For CEOs who’ve tried AI and been disappointed
Stop feeding the pilot graveyard.
A useful AI decision can be a decision not to start yet. Every competitor says “do more with AI.” BCG found only 14% of CEOs can define the P&L impact of their AI initiatives. The rest are spending on experiments no one can evaluate. This page makes the opposite case: do less, but better, with a structured way to decide what to kill, defer, and test.
One-on-one AI work for executives
Michael E. Gruen — Principal, ex-CEO. I use AI in my own workflows every day.
Book the Operating Leverage Session — skip the pilot graveyardFour coaching engagements run at a time.
Who this is for
- You’ve launched AI experiments and can’t tell which are working
- Your team has tools but no one can name the P&L impact
- You suspect some initiatives should stop — but lack the framework to decide
Most AI experiments can’t be evaluated — because no one defined what they were for
BCG’s AI Radar 2026 found that only 14% of CEOs clearly define P&L impact for all their AI initiatives. That number tells the story: most experimentation isn’t undirected because CEOs are careless. It’s undirected because “do something with AI” is not an operating question. Without a specific operating constraint to test against, every experiment looks reasonable in isolation — and none of them adds up to a decision.
The cost isn’t just the experiments that fail. It’s the attention spent on experiments that were never going to matter. That attention could have gone to the one or two operating moments where AI actually makes a difference. Doing more with AI isn’t the answer. Doing less, with a clear test, is.
5 AI projects a CEO should probably kill
Not because they’re bad ideas. Because they don’t address an operating constraint the CEO owns — and continuing them spends attention that should go somewhere it matters.
- The “everyone gets a copilot” rollout. Tool licenses without an operating question. If you can’t name the decision or workflow each license is meant to improve, you’re spending money on access, not leverage.
- The auto-generated content engine. Producing more content faster doesn’t help if the bottleneck isn’t content volume. If your team’s problem is judgment or coordination, more output makes it worse, not better.
- The “AI strategy” slide deck. A deck describing what AI might do, commissioned to show the board you’re on it. If the strategy is “we’re exploring,” the deck is a placeholder, not a decision.
- The custom chatbot for a process nobody owns. Automation layered over a process with unclear ownership. The bot will answer questions; the process will still be broken.
- The experiment with no stop condition. A pilot that’s been “promising” for months without a defined success metric, failure metric, or deadline. If you can’t say what would make you stop, you’re not experimenting — you’re hoping.
[Sample — anonymized pattern, mg approves]
[Sample — a “what we stopped doing and why” pattern: a company that reviewed its active AI initiatives against an operating-leverage test and killed several. Not because the technology failed, but because the experiments didn’t address an executive constraint the CEO owned. Stopping them freed attention for the one or two that did. Details withheld under NDA; client reviewed and approved this summary.]
The pattern: the projects worth killing aren’t the ones that failed technically. They’re the ones that were never connected to an operating question. Killing them isn’t a defeat — it’s the first real decision about AI most companies make.
Test, defer, or kill — a framework for deciding
Not every AI idea needs to be killed. Some should be deferred — either the constraint they address isn’t the one that matters now, or the conditions for a useful test aren’t in place. For each AI initiative, ask three questions.
The three-question test
- What operating constraint does this address? If you can’t name a specific recurring decision, communication, or delegation moment, it doesn’t have one. Defer.
- Who owns the decision this experiment is meant to improve? If the answer is “the CEO” and the CEO hasn’t chosen this as a priority, it’s spending attention without a sponsor. Defer.
- What would make you stop? If there’s no success metric, failure metric, or deadline, you can’t evaluate it. Define the stop condition first — then test, or defer until you can.
If an initiative passes all three — specific constraint, clear owner, defined stop condition — it’s worth testing. If it fails any one, defer it until it passes, or kill it. The goal isn’t fewer experiments. It’s experiments that can actually be evaluated against an operating question.
A structured way to decide
- 01 Baseline Install the stack, land early wins
- 02 Judgment Where AI fits your decisions
- 03 Practice Real work as weekly homework
- 04 Leverage Your team, your communication, your delegation
- 05 Hand-back Run your team, answer the board yourself
Private Executive AI Coaching applies this framework to how you actually operate. Not a checklist — weekly sessions that start with where your attention is stuck and build, week by week, toward a decision: what to test, defer, and kill. The levers we build:
- Judgment — reduce avoidable friction around important choices.
- Decision speed — clearer inputs, framing, and follow-through for recurring decisions.
- Communication — executive messages that say what you decided and what happens next.
- Delegation — clearer ownership, context, and escalation so the CEO isn’t the default path.
What changes, month by month
- Fluency — use the tools daily on your own work, without help.
- Judgment — know, for any decision on your desk, whether AI belongs in it.
- Practice — produce better work faster, verify it, and know when not to trust it.
- Leverage — run your part of the company with AI in it, without becoming its bottleneck.
What changes
- Fewer experiments — each tied to a decision you own.
- Initiatives you’ve stopped, with the reason on record.
- Decision rights your team can name without asking you.
- Able to answer the board yourself.
The six-month arc
Month 1
Baseline
Apply the deferral framework. Decide what to test, defer, and kill.
Months 2–4
Judgment & Practice
Ship changes tied to constraints you own, week by week. Stop the rest.
Months 5–6
Leverage & Hand-back
Hand off the system so your team owns it. Answer the board yourself.
Confidentiality
- The company may pay. Confidentiality runs to you, the executive — your own team doesn’t need to know.
- You share only minimum-necessary, non-sensitive context and sanitized examples.
- No production credentials or sensitive datasets enter the advisor’s tools.
- No case study, testimonial, or use of your name without written permission.
Book the Operating Leverage Session
If you’ve tried AI and been disappointed — or you’re accumulating experiments and suspect most of them won’t matter — the Operating Leverage Session is the place to start. The form asks a few short questions so the conversation is useful from the first minute.
Book the Operating Leverage Session — skip the pilot graveyard.One-on-one work with the executive — not implementation, team training, legal advice, or a guarantee of results.
The Operating Leverage Session is also a working-style fit. Business is a team sport — it’s more important that you find a great fit than that it be me, specifically. If I can unblock you with a recommendation, a referral, or a different approach, that’s a better outcome than a forced engagement.
Not ready to book? Take the AI•DOOM Assessment — the 7-minute Artificial Intelligence Diagnostic of Operational Maturity.