You bought the tools.Your people are using them.So why can't you point at what changed?
I'm Glen Caruso. I work with senior leaders on the decisions that come before the software. Which problems AI should touch, which ones it shouldn't, what gets built first, and what should never get built at all.
Most failures are decided here, long before anything is built.
Six months in,nobody can point at a number that moved.
"We're not seeing much impact from AI."
"Adoption has been slower than expected."
"It's helpful, but not really changing anything."
"It's hard to measure ROI."
"We thought we'd be further along by now."
The tools work. The decisions never got made.
A company buys Copilot or ChatGPT. Leadership expects people to use it. Someone gets handed "figure out AI" on top of their actual job. A few people run with it. Most don't. The tools end up layered on top of the same work, so the work stays the same. That's not a software failure. Four things never happened.
- 01No clear decision on where AI should and should not be used.
- 02No ownership for outcomes.
- 03No workflows actually redesigned.
- 04No definition of what success looks like.
Most of this gets decided months earlier, by nobody in particular, in a meeting that didn't feel like a decision.
Every engagement comes back to four questions.
You could answer them in a room with a whiteboard and no software at all. Most companies never sit down and do it.
Three ways in.You can stop after any of them.
A session with your leadership team to frame the real decision.
Not a workshop. Ninety minutes with the three or four people who actually decide, and one question: where does AI belong here, and where would it quietly create risk? You leave knowing what is at stake before anyone commits capital, headcount, or reputation to it.
The Decision Context Diagnostic. The work that should happen before a vendor walks in.
I go through what you have already committed to and what you are being sold, and I come back with the ones to stop, the two or three worth the money, and who owns each call. You walk into your implementation partner with the decisions already made instead of paying them to make the decisions for you.
Ongoing advisory, for when the decisions keep coming.
Working sessions with your senior team as new calls arrive. The goal is that your leadership makes decisions the same way whether or not I'm in the room. Clear framing takes drag out of an organization, and that's something you actually control.
What Phase Zeroactually produces.
I call the method DecisionLabAI. There's a working version of it at decisionlabai.com.
Phase Zero happens before anyone builds anything, and it decides what gets built. Five answers come out of it. The first one is usually the one that saves you the most money, and it is the one no vendor will ever give you.
What you should stop
The initiatives already in motion that will not return what they cost. Named, with the reasoning, so you can defend killing them.
What actually deserves the money
The two or three places where AI changes a number you report. Scored against what your organization can realistically absorb.
What order it happens in
Sequenced so each thing makes the next one easier, instead of four pilots running in parallel and none of them finishing.
Where AI has no business being
The judgment calls that stay human, decided on purpose and written down, before someone automates one by accident.
Who owns the call
A name against every decision. Most AI programs die because the answer to "whose is this?" is everyone.
// Engagements are scoped in conversation, not sold from a menu.
Some of this will land.Some of it won't.
✓ This is for
- Senior leaders and revenue executives operating under real accountability.
- Founders and investors pressure-testing high-stakes decisions before they commit.
- Institutional decision-makers shaping where AI capability gets built.
- Operators who value judgment over hype.
- Teams making decisions where failure is visible.
× This is not for
- Anyone who wants a tool demo and a box checked.
- Buyers looking for a vendor to hand the problem to.
- Anyone who needs the answer to be flattering.
- Work where the stakes, and the standards, are low.
Glen Caruso
- Enterprise dataOracle Data Cloud
- Enterprise salesAdobe, 3.5 yrs, quota‑carrying
- CategoryFinServ & capital markets
- InstitutionalUSC CIC, 18 months
- Teaching AI sinceJan 2023, UGA
- VerifyLinkedIn
I sold enterprise software before I advised anyone about it. Oracle Data Cloud, then three and a half years at Adobe carrying a number in financial services and media.
Which means I've been the person who has to defend the spend in a QBR, not the one who presents the deck and leaves. I know what it costs you personally when something you championed doesn't work.
I started teaching AI in January 2023, two months after ChatGPT launched, at the University of Georgia. Eighteen months with the University of South Carolina's College of Information and Communications since then, plus private K-12 academies and a state AI collaborative.
Across all of it I kept seeing the same thing. Tools got bought. People got told to use them. And the decisions that would have made the tools matter, what to stop doing, who owns the outcome, what good actually looks like, never got made by anyone in particular. That's the work. Upstream of the software, downstream of the strategy.