Operational excellence in my world is not a slogan. It is whether gross profit holds when a site’s mix shifts, when overtime spikes, or when a client starts shopping the contract. I own P&L in multi-site operations across union and non-union labor. I have taken a founder-led company to professional scale. I have held 99 percent of recurring revenue on a book I also helped originate. That is the lens I use for AI. Not “AI can help teams.” Whether decision-making gets faster on the controls that already run the business.

I use Claude, Grok, and custom automation. They get four jobs. Everything else is noise.

Job one: show the leak before payroll closes.
Labor cost is the business. I watch overtime, direct-labor percentage, and average wage rate by location, across CBA-governed and non-union crews in the same company. Monthly reviews are too late. The first job for AI is exception handling against that KPI set: which sites broke pattern, in language a regional manager can use this week. If the model cannot point at a location, a rate, and an owner, it is not operational. It is a summary.

Job two: hold the standard when I am not in the building.
I run a KPI system on inspection scores, training compliance, safety, and turnover. That scorecard already exists. AI does not invent it. It shortens the lag between a red cell and a coaching conversation. Multi-site operations fail when the standard lives in one director’s head. The tool’s job is to make the same inspection and turnover conversation available in every region without waiting for me to write it again.

Job three: make the company transferable.
At Westcoast Gate & Entry Systems I owned the full P&L, cut lead times 60 percent, took employee turnover to zero, raised client satisfaction 35 percent, and moved a founder-managed business into a professionally scaled operation, including multi-state acquisition integration. The constraint was never vision. It was turning practice into something a new state could run. First drafts of SOPs, manager scripts, integration checklists, and client updates used to consume the calendar. Now the first draft is cheap. My time goes to what we will enforce.

Job four: protect the commercial engine.
Business development in contract services is retention and add-on work, not a funnel diagram. At GMI I won UCLA Athletics — $4.9 million in recurring revenue, plus $1.5 million to $2 million in additional services — then ran a $25 million to $35 million P&L across 72 sites and 300 to 550 employees, with turnover down 40 percent and margins up 50 percent. Recurring revenue held at 99 percent. AI does not win that account. It does accelerate the QBR, the scope clarification, the internal brief, and the client memo that used to sit until Friday. That is the only “content” use I care about: commercial writing tied to client retention and decision speed.

Three things I will not automate.

Labor relations is the first. Mixed union and non-union workforces are a legal and cultural system, not a prompt. Safety is the second. A confident sentence is not a safe building. Pricing discipline versus a retention threat is the third. Those are judgments with board-level tails. AI can prepare the file. It cannot take the seat.

I also will not let a model invent a number I cannot trace to a site. Bad data plus a fluent paragraph is how executives embarrass a company. If the source is late or sloppy, the answer is wrong no matter how polished it sounds.

Boards evaluating AI spend can use the same test I use in operations. Name the KPI. Name the owner. Name the decision that should happen sooner. If a project cannot do that in one sentence, it is branding. Branding has a place. It is not operational excellence.

I am not building an “AI company.” I am running a labor-intensive service platform — now a 13-state facilities footprint at Platinum, after founder scale-up and a $25 million to $35 million regional P&L — and I am using AI to run that operating system with less lag. Fewer surprises in gross profit. Faster coaching. Cleaner handoffs. A business that can grow without the national director in every thread.

That is AI-driven efficiency I will sign my name to.