Plain-English notes on setting up AI properly — written for business owners, not engineers. Every post answers one real question, honestly: no invented statistics, no ROI promises, and our biases disclosed when we have them.
A missed call is a customer already holding their phone. What the first message should actually say, the three-message sequence that books work without nagging, the five ways automated follow-up turns insufferable, and where the approval line honestly goes.
Nobody owns the model — everyone rents that. What you can own is the layer around it: your business context, your instructions, your output, your data. The four layers of any AI setup, five questions to ask before you buy, and the leaving test that tells you the truth in an afternoon.
The single highest-leverage upgrade to any AI setup is one page: a "Who We Are" file that tells every tool who you are, who you serve, how you talk, and your firm rules. The seven short sections, what to leave out, where to keep it, and why it gets smarter every time you correct it.
The honest version for a small brokerage: the four things AI reliably helps with today — fast lead follow-up, first-draft listing copy, routine questions, paperwork prep — and the four places it should stay out, from fair-housing advice to a client's private file.
From the people who install it for a living: the six jobs to keep away from AI — the expensive public mistake, the apology, the answer you can't check, the licensed call, the judgment about a person, the one-time task. Plus the gray zone, and a three-question test.
Before you hand an AI tool your customers' names, emails, and history, ask six plain questions — where it lives, whether it trains their models, who can see it, whether you can delete it. What a trustworthy answer sounds like, and where to keep the data close.
When ChatGPT, Claude, or Perplexity reads your site to answer a customer, it sees text — not your design. What it reads well, the four things it usually can't see, and a two-minute test to check your own pages.
The whole starter setup is three things, not thirty: one folder your AI works out of, one brief that teaches it your business, one automation that saves real time. What goes in each, and why small beats big.
Let AI draft; keep a person approving before anything reaches a customer or moves money. Why that trade beats full auto, where full auto is genuinely fine, and how to draw the line without slowing down.
Most first AI setups aren't a months-long engagement — they're one focused afternoon. How to tell which situation you're actually in, what an afternoon delivers, and when paying for help is the honest call.
Not a robot version of you — a written copy of how your business actually works, kept where your AI can read it every time. What goes in one, what it honestly can't do, and how to start this week.
Yes — and it's good at the calls you're most likely to miss: booking, hours, after-hours. What a voice agent does well, where it falls short, and how to set one up without pretending to be human.
AI writes generically because it's averaging everyone. Pull it toward you with three things: real samples of your writing, a short rules list, and feedback that sticks — an afternoon, not a course.
Customers now ask an assistant who to hire instead of Googling. The honest playbook for being the business it names — readable pages, direct answers, consistent facts — and why guarantees are a red flag.
Three verbs tell them apart: a chatbot answers, an automation repeats, an agent acts. When each is the right fit — and the approval rule that keeps an agent safe.
The app, the one folder, the one-page brief that teaches it your business, and a first automation you'll actually use — an afternoon of setup, step by step.
The biggest upgrade to any AI tool isn't a better prompt — it's standing context. The five things your AI should know, and why memory compounds week over week.
The plain-text file that helps AI assistants represent your business accurately — what it honestly does and doesn't do, and the rest of the AI-readable checklist.