It looked fine. It was quietly broken. I found it by measuring.
That sentence describes every engagement I have run. Most of what holds a small business back online is not missing. It is there, it looks like it works, and nobody has checked. The first job is always to take a reading.
Every rule on this page came from one job. For ten years I was the business manager at boat dealers, the person in the finance office whose job is the real number. Each section below starts with the finance-office habit it grew out of.
Here is what that looked like with real clients this year.
Four accounts. Four quiet failures.
Where it comes from: in the finance office, a deal on paper and a funded deal are two different things until someone checks. So I check.
A two-store MasterCraft dealer whose ads had never run. Action Watersports’ Google Ads account showed zero lifetime spend. Every campaign setting checked out. The cause was a hidden $0.00 daily spending limit that Google had applied nine months before I arrived. Once it was found and released, the account served its first impressions within days.
A water-treatment company blocking its own name. A broad negative keyword in Montgomery Water Softening’s shared list was stopping ads from showing for the company’s own legal name, on all three campaigns. Blocked searches never appear in the search-terms report, so nothing flagged it. It was found and removed.
A contact form that said “sent” and sent nothing. During Long Boy Rentals’ rebuild, the form showed a success message while the server delivered no mail at all. It was caught before launch and proven end to end, from the form to the owner’s inbox. A success message is not delivery.
A charter selling trips on boats that can no longer sail. Reel Adventure Charters’ third-party booking listings were still taking deposits for two boats retired from the fleet. Finding it meant reading every channel where the business can be booked, not only the website.
How every change ships.
A deal isn’t done until it funds. A change isn’t done until the guardrails pass the next morning.
One experiment at a time. Each change starts with a written theory of what it should do and a condition, decided in advance, that means it gets reversed. The next morning, the guardrails get checked: spend, impressions, tracking. If something broke, it comes back out the same day.
Waste gets cut immediately and logged. Experiments wait their turn, so when a number moves, I can tell you why, or tell you honestly that I can’t.
Every report includes the numbers that went the wrong way and what I’m testing next. No “great month” without the data behind it. A business manager reconciles the deals that fell apart along with the ones that funded; the weekly report works the same way.
Automation that fails loudly is safe.
Where it comes from: a deal jacket missing a signature has to stop the deal, not slide through to the lender. Automation should stop the same way.
Automation that fails quietly destroys data. The jobs I build are designed to stop and say so when they can’t read their inputs, rather than guess. A job that cannot tell “nothing new” from “nothing readable” will eventually delete the wrong thing.
Measurement runs underneath all of it. When a scheduled job misses a day, the gap is written down, not papered over. A silent skip and a clean run look identical unless something is watching.
Everything is yours.
In a dealership, the deal files belong to the store, not to the business manager. Your accounts work the same way. Domains, ad accounts, Business Profiles, content and code are registered to you from day one, and the AI runs self-hosted, on my hardware or on a Mac mini in your office.