JUL 2026 · 5 MIN
When not to automate
The most useful sentence in this business is: do not automate that.
By Daniel Barraza · dbar Labs
Boerne, Texas
Most of my work is about where automation helps. This one is about where it does not, because knowing the difference is what keeps a small project from becoming an expensive regret. Here is the honest version.
Do not automate low-volume work
If a task happens twice a month and takes ten minutes, automation will cost more to build and maintain than it will ever save. Automate what happens every day, not what happens occasionally. Frequency is the first filter.
Do not automate a process that changes every week
Automation loves stability. If your intake process, pricing, or scheduling changes constantly, any workflow built around it will break constantly. Get the process stable on paper first. Then automate the version that survived a few months of reality.
Do not automate judgment, empathy, or negotiation
Angry customers. Pricing conversations. Complaints. Anything where being wrong carries a real cost. AI can draft a reply, and a person should decide whether to send it. Keep humans on every conversation where someone needs to care about the answer.
Do not automate on top of messy data nobody owns
If customer information lives in three places and none of them are current, automation will not clean it up. It will spread the mess faster. Fix the source of truth first, even if the fix is a spreadsheet and a habit.
Do not automate to dodge a staffing or pricing problem
If the real issue is that your prices are too low or you are short a technician, software will not fix it. It will just help you lose money more efficiently. Be honest about which problem you actually have before buying a solution to a different one.
Do not automate what you have not watched a human do well
If nobody on your team can write the steps on one page, nobody can automate them. The best automations start as a clear manual process that already works.
A good audit will talk you out of things.
Rules of thumb
- Automate the frequent, the repetitive, and the boring.
- Keep people on relationships, exceptions, and anything regulated.
- If a step requires judgment, a person keeps it.
- Start with one workflow. Prove it. Then decide about the next one.
If you remember one thing
Automation is a multiplier. It multiplies whatever process you point it at, good or bad. Point it at a clear, stable, frequent process and it gives you hours back. Point it at chaos and it gives you faster chaos.
How this shows up in my work
Every AI Opportunity Audit ends with a written plan, and that plan always includes a "what not to automate" section with reasons. Sometimes the honest recommendation is "do nothing yet," or "fix this by hand first." You deserve to hear that before you pay for a build, not after.
Why say all this out loud? Because a vendor who will tell you no is easier to trust when they tell you yes. The businesses that get the most from automation are the ones that said no to most of it.