Does automation mean AI?

Usually not. Most of the work is plumbing: reading a file, doing arithmetic, producing an output on a schedule. AI is worth adding where a step involves reading or writing language, and it is worth leaving out where the answer needs to be exactly right every time.

What happens when the process changes?

The automation has to change with it, and that should be a small job rather than a rebuild. It is worth asking how a change gets made before anything is built, because the answer varies a lot.

Can I automate something that lives in my accounting software?

Often, and it depends on whether that software will let anything else talk to it. That is the first thing to check, and it is a five-minute question rather than a project.

What if it gets something wrong?

Then you want to find out from a draft rather than from a customer. Anything that produces something a person sends should produce a draft on the first run, and stay that way until it has earned otherwise.

Which part of my week should I automate first?

Not the biggest job. The one that is repeated, boring, and always done exactly the same way, because that is the one a machine can do without being wrong.

Most businesses have somewhere between five and twenty jobs that get rebuilt from scratch on a schedule. A report, a calendar, a price sheet, a list copied from one system into another. None of them are anybody's actual job and all of them take an evening a month from somebody.

The first one to automate is not the one that takes longest. It is the one that scores highest on the four things below.

What makes a job worth automating

Run every candidate against these four. A job that fails any of them is usually the wrong first project, however annoying it is.

It repeats on a schedule Weekly, monthly, every time an order comes in. A job you do twice a year is not worth the setup, however tedious it is.
It is done the same way every time If there is judgment in the middle, the automation can do the parts either side and hand you the decision. What it cannot do is guess.
The inputs already exist somewhere A schedule in a file, records in a system, a folder of photographs. If somebody has to type the inputs in first, you have moved the work rather than removed it.
Getting it wrong is visible, not expensive Start where a mistake is obvious and cheap. A wall calendar, a report, a draft email. Not payroll, not invoicing, not anything that touches a bank account on its first outing.

The arithmetic that decides it

Take the job. Write down how long it takes and how often it happens. Multiply it out for a year. That is the number the automation is competing with, and it is usually larger than people guess because the monthly ones hide.

Then do it again for every other job on the list. The point is rarely the first one on its own. It is what the whole list adds up to, which is usually more of the year than anybody expected to be spending on work nobody was hired to do.

What to do before you buy anything

Write the job down as steps, in the order somebody actually does them, including the part where they check something. Most automation projects that go wrong were scoped from a description of the job rather than from the job.

The writing-down is also the test. If you cannot get it onto a page as a list of steps, there is judgment in there that has not been made explicit yet, and that is the thing to work out first.

What it should feel like afterwards

Nobody should have to remember to run it. If the automation needs somebody to press a button on the first of the month, you have swapped a two-hour job for a two-minute job plus a reminder, which is better but not what you paid for.

The other half is that the outputs come from one place. When a schedule changes in the source, everything downstream changes with it, so nothing can quietly disagree with anything else. That is the property that makes the saving stick rather than decay.

The questions we get asked.

Does automation mean AI?: Usually not. Most of the work is plumbing: reading a file, doing arithmetic, producing an output on a schedule. AI is worth adding where a step involves reading or writing language, and it is worth leaving out where the answer needs to be exactly right every time.
What happens when the process changes?: The automation has to change with it, and that should be a small job rather than a rebuild. It is worth asking how a change gets made before anything is built, because the answer varies a lot.
Can I automate something that lives in my accounting software?: Often, and it depends on whether that software will let anything else talk to it. That is the first thing to check, and it is a five-minute question rather than a project.
What if it gets something wrong?: Then you want to find out from a draft rather than from a customer. Anything that produces something a person sends should produce a draft on the first run, and stay that way until it has earned otherwise.

Want help finding yours?

Tell us what your business does in a week and which part of it is done by hand. That is the whole conversation, and it is the same one every one of these starts with.

Hello@GeekThangs.com Start a Project

Based in St. Clair Shores, working across Metro Detroit.