Automation Systems
Three automations that save 10 hours a week for service businesses
August 15, 20264 min
Benji's Buns is a cinnamon bun bakery in New York's West Village, and its flavors change every week. We built one automation that notices the menu changed, writes the announcement, and publishes it to the bakery's blog, Facebook, Google Business Profile, the Bun Club community, and a Bun Club email. The part that made it reliable was not the AI writing. It was the rules that stop the system from posting when something is wrong.
Disclosure: I hold an ownership interest in Benji's Buns. The numbers below come from the workflow's own run records.
Key takeaways
- A weekly flavor change at Benji's Buns now publishes to five places from one trigger, with no one retyping the announcement.
- The blog post has to be live and reachable before any social post or email goes out, so every channel links to a page that actually exists.
- On September 9, 2026, the email step reached 894 of 953 Bun Club members; the rest were invalid or unsubscribed addresses the system skipped on purpose.
- Four real failures showed up in the first live weeks, and each one became a permanent guardrail instead of a one-time patch.
- The hardest problem was not writing copy. It was making sure the same announcement never posted twice.
What did Benji's Buns need?
A bakery with rotating flavors has a simple marketing problem that eats hours. Every week someone has to write the new flavors up, post them on social, update the Google listing, tell the loyalty club, and put something on the website. Miss a week and regulars do not know the new buns exist. Do it by hand and it competes with actually running a bakery.
The goal was one change in the bakery's menu system producing every announcement automatically, in Benji's voice, on the same day.
What did we build?
The automation runs on n8n, a workflow tool SimplySync hosts, and connects to the SimplySync platform that runs Benji's website, social accounts, community, and email list. Here is the order it works in:
- Detect. On a schedule, it reads the current flavor lineup from the bakery's menu and builds a fingerprint of it. If the fingerprint matches last week's, it stops. Nothing publishes on a week with no change.
- Write. Claude, an AI model, drafts the blog post, the social captions, and the email in the bakery's voice.
- Check the copy. A validation step confirms the draft has every required field. If it fails, the system asks the AI for one corrected draft and checks that one too.
- Tell a human. Before anything goes public, I get a notice that it is publishing now.
- Publish the blog first. The blog post must come back saved, published, and loading at its real web address.
- Fan out. Facebook, Google Business Profile, and the Bun Club community get posts that link to that blog post, using the same rotating announcement image.
- Email the Bun Club. The list is pulled fresh, bad addresses are skipped, and each send is counted.
- Report. The run reports sent or partial, never a blanket success.
What worked?
The "no change, no post" rule. The fingerprint check means the automation can run as often as it wants without spamming anyone. On September 10, 2026, the day after a live drop, it ran, saw the same three flavors, and stopped without publishing or emailing.
Blog-first publishing. Every social post and email points at the blog article. Requiring that article to be live before anything else goes out means no one clicks a link from Facebook and lands on an error page.
Honest counting. The September 9, 2026 run processed 953 Bun Club contacts. 894 emails were accepted for delivery, 18 were rejected because the address itself was invalid, and 59 contacts were skipped because they had no usable email or had unsubscribed. The report said exactly that, instead of "sent."
One image system. Five square announcement graphics rotate in a fixed order across the blog, Facebook, Google Business, and the community, so the brand looks consistent without a designer touching each week.
What did not work, and what did we change?
This is the part most automation write-ups skip. Every one of these happened on a live system.
1. Text messages tripped a carrier limit (August 21, 2026). The original design also sent the flavor drop by text. Roughly 870 messages went out in one unstaggered burst at about two per second, and the carrier's velocity cap stopped the send partway through. We turned SMS off for this campaign. If it comes back, it will be batched, paced, and capped per day before a single message goes out.
