Do not automate a broken process. You will just break it faster
Most businesses automate the wrong thing first and buy themselves a faster mess. Here is how to tell which processes are ready, which four usually pay off first, and what to leave alone.
By Yannis Spanenburg · · 6 min read
Every week a business owner asks us about AI agents. Almost every week, the honest answer is that their first automation should be something far more boring, and that the agent conversation is three steps early.
This is not a lecture about hype. It is about sequence. Automation multiplies whatever process you point it at. Point it at a good one and you get leverage. Point it at a bad one and you get the same bad outcome, more often, with less chance of anyone noticing.
What makes a process ready to automate
A process is ready when four things are true. It happens often enough to be worth the build. It follows the same steps every time. Its inputs are predictable. And you can state plainly what a correct result looks like. Miss any one of those and you are automating a guess.
The test that costs nothing
Write the process down, step by step, as if handing it to someone starting on Monday. If you can do that in one sitting, it is a process and it can probably be automated. If you cannot, it is not a process. It is a habit living in one person's head, and automating it will simply encode whatever that person happened to do last Tuesday.
Most automation projects that fail were never really automation projects. They were undocumented processes that got expensive.
What to automate first
Speed of first response
A new enquiry that waits until tomorrow morning is competing against whoever answered within the hour. Automatic acknowledgement, routing to the right person, and a booking link is usually the highest return per hour of build in the entire business. It is also the least glamorous.
Intake and qualification
Every business asks new customers the same opening questions. Collecting those answers before the first conversation rather than during it shortens your sales cycle and filters the people who were never going to buy. This is where language models genuinely earn their place, turning messy free text into structured fields.
Reporting nobody enjoys building
If someone on your team rebuilds the same report every month by copying numbers between tabs, that is pure recoverable time. It is also low risk: if the automation breaks, a report is late rather than a customer being harmed.
Follow-up that depends on memory
Quotes that never got chased, carts left behind, customers who bought once and were never contacted again. Nobody decided to drop these. They fell through because following up relies on someone remembering, and people are unreliable at remembering by design.
What to leave alone
Leave anything that is your actual differentiation. If customers choose you because of how you handle a difficult conversation, automating that conversation removes the reason they chose you. Leave anything where judgment meets consequence, such as pricing exceptions, credit decisions or complaints that have already escalated. And leave anything that changes shape every few weeks, because you will spend more time maintaining the automation than doing the work.
The order that works
Document it. Then simplify it, because roughly half of most processes exists to solve a problem that no longer exists. Then automate what survives. Then measure whether anything actually improved. Skipping straight to step three is the standard failure, and it is expensive because it feels like progress the entire time.
Automation is a multiplier. Multiply a mess and you get a bigger mess, delivered on schedule.
The five minute sort
List the five things your team does most often. Next to each, write how long it takes and how much judgment it needs. Automate the one that is frequent and low judgment, and leave the one that is rare and high judgment exactly where it is. That single sort answers most of the question, and it does not require anyone to know what an agent is.
Frequently asked questions
What should a small business automate first?
Usually the speed of first response to a new enquiry: automatic acknowledgement, routing to the right person, and a way to book time. It is frequent, low risk, requires little judgment, and it competes directly against whichever competitor replied faster than you did.
How do I know if a process is ready to automate?
Write it down step by step as if you were handing it to a new hire. If you can do that in one sitting, it is a real process and it can likely be automated. If you cannot, it is a habit that lives in someone's head, and automating it will only encode their most recent improvisation.
What is the difference between automation and AI?
Automation follows rules you define, so the same input always produces the same output. AI handles the parts where the input is messy or the rules cannot be written down in advance, such as reading unstructured text. Most businesses need far more of the first than the second.
What should I never automate?
Anything that is the reason customers chose you, anything where judgment meets real consequence such as pricing exceptions or escalated complaints, and anything that changes shape every few weeks. In the last case the maintenance will cost more than doing the work manually.
Will automation replace my team?
In small businesses it usually moves people off repetitive work rather than removing them. The processes worth automating first are the ones nobody wanted to own anyway: chasing, copying, reformatting and remembering.