lesson learned

It told me all 400 were updated. One of them still had the old paragraph.

A few hundred emails were sitting ready to go, each one written for a specific business, and every one needed its middle paragraph rewritten before it went out. By hand that is a day. Automated it is a few minutes, which is exactly why this is one of the first jobs people hand to a machine.

I ran it. It came back and told me everything was updated. Three things were wrong and not one of them made a sound.

Some came out with broken links, because every time one of those drafts gets saved the link is rewritten, and saving twice wrapped it twice. Forty six of them lost the greeting entirely, so the first thing a stranger would have seen was their own name missing. And the counts it reported were wrong in both directions, calling some batches half done when they were finished, and calling one finished when an email in it still carried the old paragraph.

None of that showed up as an error. If I had trusted the summary, a few hundred people would have gotten a broken first impression from a guy who sells automation.

What I do now is boring. Run it on one. Open that one where the customer would see it, in the real inbox, not in the tool. Then run the rest, then pull five at random and look at those too.

It holds for anything that changes your customer records in bulk. Price updates across a catalog. A new template on your invoices. Moving everybody’s service date after a schedule change. The summary at the end is a claim about what happened. The proof is the thing itself, seen from where your customer stands.

That is the difference between an automation that saves you a day and one that costs you a week of apologies.

> I send a short AI report every week. What shipped, what matters if you build with AI.

Subscribe free →