The Small Business AI Office Playbook

The Small Business AI Office Playbook · Chapter 1

Map your day before you add AI

Most owners start with the tool. They sign up for a chatbot, connect it to something, and a week later it’s texting customers things nobody approved. The fix is boring: before you automate anything, find out what actually comes into your business and who has to touch it.

This chapter is a one-week exercise. You don’t buy anything and you don’t install anything. You write things down. At the end you’ll have a list of what to automate first, what needs your eyes, and what stays human.

Day 1: list every way work comes in

Get a notebook or a spreadsheet with one row per channel. Most small businesses have more than they think:

• Phone calls (answered)
• Phone calls (missed)
• Voicemails
• Text messages
• WhatsApp
• Email (main inbox)
• Email (the other inbox you forgot about)
• Website contact form
• Facebook and Instagram messages
• Google Business messages and reviews
• Referrals from a partner, supplier, or past client
• Walk-ins or job-site conversations

Leave room to add more. Somebody will call your personal cell by Wednesday and you’ll realize that’s a channel too.

Days 1 to 5: count them

For five working days, make a tally mark every time something comes in. Next to each mark, write one or two words about what it was: “new lead,” “schedule change,” “invoice question,” “supplier,” “spam.”

You’re after three numbers per channel:

1. How many came in
2. How many were missed or answered late (more than a few hours)
3. How many needed you, the owner, specifically

That third number is the one that matters. Mark it with a star. A message needs you when it involves a decision only you can make: a price, a date you’re committing to, a problem with a customer, or money leaving the account.

Be honest. Plenty of owners find that most of their starred items don’t really need them. They just never gave anyone else the rules.

Day 5: read the tally

Look at the sheet and answer:

• Which channel has the most volume?
• Which channel has the most misses?
• Which starred items could someone else handle if they had a clear rule?

Here’s what an EXAMPLE week might look like for a small contractor. Your numbers will be different.

In this example, the biggest leak is obvious: 17 missed calls, and 6 of those were people who needed the owner. That’s where you start.

Day 6: set your three approval gates

Before any AI touches your business, write down the three kinds of things it is never allowed to do on its own. These are your approval gates. Anything that hits a gate stops and waits for a human to say yes.

Gate 1: anything a customer reads. Texts, emails, estimates, social replies, review responses. AI can draft them. A person approves them before they go out. You can loosen this later for low-risk messages like “Got your message, we’ll call you back today,” but only after you’ve read a few weeks of drafts and trust them.

Gate 2: anything that spends money. Orders, subscriptions, payments, refunds, ad spend. AI can prepare the order or flag the invoice. A person clicks the button.

Gate 3: anything that commits a date or a price. “We can start Monday.” “That’ll be around fifteen thousand.” These sentences are promises. Once a customer hears them, you own them. AI never says a date or a number to a customer without your sign-off.

Write these three gates on one page and give a copy to anyone who sets up or runs your tools. If a tool can’t be configured to respect them, don’t use it.

Day 7: pick your order

Automate in this order. Each step is small, and each one builds on the one before it.

1. Missed-call text-back. When a call goes unanswered, the caller gets a text within a minute: who you are, that you saw the call, and when you’ll get back to them. This passes Gate 1 because you write the message once, approve it, and it never changes. It usually plugs the biggest leak for the least effort.

2. Call summaries. Calls get recorded (where the law allows it and with the notice your state requires) and turned into a short summary: who called, what they want, what they were promised, what happens next. You stop relying on memory and sticky notes.

3. Follow-up drafts. Using the summaries, AI drafts the next message for each open lead or job, such as the follow-up on an estimate you sent last week. Drafts sit in a queue. You or your office person review each one before it goes out. Gate 1 stays closed.

4. Daily digest. Once the first three are running, one message arrives at a set time each evening with new leads, missed calls with their status, drafts waiting for approval, and anything that hit a gate. It replaces checking six apps.

Don’t skip ahead. A daily digest built on messy inputs just gives you a neat summary of a mess.

What AI should not do

Being honest about limits is part of the job.

• It shouldn’t price your work. It can do arithmetic on your quantities and your rates. It shouldn’t come up with rates. Anything it knows about prices is general and probably not from your market.
• It shouldn’t promise dates. It doesn’t know your crew is short this week or that the inspector is backed up.
• It shouldn’t handle an angry customer alone. A draft is fine. The call is yours.
• It shouldn’t touch money without a person. No auto-pay, no auto-refund, no auto-order.
• It shouldn’t be the only record. Summaries can be wrong. Keep the recording or the original message for anything that matters, and correct the summary when you catch a mistake.
• It shouldn’t pretend to be you. If a customer asks whether they’re talking to a person, the answer is the truth.

Many owners find that after this week they need less AI than they expected. Often the biggest gain comes from the missed-call text and a clear rule for who answers what. That’s a good result. You’re buying back hours, not collecting tools.

Your worksheet

My three gates (write them in your words):

1. A customer reads it:
2. It spends money:
3. It commits a date or price:

First thing I’ll automate:

In Chapter 2 we set up the missed-call text-back step by step, including the exact message and how to test it before a real customer sees it.