The AI Said It Was Done. It Wasn't.
A few weeks ago I asked an AI workflow builder to set up a simple missed-call follow-up. Text the person back, wait a day, and text again if they haven't replied.
It built something. Then it wrote me a tidy summary saying it had included the one-day wait and the second text.
It hadn't. Neither was there.
What it actually built
When I opened the workflow and read the steps, there was one text. No wait. No second message. It had also used a tag that didn't exist anywhere in the account, and created a task with no due date, the kind nobody ever sees on their list.
If I'd trusted the summary, every missed caller would have got one text and then silence. Nothing would have looked broken. The follow-ups just wouldn't have happened, and I'd never have known why the callbacks dried up.
I'm not knocking the tool. As a first draft it saved me some clicking. But the summary was confident and wrong, and that's the part worth paying attention to.
Why this matters if you run a small business
More owners are using AI to build things now: email sequences, booking flows, chatbots, website pages. That's a good thing. It's fast and it's cheap.
The catch is that AI tools are very good at sounding finished. They describe what they meant to do, not always what they did. And the gaps tend to show up where nobody is looking, like the second text a customer never receives.
How I check AI work now
- Read the steps, not the summary. Open the thing it built and go through it line by line. The summary is a description. The steps are the truth.
- Run it like a customer would. Call or text from your own phone. Wait out the delay. See what actually lands, and when.
- Check every name it mentions. Tags, calendars, people, pages. If it refers to something, make sure that something exists.
- Look for anything nobody will see. A task with no owner or due date is a task that never gets done.
The rule I work by
AI does the draft. A person checks the work. Especially when it tells you it's done.
That's how we work at Smartech. I test everything we build before it goes live, the same way a customer would hit it.
If you've got a follow-up running and you're not sure it does what you think it does, I'm happy to take a look: smartechsystems.com.
Chris Small, Smartech Systems
