Recently, the automations behind our operations changed noticeably in just a couple of weeks. Nobody sent an announcement. I found out on a Monday morning, from a routine check.
Some automations were new, some had been switched on or off, and a few had quietly stopped. One of the ones switched off fed data into finance. Everyone had a good reason. None of it was written down anywhere I would have looked.
That is normal. It is also how a company ends up closing the month with numbers nobody can explain.
Why your systems drift
Drift isn’t caused by careless people. It comes from fast, reasonable ones. In a small team, the person who notices a problem is usually the person who can fix it, so they fix it on the spot.
Three things make this worse:
- No-code tools make changes cheap. Switching an automation off takes one click. Writing down why takes ten minutes, so nobody does it.
- Ownership is shared. Ops, finance and engineering all touch the same automations. Each person sees their part, nobody sees the whole picture.
- Documentation is written once. The process doc was accurate the day it was written. Every change since then has made it a little more wrong, and nothing tells you how wrong.
The result is two versions of your company: the one in your docs and the one actually running. Month-end close, audits and handovers are where the gap shows, usually at the worst possible moment.
The weekly drift check
The fix isn’t more documentation. It’s a short weekly comparison between what your docs say and what is actually running. You aren’t rewriting anything, only looking for differences.
| Check | Compare | Red flag |
|---|---|---|
| Automation inventory | Count and list of automations vs. last week | New or deleted automations nobody mentioned |
| On/off state | Which automations are active vs. what the docs say | A core sync switched off, or something marked “test” running live |
| Errors | Automations that stopped or failed since last check | Anything that stopped on errors and is still stopped |
| Field mappings | The fields your automations read and write still exist | A renamed or deleted field that an automation depends on |
| Process docs | Last-updated date of each doc | Docs that haven’t changed while their systems have |
Each check is cheap on its own. Run weekly, they turn drift from a month-end surprise into a Monday to-do.
How I run it without writing code
I don’t code. I run this check with an AI assistant connected to our tools, and it keeps an operations wiki up to date for me.
The setup has three parts:
- A living wiki in plain language. What each tool does, how data moves between them, and why things were set up the way they are.
- Regular updates. The assistant reads team conversations and meeting notes, and files only real process changes. One-offs and chatter get skipped, which is what keeps the wiki readable.
- A weekly check. The assistant compares the wiki against the live systems and flags what changed, what broke and what went stale. I decide what to do about it.
The AI isn’t the point. You could run the same check with a spreadsheet and a calendar reminder. What matters is that someone compares docs against reality on a schedule, instead of when something breaks.
What it caught, and what to do with it
None of those changes were mistakes on their own. Together, they meant our docs described a system that no longer existed.
Finding drift is half the job. The other half is deciding what each finding means. I sort every finding into one of three buckets:
- Fix it. Mechanical gaps: a missing page, a broken link, an outdated count. Update the doc and move on.
- Ask the owner. A core automation switched off, or one stopped on errors. Was it on purpose? Is it temporary? Who switches it back on? One message, same day.
- Record the decision. If the change was deliberate, write a short decision note: what changed, who decided, why, and the date. Next quarter, nobody will remember the reason, and the note saves you from “fixing” something that was turned off on purpose.
The goal isn’t to stop people from changing things. Fast changes are a feature of a small team. The goal is that every change is visible within a week.
Steal this: the Monday checklist
Copy it into your calendar invite and run it every Monday before the week gets busy.
- ☐ List every active automation and compare against last week’s list
- ☐ Flag anything new, deleted, switched on or switched off
- ☐ Check for automations that stopped or errored since last Monday
- ☐ Spot-check that the key fields your automations depend on still exist
- ☐ Find docs that haven’t been updated in a while and mark them for review
- ☐ Message the owner of every unexplained change, same day
- ☐ Write a one-paragraph decision note for every deliberate change
- ☐ Save this week’s list as the baseline for next Monday
If you have no baseline yet, start with step one and stop there. A list of what’s running today is already more than most early-stage companies have.
The real lesson
Founders spend a lot of energy building systems and very little checking that those systems still match the docs. But a system nobody checks is a system nobody owns.
You don’t need a bigger team or better documentation to fix this. You need a short slot on the calendar, a list, and the habit of looking every week. Start this Monday.