Playbook · Tips
Things I keep saying.
Every one of these came out of a real conversation with an owner who was about to spend money on the wrong thing. Opinionated on purpose.
- 01
Never let an AI agent quote a price it cannot cite.
An agent that invents a number will do it confidently and you will find out from an angry customer.
Give the agent an explicit price list and one hard rule: if the number is not in this list, say "I will have someone confirm that for you" and create a task. Everything the agent says about money should be traceable to a document you wrote. This single rule prevents most of the ways an AI front desk can cost you real money.
AI - 02
Automate the boring middle, not the ends.
The first touch and the final decision should stay human. Everything between them is fair game.
The middle of a process — routing, data entry, reminders, status updates, moving information between systems — is where automation pays and where mistakes are cheap to catch. The first contact with a customer and the moment money or judgement is involved are where automation costs you trust. Draw that line before you build anything.
Automation - 03
An automation without an alarm is a liability.
A silent job that stops running is worse than no job at all, because you stop checking.
Every scheduled task needs two things: a notification when it fails, and a heartbeat that tells you it ran at all. The dangerous failure is not the loud one — it is the sync that quietly stopped three weeks ago while you assumed the data was fresh.
Automation - 04
If two people spell it differently, it is not one field.
"Chula Vista", "chula vista" and "CV" are three customers to a computer and one to you.
The moment you notice free-text where a dropdown should be, you have found a future reporting problem. Fix it early: it costs ten minutes now and a data migration later. The same goes for phone numbers, dates typed by hand, and anywhere someone can put "N/A".
Data - 05
Your spreadsheet already contains a database. Find it first.
Before anyone builds you software, work out what real things your columns are secretly describing.
A sheet with 40 columns is usually four or five real objects mashed together — customers, jobs, invoices, technicians. Separating them on paper takes an afternoon and makes every quote you receive afterwards cheaper and more accurate, because the developer is no longer guessing.
Data - 06
Ask what happens to your data when you leave.
Ask before you sign, not when you are already trying to go.
Specifically: can you export everything, in a format something else can read, without calling support? If the answer is vague, that vagueness is the product. Platform lock-in is rarely a technical problem — it is a pricing strategy, and you are looking at it.
Buying software - 07
Custom is not the opposite of off-the-shelf.
Most good systems are eighty percent boring tools you already pay for and twenty percent the part nobody sells.
You almost never need custom accounting, custom email or a custom calendar. You often need a custom layer that makes the five things you already use behave like one. Anyone who wants to rebuild your whole stack from scratch is selling hours, not judgement.
Buying software - 08
A per-seat price is a tax on growing.
Do the maths at the headcount you want, not the one you have.
Take whatever you pay now and multiply it by three, because that is what it costs the day the business works. If that number is uncomfortable, you have learned something useful about the tool before you have built your whole operation on top of it.
Buying software - 09
Your homepage has eight seconds. Spend them on what you do.
Not your values, not your journey, not a stock photo of a handshake.
A stranger arriving from a search wants three things: what this is, whether it is for them, and what to do next. Everything else is for the people who already decided. Put the first sentence of your homepage in front of someone who has never heard of you and ask them what you sell. Their answer is your real headline.
Websites - 10
Speed is a feature your customers feel and never mention.
Nobody says "your site was fast". They just do not leave.
Most small business sites are slow for two reasons: uncompressed images, and a page builder loading forty things nobody asked for. Fixing those two usually halves load time without touching the design. Test on a phone on cellular data, not your laptop on office wifi.
Websites - 11
Give the model your documents, not your adjectives.
"Be professional and friendly" does almost nothing. Your actual policy document does almost everything.
Tone instructions are the least useful part of a prompt. What changes the output is grounding — real prices, real hours, real policies, real examples of how you answer. If you find yourself writing more personality instructions, you probably need more source material instead.
AI - 12
Make it ask you questions before it writes anything.
Adding "ask me one question at a time until you understand" to a prompt improves the output more than any other single change.
A model given a vague request will produce a confident, generic answer. A model told to interview you first will surface the specifics that make the output usable — and often the questions themselves are the valuable part, because they are the things you had not thought about.
AI - 13
Deposits fix no-shows better than reminders do.
A reminder asks politely. A deposit changes the decision.
Reminder sequences reduce no-shows somewhat. Taking any deposit at all — even a small one — reduces them dramatically, because the customer has now made a decision rather than an intention. If you do both, take the deposit in the same conversation where they said yes, not in a follow-up email.
Automation - 14
Track the number you would actually change your week for.
If a metric moving would not change what you do on Monday, stop measuring it.
Most dashboards are built to look complete rather than to be used. Pick the two or three numbers where a bad week means you do something differently — and put those somewhere you cannot avoid seeing. Everything else belongs in a report you run when you have a question.
Data - 15
Own the domain, the repo and the accounts. Always.
Whoever controls those three things controls whether you can leave.
Register the domain in your own name, on your own card. Make sure the code lives somewhere you can log into. Be the owner on the hosting, the analytics and the payment processor, and add your developer as a user. A good developer will set it up this way without being asked. Anyone reluctant is telling you something.
Buying software - 16
Write down what happens after someone fills in the form.
Most contact forms on small business sites go to an inbox nobody has opened this week.
Follow your own form. Submit it. Time how long it takes for a human to respond. Then decide whether that is the first impression you want to be making. Speed to first reply is the single highest-leverage number in most small businesses, and almost nobody measures it.
Websites
Nothing under that topic yet.
Next
Prompts you can actually use today.
Sixteen of them, and they rewrite themselves around your business once you type your details in at the top.