Put AI inside the process, even if customers never see it

You can use AI in your service business without changing how you talk to customers. I'd start with an internal task, like organizing business notes into a checklist, and keep a person responsible for checking the work. Your calls and conversations can stay personal while you test whether AI makes the preparation easier.

By Eric Yim, Protocol Text

Customers don't have to interact with your AI

I think if you're not incorporating AI in some form into your business, it's going to be really hard to compete with new businesses coming out. That's my view of where things are going, not a prediction that every business needs the same tools.

And that's not saying that your customer facing stuff has to be AI driven, but some of your processes.

You might still answer the phone yourself, ask follow-up questions, and personally confirm an appointment. Behind that conversation, AI could help organize an internal handoff or draft a checklist from your existing instructions. The customer doesn't need to use a chatbot for that work to be useful.

The decision starts with where your time goes. If you repeatedly sort through scattered notes before someone can act, that's a task worth examining. If the work already takes very little effort, adding another tool may just create another step.

Choose a process problem before choosing a tool

Think about a cleaning business doing all its scheduling by pen and paper. That's an illustrative example, but it shows the competitive issue I mean. Another business with tighter processes may get to the lead faster and schedule people faster, leaving more time for other work.

That doesn't automatically make scheduling an AI problem. A shared calendar or a booking system may solve the actual difficulty. I'd use AI where interpreting or organizing information is the part that takes effort, rather than assuming every process needs it.

For a first test, choose something repeated, easy to inspect, and safe to redo. Turning non-sensitive operating notes into a draft checklist fits that description better than letting a tool make commitments to customers. You need a clear source to compare against and someone who knows what a correct result looks like.

A practical exercise: turn notes into a usable checklist

Here's a hypothetical example, not completed client work. Suppose a cleaning business owner has these internal notes: confirm the requested service, check the service area, ask for preferred dates, and don't promise a date until availability is checked.

The goal is a checklist for the person handling an inquiry. It isn't a new customer message or permission for AI to book anything.

Give it the source and a clear boundary

A starting prompt could be: Turn these notes into a draft inquiry checklist. Use only the supplied instructions. Keep requirements separate from suggestions. Flag anything unclear as a question for me. Don't invent policies or booking rules.

Then supply the notes without customer names or other identifying details. Asking for boundaries helps define the task, but it doesn't guarantee the tool will follow them.

Check meaning, not just wording

Compare the draft against each original note. Does it preserve the requirement to check availability before promising a date? Has it quietly added a deposit requirement? Did it leave out the service-area check?

A neat list can still be wrong. Correct additions, restore omissions, and resolve unclear instructions yourself. Only then should someone use the checklist. For this exercise, the useful output is a reviewed internal document, not an unchecked answer sent to a customer.

Count the checking time before you expand

I think AI is super overhyped, but I also think it can amplify your skills when you bring critical thinking and discernment to the outputs. Discernment here means knowing what to accept, what to question, and what to reject.

To judge a test, record how long the task takes without AI. Then count the whole AI-assisted process: preparing the notes, writing the request, checking the draft, and correcting it. A fast first draft doesn't tell you much if fixing it takes longer than doing the task yourself.

Try it on several ordinary examples, including one with incomplete notes. Look for recurring mistakes and whether another person can use the result without extra explanation. Don't expand the tool's role just because one clean example worked.

If it saves effort while preserving the important instructions, keep that narrow use. If review keeps uncovering invented rules, tighten the task or stop using AI for it. The person who understands the work still needs to own the decision.

Frequently asked questions

How can I keep customer information private while testing AI?

Start with made-up examples or internal instructions that contain no customer details. Remove names, addresses, contact information, access codes, and identifying context. Before using real business data, review the tool's retention, training, and access settings. If you're unsure what happens to the information, don't upload it.

What should I do if my business notes contradict each other?

Have AI flag the conflict rather than choose a rule. Then ask the person responsible for that process to resolve it and update the source notes. Otherwise, a polished checklist can hide a decision your business hasn't actually made.

Who should review AI work if I don't have an office manager?

For a first test, I'd have the owner or the person who normally does the task review it. Keep the scope small enough that checking fits into their work. If nobody has time or enough knowledge to verify the result, choose a simpler task before using it operationally.

Should I buy an AI subscription before I know what I'll use it for?

I'd define the task and the review standard first. Then compare tools against those needs, including privacy, cost, and the effort of moving information between systems. Check whether software you already use offers a suitable option, but evaluate its settings and output just as carefully.