What to automate first

By this point in the module you have a review-reply prompt, a quote prompt, a voice file for social posts, four message templates, and a Google Business Profile routine. That is a lot of new habits for a person who also has to run a crew, and the honest truth is that most owners who try to start all five at once keep none of them.

So this lesson is about choosing. It gives you a simple way to score any task in your business, a clear picture of what AI does well today and what it does not, and a rule for keeping a human in the loop without turning the human into a bottleneck. It ends with the one thing to do this week.

Three questions that score any task

Automation just means a task happens with less of you in it. It might be AI drafting a reply, or it might be a plain rule in your phone that sends a text. Either way, the same three questions tell you whether a task is a good candidate:

• Is it repetitive? You do it the same way every time, with small variations. Review replies, yes. Negotiating a commercial contract, no.
• Is it low risk? If the draft is wrong, is it cheap to catch and fix before it costs anything? A bad social post draft costs you the ten seconds to delete it. A bad price on a signed quote costs you real money.
• Is it high volume? Does it happen many times a week? Automating something you do twice a year saves nothing and adds a tool to maintain.

Score each task from zero to three. Three means start there. Two means it is worth doing, with a human checking every output. One or zero means leave it alone, at least for now. The scorer above lets you tap through a typical list. The interesting part is that some "obvious" automations score badly. Bookkeeping is repetitive and frequent, but the cost of a wrong category shows up at tax time, which puts it in the "use real software with your bookkeeper" column, not the chat window column.

What AI handles well today

As of August 2026, the chat assistants most owners use are genuinely good at a specific family of tasks, and it is worth naming it so you aim at the right targets:

• First drafts of anything written. Replies, posts, emails, descriptions, quotes from your notes. This is the core skill and it is reliable.
• Reformatting. Notes into a table, a rambling voice memo into a clean list, a long email into three bullet points.
• Matching a voice you show it. Given examples, it holds a tone well.
• Summarizing what you give it. Paste a long thread, get the gist. The Documents and Data module covers the limits of this.
• Explaining things. A contract clause, an insurance term, a city permit form. As a starting point for a conversation with the actual expert.

Notice the shape. Every item is "text in, better text out, with you reading it." That is the sweet spot.

What it does not handle well

The failures are just as consistent, and they are the reason the "low risk" question exists:

• Facts it was not given. Prices, availability, dates, what the weather is, whether the event is this Saturday. It will produce a confident answer that is made up. The trust and limits lesson explains why.
• Judgment calls about people. Which applicant to hire, whether a customer is lying, whether to fire someone. It is not good at this, and in some cases there is real legal exposure.
• Money and legal decisions. Tax treatment, contract terms that bind you, insurance coverage questions. It can explain; it should not decide.
• Anything with no one reading the output. A fully unattended process that sends messages, posts content, or makes commitments in your name with nobody looking is where every AI horror story you have read comes from.

The order that works

If you scored your own list honestly, the same few things end up on top for almost every local trade, and there is a sensible order to start them in:

• Week one: the missed-call text. It scores three, it is one sentence, and many business phone apps can send it without any AI at all. If yours cannot, save it as a two-tap text. Everything else waits until this is done.
• Week two: review replies. Save the system prompt from lesson one, clear the backlog in one sitting, then reply every Friday.
• Week three: the estimate follow-up. Save the template, send it on day three after every quote.
• Month two: posts and the profile. Build the voice file, batch a month of social posts, and start the weekly profile post.
• Whenever the first four are habits: quotes. Highest payoff and highest risk, so it comes last, once you trust your own checking routine.

That is one new habit every week or two. Slow enough to stick, fast enough that within two months the boring writing in your business is mostly handled.

When to call a pro

Everything in this module runs inside a chat window with you reading the output. That covers a lot. There is a line past which it stops being a chat and starts being a system, and that is where it makes sense to bring in someone who does this for a living:

• Connecting tools together. When you want a form on your website to create a quote draft, or a finished job to trigger the thank-you and the review ask automatically, that is software wiring, not prompting.
• Anything that runs unattended. The moment nobody is reading the output before it goes out, the guardrails have to be built in code and tested, not written in a prompt and hoped for.
• Customer data. If the automation touches names, addresses, phone numbers, or payment details at any volume, you want someone thinking about where that data goes and who can see it.
• Your website and your profile as one system. Most Southern Utah businesses get found through Google first. If the site, the profile, and the follow-ups are not telling the same story, fixing that is worth more than any single automation.

A good pro will start by asking what you scored a three, not by selling you a platform. If someone leads with the platform, keep your scoring grid in your pocket and ask them which of your tasks it actually changes.

Try this yourself

Do this in ten minutes, today, before the module fades. Paste this into your AI:

I run [BUSINESS NAME], a [SERVICE] business in [CITY], Utah, with [NUMBER] people. Here are the tasks that eat my week:

[LIST 8 TO 12 TASKS, ONE PER LINE, e.g. replying to reviews, writing quotes, answering the same phone questions, chasing unpaid invoices, posting on Facebook, scheduling the crew, ordering materials]

For each task, score it 0 to 3 using these three questions, one point each: Is it repetitive (same way every time)? Is it low risk (a wrong draft is cheap to catch before it costs anything)? Is it high volume (many times a week)?

Give me a table sorted by score, high to low. For each task, one sentence on whether an AI chat assistant can help with it today, and where a human should check the output. Flag any task where AI should not be used at all and say why. Be blunt.

Read the table. Pick the top item. Set it up this week using the lesson that covers it. Do not pick two. When it is a habit, come back for the next one. That is the whole playbook: one repetitive, low risk, high volume task at a time, with you reading the output at the moment it is cheap to fix, until the boring parts of the business run themselves and you are back to doing the work you actually started the business for.