1 rushed automation burnt through $2,000
You might not believe it, but business owners have a tendency to jump straight to fixing something before we've fully figured out what's wrong.
It’s true.
Now AI has given us a way to make that mistake even faster.
We can spend money automating a process before anyone has stopped to ask whether it's a good process to begin with.
You've probably heard of Elon Musk's five-step algorithm for improving processes:
- Question the requirements.
- Delete unnecessary steps.
- Simplify and optimize.
- Speed up the process.
- Automate.
Notice where automation falls on that list.
Last.
So you don’t spend time and money automating something that shouldn't even be happening.
But we business owners tend to start at step five.
We get excited about what AI can do and start automating before we've figured out how the process should work.
I talked about this recently with Jake George, co-founder of Agentic Brain. His team has built more than 700 AI employees for businesses.
He said that one of the most useful things they do is talk to the people who actually do the jobs they're being asked to automate.
The owner describes how a process works.
Then the team describes how they actually do it.
And those aren't always the same thing.
That's not surprising.
As businesses grow, processes evolve.
People find shortcuts and create workarounds.
They find other ways to get something done.
Whether it's the best way or not.
Jake told this story about a company that had designated an employee as their "AI guy."
He was “AI guy” only because he knew more than anyone else on the team about AI.
And all he had done was shared an AI hobby project he’d worked on on his own.
Anyway, they had the AI guy build an automation to process documents.
And over two weeks it ran up more than $2,000 in AI costs.
Come to find out, the system he’d built was processing every page of documents that were 50-80 pages long.
And the information they actually needed was on the first page.
So that’s 50-80 times the AI tokens needed.
That’s one way you can burn $2000 unexpectedly.
It can happen pretty easily.
I had a workflow that was running 100 operations every cycle,
and it should have stopped after the first step when the queue was empty.
Buuuuuuut we hadn't set it up to do that.
So it ran the other 99 every time.
That’s 100x what it should have cost.
Yeah, AI tokens cost fractions of a penny.
But can your business really afford to pay 50, 80, or 100x cost on something?
We were fortunate and caught it the same day.
By luck.
If we'd missed it, it could've kept burning credits for months.
Before you automate a process, you need to understand what you're trying to accomplish, how the work currently gets done, and whether that's even the way it should be done.
Then you can figure out what to improve, eliminate, delegate, or automate.
AI might be the answer.
Or maybe it's a clearer handoff, better accountability, or eliminating work that doesn't need to happen at all.
The bigger opportunity is to design a better process before you spend money making the old one faster.
That's one of the big reasons I enjoyed my conversation with Jake.
He spends a lot of time telling people when not to use AI.
He's approaching it differently, trying to help businesses figure out where it actually makes sense.
We're kicking off Season 4 of The Business Owner's Journey with that conversation.
Listen to Jake George: Why Most AI Implementation Fails and How to Get It Right