Which business functions should you automate first?

What is the right order for most businesses?

Lead scoring first, because it changes what you do with your time immediately and a wrong answer costs you one phone call. Then first-line support, then reconciliation and record hygiene. Outbound sending comes after those, because it is the one that can damage your reputation at volume.

The reasoning behind the order is blast radius. A mis-scored lead wastes an hour. A mis-sent outbound campaign burns a domain reputation you spend months rebuilding. Start where being wrong is survivable and work outwards as you learn how the system behaves.

Reconciliation is the sleeper. It is nobody’s favorite job, it is entirely rule-shaped, and it is where small errors quietly accumulate into a set of accounts nobody trusts. It also produces immediate, visible relief.

How do you know a function is ready to automate?

You can write down how you do it. If the instructions fit on a page and you would hand them to a new starter without worrying, it is ready. If you find yourself writing "it depends" more than twice, the function still lives in your head and automating it will encode the wrong thing.

A second test: has it happened enough times to have a pattern? Automating something you have done four times means guessing at the shape of the work. Automating something you have done four hundred times means encoding a shape you actually know.

Write the instructions first, whichever platform you use. The exercise usually reveals that two of the steps were unnecessary and one was wrong, which is worth more than the automation.

What should you automate last, or never?

Anything where the relationship is the product, anything you do fewer than a few times a month, and anything irreversible. Rare tasks are the worst return on effort: you will spend more time specifying them than you will ever save, and get them wrong because there are no examples.

Relationship work is the harder call, because it is often the work founders most want off their plate. The test is whether the other person would mind knowing it was automated. If a customer would feel differently about the message on learning software wrote it, that is information about what the message was for.

There is also a category worth protecting deliberately: the work that keeps you in contact with reality. Founders who automate all customer conversation stop hearing what is actually wrong with the product. Keep a channel open on purpose.

What is the most common ordering mistake?

Automating outbound first, because it feels like the growth lever. It is the function with the largest blast radius and the one most dependent on an offer that already works. Automating a message nobody wanted just means more people receive a message nobody wanted.

The second most common is automating the function the founder personally dislikes rather than the one that consumes the most time. Those are frequently different. Invoicing is annoying and takes two hours a month; support is fine and takes twelve. The disliked task is the one that gets attention, and the expensive one continues.

The third is automating something that has never been done manually. If nobody has run the process even once, there is no evidence of what good output looks like, and the automation encodes an assumption instead of a practice. Do it by hand five times first, badly, and the specification writes itself.

A simple ordering that avoids all three: whichever function consumes the most hours, has the clearest rules, and would cause the least damage if it went wrong for a day.

A practical first month

Week one, write down how you currently do the function that costs you the most evenings. Week two, automate it and watch it closely. Week three, tune the thresholds — what stopped that should not have, and what did not stop that should. Week four, do the next function.

On newc0 the four functions arrive together, but the same discipline applies: turn thresholds down at the start, watch what reaches the morning queue, and loosen them once you have evidence rather than hope. The agent workforce page sets out what each function covers.