Home / Playbook / Integrations

Playbook · Integrations

Why your automations keep breaking.

The Zap that quietly died. The "success" that did nothing. The bill that tripled. I'll show you what's really going wrong — and how to hook your tools together so they stay hooked. No pitch — just straight answers.

Quick reason to trust me here: my day job is wiring tools together for a company with hundreds of locations. I once took 162 tangled workflows and boiled them down to a handful — and proved the new ones did the exact same thing before I shut the old ones off. Keeping this stuff from breaking is the whole job. This is that same know-how, shrunk down to what a shop your size actually needs.

My Zap keeps turning itself off — but there's no error.

Zapier will quietly shut a Zap off on its own. It does this when your plan or free trial changes, when you edit the Zap, or when it fails on almost every run for a few days. It's supposed to email you — but that email is easy to miss, or it never shows up. So you find out days later, when a customer asks why nobody got back to them.

Checking Zapier more often isn't the fix. The fix is your own alarm — a little check that pings you when a job doesn't run, instead of trusting Zapier to tell you it broke.

The catch: it turns itself off, and the warning email is easy to miss.

It says "success" — but nothing actually happened.

This is the scary one. A green "success" only means the other app said "got it" — not that it actually did anything. It can still toss your thing as a duplicate, stick it in a line to deal with later, block it on some rule of its own, or flag it as spam right after saying "yes."

A checkmark is a promise, not proof. So don't trust the green light — go check. After the automation runs, make sure the thing it was supposed to create actually shows up where it belongs. "Done" on the screen and "the email actually sent" are two different things.

The catch: "success" means they heard you, not that it happened.

It worked for months, then suddenly stopped.

Something changed under the hood. It's almost always one of three things. A login quietly expired — Google, Microsoft, and QuickBooks all make you sign back in on a schedule, whether you notice or not. Or the app on the other end changed how its info is laid out. Or a field got renamed or left blank, so a filter skipped the run without a word.

The automation can't tell a connection is dead until it goes to use it. So the misses pile up out of sight, stuck in a "held" pile you never look at.

The catch: a login or a field quietly changed, and nothing told you.

Why is Zapier so expensive now?

Zapier charges you per task — and every single step of every run counts as a task. So a Zap with five steps that fires 200 times a month is 1,000 tasks. That means your bill grows as you get busier, not as you add more Zaps. Plenty of shops start around $30 a month and are paying $300–600 six months later.

Make and n8n charge a bit differently and usually cost less once you're doing a lot — the trade is they're fiddlier to set up. But jumping to a new tool doesn't fix the real problem (that's below). Making it reliable does.

The catch: you pay per task, so the busier you get, the more it costs.

It failed silently and my client noticed before I did.

This is the nightmare, and it's baked in: none of these tools warn you when something stops. Out of the box, they say nothing. n8n won't make a peep unless you build a separate "if it breaks" path yourself. Make just switches your workflow off after it fails a few times, sometimes without even telling you why.

So you have to set up the warning yourself. Two parts: an alert for when a job doesn't run when it's supposed to, and an alert that texts or emails you the second something errors out — loudly, not quietly. A break you hear about right away is a hiccup. A break your customer finds first is a lost account.

The catch: nothing warns you when it stops — you have to build that yourself.

My apps keep randomly disconnecting.

Logins expire on purpose (Google and Microsoft are the big ones), or someone changes a password, or an admin locks down which apps are allowed to connect — and your automation only finds out the next time it tries to run. Two things cut it way down. First, connect the automation using its own dedicated account, not a real person whose password is going to change someday. Second, keep an eye on the connection so you reconnect before a pile builds up.

The catch: it's plugged in through a person's login, which won't last. See access & logins.

I have too many tools that don't talk to each other.

Every tool keeps its own copy of your customer list and your jobs. So unless you pick one place to be the boss of each kind of info, every new app you add is one more copy that can quietly drift out of line — the same customer twice, an old invoice, a follow-up that never goes out.

More Zaps won't fix that. Pick one home for each kind of info — customers live in your CRM, money lives in your accounting app — and make every other tool pull from that home instead of editing its own private copy.

The catch: nothing is the boss, so the copies drift apart.

How I'd approach it

Broken automations don't need more automations. They need to be built so they don't break.

Here's how I'd think about it. A few plain rules turn a pile of flaky Zaps into something you can trust with real money. Make every step safe to run twice without doubling up — so a re-run never sends the same invoice again. Make it able to pick back up where it left off if it trips. And make it yell when something's wrong — an alert when a job doesn't run, plus a quick check that the result really landed. Keep one home for each kind of info. And run the old way and the new way side by side until they match, before you trust the new one.

That last part — proving they match before you flip the switch — is exactly how I move the automations a business runs on without anyone feeling a bump. It's way too much for a little hobby Zap. It's the whole game for the ones your money rides on.

Make it reliable

Tired of babysitting automations that keep breaking?

Just tell me which ones keep dying on you — or which tools still won't talk to each other — and I'll tell you straight what it'd take to make them boringly reliable.

Tell me what keeps breaking →