AI for Operators
What to Automate First (and What to Never Automate) in a Small Business
Most founders automate in the wrong order. They start with whatever is shiny, or whatever a tool's marketing told them to, and they end up with a pile of half-built systems solving problems they never had. I have built three businesses and helped operators build many more, and the order you automate in matters as much as what you automate. Here is the order I actually use, and the line where I stop.
First, the work that is frequent, rules-based, and low-stakes
The first things to hand off share three traits. You do them often, they follow clear rules, and getting one wrong is cheap to fix. Scheduling. Data entry. Sorting and routing inbound messages. Pulling the same report every week. These are boring on purpose. They are also where you are quietly losing the most hours, and the risk of letting a machine handle them is close to zero. Win here first, build the habit, then move up the ladder.
Next, the work that is valuable but needs a safety net
Once the boring wins are in place, move to the higher-value tasks that still tolerate a human checkpoint. First drafts of writing. Research and synthesis. The scaffolding of investor or customer follow-up. These give back more than hours; they give back creative energy. The rule here is the one I wrote about in AI agents for founders: let the machine do the work, but keep a human signature on anything that leaves your hands.
The test I use before automating anything
Before I automate a task, I ask two questions. Is it frequent enough that automating it is worth the setup? Is a mistake cheap enough to tolerate? Frequency tells you whether it is worth doing. Reversibility tells you whether it is safe to do. A task that happens rarely is usually not worth the effort. A task where errors are expensive should keep a human in the loop, or stay manual. Most of the regret I see comes from skipping these two questions.
What to never automate
Some work is not a bottleneck to remove; it is the job itself. The real investor conversation. The decision to hire someone or let them go. The apology when something goes wrong. The strategic call about where the business goes next. Moving money. These carry your judgment and your reputation, and handing them to a system is how trust gets quietly destroyed. Automating the prep around these is fine. Automating the moment itself is a mistake.
The error that wastes the most time
The most common mistake is automating something that was never your real constraint. Founders love to build a clever system for a task that, looked at honestly, they should simply stop doing. Before you automate, find the actual bottleneck, the thing that would free the most if it disappeared. Often the answer is not a tool at all. It is a decision to drop the task entirely.
What you are really buying
Automation done well is not about doing more things. It is about buying back your attention for the small number of things only you can do. That is the same thread running through everything I write about AI, which I traced in the evolution of AI: the tools change the cost of the work, never your responsibility for the outcome. Deciding what to hand off and what to protect is most of what I help operators get right in how I work.