Start with a list of steps, not a list of tools
Picture a small service company that gets excited about automation. In a few weeks, it sets up automatic texts for every new lead, automatic invoices when a job is marked done, an automatic thank-you after every payment, and an AI chat that answers questions on the website. Then a customer gets an invoice for a job that was paused for a warranty question. A long-time client gets a cheerful "welcome" text. A Spanish-speaking customer gets an English reply that misreads her question. The owner spends more time apologizing than the automations ever saved.
That scene is an original illustration, not a TwoQM client and not a measured result. Nothing in it was a bad tool. Each automation did exactly what it was told. The trouble was that nobody decided, up front, which steps should never run without a person.
This article is about drawing that line. It applies to any industry: contractors, clinics' front offices, restaurants, property managers, nonprofits, retailers. If you are already weighing AI tools, the article on budgeting for AI as an operating line item covers the cost side; this one covers the judgment side.
Before choosing software, write down the steps in one workflow you want to improve, such as quote to job to invoice to payment. For each step, write who does it today and what they check before moving on. That last part matters most. The checks people do without thinking are exactly what an automation skips.
Four questions for every step
Ask these about each step on your list:
Can it be undone easily? Sending a reminder can be followed by a correction. Charging a card, cancelling a booking, or deleting a record is harder to take back.
Does it touch money, safety, legal terms, or someone's eligibility? Those steps usually need a named person to approve them.
Does it need judgment about a relationship? A long-time client, an upset customer, or a sensitive situation deserves a person, not a template.
Are the inputs reliable? If the step depends on a status field that people forget to update, automating it spreads the mistake faster.
A step that is easy to undo, low-stakes, routine, and fed by reliable data is a good candidate. A step that fails any one of these questions usually should keep a person in the loop.
Good candidates for automation
These tend to be safe places to start, once the data behind them is clean:
Reminders built from approved templates, such as appointment or document reminders.
Moving data between systems so staff do not retype it.
Creating a task or alert when something is overdue.
Summaries and reports pulled from records you already keep.
First drafts that a person reviews before they are sent.
Steps to keep with a person
These usually stay manual, or become "draft, then approve":
Approving refunds, credits, discounts, or write-offs.
Final pricing and quotes. Link customers to your published pricing instead of letting a tool restate it.
Decisions about eligibility, priority, or who gets a limited spot.
Messages to an upset customer or about a complaint.
Anything that cancels, deletes, or charges.
Translations that will be sent as final, until a bilingual staff member has reviewed them.
Draft-then-approve is the middle ground
Most steps do not have to be all manual or all automatic. A system can prepare the work and stop at a clear approval point: the invoice is drafted, the message is written, the record is updated in a pending state. A person reviews and presses the button. The article on AI-drafted follow-ups with human send approval walks through that pattern for messages.
Record who approved each step and when. That log is what lets you trust the system later, and loosen it when it has earned trust.
Write your boundaries down
Keep a one-page list: what runs on its own, what is drafted for approval, and what stays fully manual, with a named owner for each. Review it when you add a new tool or a new service. When something goes wrong, check the list first: was this step on the right side of the line?
Where TwoQM fits
Business automation handles the routine steps on your "runs on its own" list, with approval points built in where you want them. AI automation can draft, summarize, and sort for staff review. Connected systems move data between the tools you already use, and custom software can add approval screens where off-the-shelf tools have none. What fits your setup is confirmed during scoping, and part of scoping is agreeing on what not to automate.
Where AI helps, and where it stops
AI can draft messages, notes, and summaries, sort incoming requests for review, and point out records that look incomplete.
AI does not approve money, make decisions about customers or eligibility, send final messages, or delete anything on its own. A person on your team does.
Where to start this week
Pick one workflow and list its steps. Mark each one as runs on its own, draft then approve, or stays manual, and name who owns it. The steps you are unsure about are the ones to talk through before adding any new tool.
TwoQM builds automation, AI tools, connected systems, and custom software for small businesses in the Rio Grande Valley and beyond. If you want help drawing the line in your own stack, TwoQM can scope it with you. Pricing explains how business apps, systems, and custom software are scoped.
Talk Through Your Automation Boundaries · Explore TwoQM Business Automation