing
Here's what nobody tells you: the highest ROI in AI isn't sexy. It's not chatbots. It's not personalization engines.
It's eliminating business process outsourcing. It's cutting external agency costs. It's automating the tedious garbage work that humans hate doing.
Over 50% of AI budgets go to sales and marketing because that's where executives see the demos. But the money is in back-office automation where mistakes don't end up on Twitter.
The test: If the AI touches customers before it touches internal operations, you're optimizing for "cool demo" over "actual ROI."
4. You Have Adult Supervision Built In
The companies that cross the "GenAI Divide" don't run perfect pilots. They run contained failures.
They expect things to break. They design blast radius. They have kill switches. They treat early deployment like a controlled explosion, not a product launch.
They don't avoid friction—they weaponize it to learn before scaling.
The test: If your implementation plan doesn't explicitly include "what we'll do when this fails catastrophically," you're not ready. You're just optimistic.
5. Someone Besides IT Signs the PO
When AI spending moves from "team discretionary" to "executive budget authority," something magical happens: accountability.
Suddenly there are metrics. Suddenly there are reviews. Suddenly "we're collecting data to train the model" doesn't pass as a success metric.
The shift from "innovation theater" to "this better generate revenue" is the difference between pilot purgatory and actual deployment.
The test: If the budget owner can't explain in 30 seconds what business metric this improves, kill it. They're not ready to own the outcome.
5 REASONS TO RUN LIKE HELL (Or at Least Slow Your Roll)
1. The Vendor Can't Show You Where It Failed
If someone pitches you AI without talking about failure modes, they're either lying or dangerously inexperienced.
Deloitte just had to refund $440,000 to the Australian government because they used ChatGPT to generate a report with fabricated academic citations. They didn't disclose the AI use until after the fake references were discovered.
That's not a "whoops." That's a governance failure that ends consulting relationships.
The red flag: They only show you success stories. They can't articulate specific risks for your use case. They say "the model will learn." Run.
2. Data Quality is "Something We'll Figure Out"
43% of companies cite data quality as their primary AI obstacle. Not model selection. Not compute power. Data.
If your data is garbage, AI just automates garbage at scale. It doesn't magically clean itself. It doesn't "get better with more data" if the data is systematically biased or incomplete.
Walmart spent TWO YEARS building new data collection infrastructure before deploying AI for inventory. They didn't try to force existing data into new models.
The red flag: The pitch includes phrases like "we'll start with what you have" or "the model will normalize the data." Translation: they're hoping your data isn't as bad as they suspect.
3. ROI is Measured in "Efficiency" Not Dollars
"Our AI will make your team 30% more productive!"
Cool. Show me the headcount reduction plan. Or the revenue increase. Or the cost avoidance.
"Well, they'll just be able to focus on higher-value work..."
Nope. That's code for "we have no idea if this actually makes money."
Shadow AI exists because 90% of workers prefer consumer tools to enterprise AI. Your employees are already using ChatGPT because your "official" tools are slower and dumber.
The red flag: The business case is built on "time saved" without showing where that time converts to money. If you're not reducing costs or increasing revenue, you're buying a nice-to-have.
4. Compliance is "We'll Add That Later"
The EU AI Act went live February 2, 2025. Colorado's AI law hits June 30, 2026. The Trump administration just created an AI Litigation Task Force to challenge state laws, which means regulatory uncertainty for the next 12-24 months minimum.
If your vendor's answer to "how does this comply with [regulation]" is "we're monitoring the situation," you're the beta tester for their compliance strategy.
The red flag: No documented impact assessment. No decision-rights architecture. No clear ownership when the model makes an expensive mistake. They're punting governance to "later" because it's hard and expensive.
5. Your CEO Thinks AI is Magic
This is the big one.
If your CEO believes AI "learns on its own" or "gets smarter over time" without understanding the governance required, you're headed for disaster.
The companies failing aren't failing because the technology doesn't work. They're failing because leadership doesn't understand that AI implementation is an operating model problem, not a technology problem.
You're not just deploying software. You're redesigning how work gets done, how decisions get made, and how value gets created.
The red flag: Leadership can't articulate specific risks. They use phrases like "we need to stay competitive" instead of "this solves X problem costing us $Y annually." They're buying AI because they feel like they should, not because they know why they must.
THE BOTTOM LINE
The question isn't "should we do AI?"
The question is: "Are we doing AI for adult reasons or because we're afraid of looking stupid?"
If you're solving a real problem, buying proven solutions, starting where mistakes don't go viral, building accountability from day one, and operating with executive ownership - you're probably fine.
If you're experimenting because everyone else is, hoping data will magically organize itself, measuring success in vague productivity gains, and treating governance as optional - you're about to join the 95%.
The difference between the 5% and the 95% isn't technology.
It's whether there's an adult in the room asking "will this actually work?" before the check clears.
Mike McKenna runs TEAM Solutions and the Founder's Ally initiative, advising CEOs of $3M-$30M companies on AI-first pivot positioning and governance frameworks. He's the guy who tells you if it will actually work before you spend the money.
----
Disclosure:
Article by Mike + AI
Header image by AI