Listen to the audio here:
There’s a number buried in a recent industry report that should stop you mid-scroll.
72% of AI initiatives have failed to scale across the business. 20% have already stalled, failed outright, or been abandoned. Only about a third are hitting the ROI targets that justified them in the first place.
Not “underperformed.” Abandoned.
Quietly, the way you abandon a gym membership: no funeral, no announcement, just a card that keeps getting charged for something nobody uses anymore.
Now watch what happens when you point that lens at the biggest name in the room.
This Isn’t a Salesforce Story
Salesforce has bet the house on AI agents.
Its platform, Agentforce, is selling fast. Paid deals more than tripled in about nine months: from 3000 deployments in January to 9500 in August.
Then an investment bank surveyed Salesforce’s own partners. These are the firms paid specifically to make Agentforce succeed for customers. When asked what was driving their new business, not a single respondent named it.
Not one.
It’s tempting to read that as a Salesforce problem. It isn’t. Salesforce is simply big enough, and public enough, that its version got measured and printed.
The same pattern is sitting inside a huge share of the “we added AI” initiatives launched this year. It shows up in companies of every size, on every platform.
Everybody’s hiring the AI workforce. Almost nobody’s checking whether it showed up for the shift.
So, what’s actually going wrong?
There’s a clean answer, and once you see it, you’ll start spotting it everywhere.
Two Businesses, Two Invoices
Picture two owners buying the exact same technology.
The first hears “AI agents can automate anything” and thinks big. Ten agents. Ten workflows. Customer service, follow-up, scheduling, the works. Full rollout.
Six months later, nothing is obviously broken, so nothing is obviously better either. The only thing in the whole project with perfect clarity is the invoice.
The second owner asks a smaller, sharper question: “My team spends ten minutes on every order-status call, two hundred times a day. Can this fix that, specifically?”
One agent. One job. One number. Six months later, that owner has a sentence they can say out loud in a board meeting without flinching.
Same technology. Opposite outcomes. The first bought a hack. The second bought a result.
Hacks Have a Shelf Life
Here’s the part almost nobody is pricing into their AI budget.
A hack is milk, not wine. The moment something works, it gets copied.
Competitors pile in, the vendor turns it into a template, and the edge evaporates on its own schedule. Whatever clever workflow you deployed this quarter, somebody will sell it cheaper next quarter.
What doesn’t expire is the discipline of knowing. You know precisely and honestly what worked, what didn’t, and what it cost you to find out.
That discipline is rare. Most companies rushing into AI don’t have it. What they have instead is a dashboard that’s afraid of bad news.
The Dashboard That’s Afraid of Bad News
Here’s the tell.
A measurement system that has never once delivered an uncomfortable answer hasn’t been tested.
It’s been decorated.
Real measurement is occasionally rude. It says things like:
We don’t have enough data to call this a trend.
We didn’t track the cost, so we can’t tell you the return.
This result is unknown.
And unknown is not zero. Stop treating it that way.
Most dashboards won’t say any of that. They quietly round the gaps toward “fine.” A missing cost becomes $0 instead of “we don’t know.” Eleven results become a “conversion rate.” That rate gets printed to one clean decimal and carried into the next leadership meeting as fact.
Nobody’s lying, exactly. They’re just letting the software’s defaults do the flattering.
Before I spent two decades inside P&Ls, I trained as a physicist. Every physicist learns this rule early: a measurement without error bars isn’t a measurement. It’s an opinion holding a ruler.
Most AI dashboards are all ruler, no error bars.
When AI Guesses Beautifully
There’s a second cost hiding underneath the first, and it’s the one that should worry you most.
AI doesn’t always fail loudly. Sometimes it fails beautifully. The sentence is polished. The spreadsheet is clean. The task is marked complete.
But the AI was guessing. It wasn’t lying. It filled a gap with something plausible instead of admitting it didn’t know.
When nobody is honestly checking the work, that guess gets logged as a finished task. It feeds the next workflow. It rolls up into the dashboard. It lands in the boardroom looking exactly like progress.
It isn’t progress. It’s a decision nobody actually made, wearing a decimal point.
At that stage you don’t have an AI problem. You have a measurement problem in an AI costume.
The Hunter and The Verifier™
This is the distinction I keep coming back to, and it reframes the whole AI conversation.
Think back to the Gold Rush. Thousands of prospectors swarmed the hills, and they were very good at finding shiny rocks. But the people who actually knew who was rich weren’t in the hills. They sat in the assay office, weighing every rock and delivering the verdict: gold or fool’s gold.
The prospector is the Hunter. The assayer is the Verifier™.
AI has become a phenomenal Hunter. Agents can now find the broken workflow, the wasted hour, the leaking dollar and the unclaimed opportunity with almost no hand-holding. That capability is arriving fast, and it gets cheaper every month. Hunting is close to solved.
The Verifier is the hard part…on purpose. It stands at the end of the hunt and asks the questions the Hunter never will:
✅ Was this worth finding?
✅ Did fixing it move a number that matters?
✅ What did it actually cost?
✅ Did the result hold, or did it fade after the demo?
✅ Would we make the same investment again?
Here’s what separates a real Verifier from a decorated dashboard: it is willing to say “this didn’t work” as plainly as it says “this did.” It says so before the number gets rounded into something flattering, and before the next invoice.
A Verifier that has never delivered bad news isn’t a Verifier.
It’s a cheerleader with a login.
That’s not a technology problem. It’s a discipline problem. And discipline is the one thing you can’t download.
▶ Want to see The Verifier™ in action?
Watch the full IconicTV LIVE episode, The Verifier Gap: What Every AI Hack Forgets to Build, on IconicTVLive.com.
What We Build Differently
Any consultant can hand you agents right now. That part of the market is being commoditized in real time. Anyone with a demo can sell you a Hunter.
What we build alongside the agents is The Verifier™.
It’s the same forensic discipline I built TIPS™ around, years before “AI agent” was a phrase anyone used. TIPS™ is an honest read of what a business is actually doing with its money, with nowhere for the gaps to hide. It follows the evidence through a P&L. The Verifier™ points that same forensic eye at your AI workforce.
The question stops being “Did the bot do something?”
It becomes “Did it do the right thing, at a cost you can defend, with a result you’d put your name on in front of your board?”
That’s the whole bet behind AI Workforce Design.
Not more automation. Automation you can actually trust the report on.
So Here’s the Question:
Forget Salesforce. Forget Agentforce. Forget the demo.
Look at your own business and ask:
If I pointed an honest Verifier at everything I automated, funded, or delegated this year, how much of it would survive? Not the dashboard version.
The real one.
Most owners don’t know. That’s not a failure of effort, because you can work incredibly hard and still measure the wrong thing. It’s a failure of measurement, and it’s exactly the gap TIPS™ and The Verifier™ were built to close.
Two Ways In
If you want the honest answer instead of the flattering one:
Take the Revenue Leak Detector Quiz
A fast, no-fluff diagnostic that shows where money is quietly leaving your business, before you spend another dollar automating around the problem.
A forensic, one-on-one read of what’s actually working in your business, and what’s just decorated to look like it is.
Either way, you leave with a real number.
Not one wearing a decimal point for show.
