Zapier Sidekick just got its first public receipts, and honestly they beat most automation case studies I’ve read this year. Zapier’s engineering team shared five jobs the tool really did inside the company, numbers included: a 55-cent report that writes itself, and an AI that cleared out 800 dead Slack channels. Here’s what worked, and how to copy the recipe.
Most automation advice is marketing
Search for AI automation examples and you’ll get the same thing over and over: a screenshot of a Zap someone built once, a stock photo, and a promise about saving 10 hours a week. No numbers. No follow-up. Often no proof the thing still runs a month later.
That’s why the post from Zapier’s engineering team caught my attention. It covers automations running inside Zapier itself, not customer fairy tales. They kept the numbers wherever people measured them, which almost nobody does.
Five examples, all real. Let’s go through what each one does, what it cost, and the pattern you can steal.
What Zapier Sidekick actually is
Zapier Sidekick is the company’s AI workspace, and it connects to more than 9,000 apps. You describe a job in plain English, it builds the automation, and the thing runs on a trigger or a schedule after that. No flowchart editor required.
Results land somewhere you can actually find them again. Reports go to pages, structured data goes to tables, and the team can search all of it later. If you’ve ever built something clever in ChatGPT and then lost it in your chat history, you already see why that part matters.
There’s a fun detail in the original post too: Zapier ran Sidekick as an internal pilot first, and someone had to open a Slack channel just for wins because the regular feedback threads were overflowing. When a tool gets out of people’s way, this is what adoption looks like.
Five real automation examples
All five ran on Zapier Sidekick, either inside the company or with a customer. I’ve compressed each story down to the part worth stealing, but the numbers are theirs, not mine.
Deal reviews that write themselves overnight
When an Enterprise deal closes, a Zap pulls the deal facts, runs an analysis, and drops a report on a Sidekick page. By the next morning, the team knows why they won, who was in the room, and where the next expansion could come from. Each report costs about 55 cents, and 58 of them were already saved when the story went up. Before this, one person wrote these by hand whenever they remembered, so most weeks nobody got one at all.
Out-of-office replies turned into a lead database
A sales rep noticed that out-of-office replies are quietly full of leads: who left the role, who replaced them, who to contact instead. Nobody reads that pile, so the information evaporated every single time. In about a week, the rep built a pipeline that catches each reply, separates bounce messages from humans, uses AI to pull out names, titles, and contact details, and writes everything to five connected tables. Weekly rollups. Nightly retention checks. Built by someone who doesn’t code for a living.
200 calendar invites without the scheduling pain
Zapier’s CEO and CTO wanted 10-minute conversations with a big chunk of the company. Scheduling that across time zones, vacations, and busy weeks is exactly the kind of work that eats a whole week and leaves nothing to show for it. A Sidekick Slack app booked most of the calls, dodged working hours and out-of-office days, and flagged the conflicts for humans. Roughly 200 invites went out, and the assistants only handled the exceptions.
Test coverage jumped from 25% to 75% in two weeks
Zapier’s observability team moved an internal AI feature onto Sidekick, and coverage on their target use cases went from about 25% to about 75% in roughly two weeks. The difference came from unglamorous additions: code execution let the assistant actually run things, and file handling let it produce real files back for users. Sometimes the unlock isn’t a smarter model. It’s hands.
800 dead Slack channels, finally gone
A customer had spent months trying to bulk-archive inactive Slack channels through the API and got nowhere. Then they described the chore to the agent instead. More than 800 channels were archived, with nobody clicking through them one by one. The best part of the story is the order of events: the technical route failed first, and the plain-English route worked after.
| Automation | Who built it | Result |
|---|---|---|
| Deal breakdown after every Enterprise close | Product marketing (one Zap) | About 55 cents per report; 58 saved |
| Out-of-office replies into a contact database | A BDR, in about a week | Five connected tables plus nightly checks |
| Listening tour scheduling | Executive assistants | About 200 invites handled |
| Internal AI feature rebuilt on Sidekick | Observability team | Coverage 25% to 75% in two weeks |
| Slack channel cleanup | A customer | 800+ channels archived |
The pattern behind all five
Every story started the same way: someone described a chore in plain words. Not a workflow diagram. Not a prompt library. Something closer to “this weekly thing I do is miserable.”
Then two things turned a demo into infrastructure. First, the automation got a trigger or a schedule, so nobody has to remember to run it. Second, the results got delivered somewhere the team already looks, like Slack or a morning page. One salesperson who built six connected automations in three weeks said it best: have the results come to you.
That’s the whole recipe. It also explains why most people’s ChatGPT workflows die after three days: nothing schedules them, and nothing delivers the output on its own.
How to try it without wasting a weekend
Pick one chore you actually do every week. Not a chore you wish you did. One with an inbox, a file, or a report attached to it.
Then describe the job to Zapier Sidekick like you’re onboarding a new teammate: where the inputs live, what good output looks like, and where the result should go. Run it manually a few times and read the output like a skeptical editor, because first drafts from any AI deserve that. Only after it survives your review should it get a schedule and a delivery destination.
You can start at sidekick.zapier.com. And if you want to see this same loop applied to a different AI, we already walked through a full Grok automation setup with Zapier, which connects the AI that watches X to thousands of apps.
The catch nobody mentions
Two honest caveats before you clear your Saturday. First, these numbers come from Zapier’s own team talking about Zapier Sidekick, so some glow is guaranteed. Second, the deal-review example assumes clean CRM data; most pipelines are messier than the demo, and messy input means messier reports.
Still, 55 cents per report is the kind of number you can test yourself this week. If the output lands at 80% of a human draft, you’ve already won, because the human only ever wrote it when they remembered anyway.
Your next step
Pick one weekly chore, hand it to the agent, and demand a number from it: cost per run, minutes saved, replies processed, anything you can measure again next month. If your routine doesn’t exist yet, our 7-day AI adoption plan builds one first. And if you’d rather have an agent hunt down busywork on its own, see how Perplexity Teammates works, because it’s the same shift toward agents that finish jobs instead of chatting about them.
One weekend, one chore, one number. That’s a better automation portfolio than ten unfinished Zap templates.