Skip to content
caadoo
de

Automations that survive the builder leaving

Some automations run only because the one person who built them is still around. That works until they are on holiday, off sick or gone. This is how to take stock of what already runs and give every automation an owner, a one-page runbook, a team alert address, a shared vault and a quarterly check.

By T. Krause8 min read

Here is an invented example: a 60-person machine builder. Every web enquiry lands in the CRM with the right sales rep assigned, the quote template pre-filled, and a reminder set for day three. Nobody in sales thinks about how. Then the colleague from inside sales who set all of it up in Zapier two years ago goes off sick for three weeks, and on day four the enquiries stop arriving. The login he used for the CRM connection had expired. The error emails went to his inbox.

Sales notices a week later, when a prospect calls to ask why nobody replied. By then there are forty enquiries in a web form backend that nobody on the team knew existed.

The colleague did nothing wrong. He solved problems no one else was solving, and nobody ever asked him to write anything down. The risk was never him. It was that everything ran through him.

Photo: Unsplash

Why the hero admin is a risk

If your company has one, you know who I mean. The person who taught themselves Power Automate or Zapier and quietly wired up twenty small things that everyone now relies on. Their work is usually good. How it is set up is the problem.

Alerts go to one inbox. When someone builds an automation for their own work, the failure emails go to their own address. When that person is on holiday, the alert sits unread. The failure surfaces later, through a customer, an auditor or a missing delivery.

Credentials live in a personal account. The connections to the mailbox, the ERP and SharePoint run on that person's login. Microsoft's documentation has a name for a flow with no valid owner left, an "orphaned flow", and warns that it can fail when it uses connections tied to the departed user's account. A co-owner can keep a shared flow running after the creator leaves, but according to Microsoft a co-owner cannot change the credentials of a connection another owner created.

Some flows switch themselves off. In Power Automate, a cloud flow whose trigger or actions fail continuously for 14 days is turned off. A flow that isn't triggered for 90 days may be turned off as well, unless its owner has a premium licence or the flow has its own Process licence. Microsoft notifies the owners and co-owners 30 days before. Think of the flow that runs once a year for stocktaking. If the person who built it has left, that notice goes nowhere.

Nobody knows the automation exists. You can't miss what you never knew about. The first time most people hear of an automation is the day it stops.

Take inventory of what already runs

Before writing any rules, find out what you have. Expect more than IT knows about.

Read out the platforms. For Power Automate, an admin can open the Power Platform admin center, pick an environment, go to "Resources" and then "Flows", and look at the "Owners" column. A flow with no owner listed is orphaned. For Zapier, Make and similar services there is often no central view, because every user signed up with their own account. Ask each of them for a list or a screenshot of their active automations.

Follow the money. Credit card statements and cost centre reports show which services the company pays for. Any subscription nobody can explain points to an automation nobody can explain. Free accounts won't show up there, which is why the next step matters too.

Ask every department one question. In person if you can, because a round-robin email gets skimmed: "What happens in your team automatically, without anyone clicking?" Count Excel macros, scheduled tasks on a server and scripts someone wrote years ago. To the people using them, those are automations too, whatever tool they run in.

Put it all in one list. One list, one place, for the whole company. A SharePoint list or a shared spreadsheet is fine. One row per automation with its name, purpose, platform, trigger, business owner, technical contact, the account its credentials belong to, where it logs, where alerts go, the date of the last check and a link to its runbook.

Anything nobody can explain after that doesn't get switched off quietly. Announce a date, switch it off on that date, and keep it restorable. If nobody complains within a month, delete it.

What every automation needs

Every row on the list gets the same basics, the twenty old ones and every new one before it goes live.

An owner in the business. This is a named person in the department that benefits, not IT and not the builder. The owner decides whether the automation is still needed and is the first to notice when its output stops making sense. Name a deputy as well, because owners take holidays too.

A one-page runbook. What does it do? What triggers it? Where does it log? What do you do when it stops? How do you switch it off and do the job by hand? That last question matters most. The manual procedure has to be written down while someone still remembers it, because a year after go-live nobody will. The runbook covers the automation. The process behind it needs its own write-up, and the process documentation template gives you one to fill in.

Alerts to a team address. Failure notifications go to a shared mailbox or a Teams channel that several people read. Test it once by breaking the automation on purpose and seeing who reacts, and how fast.

Credentials in a shared vault. The automation runs under a service account the company owns, not under a person's login. The passwords sit in a password manager's shared vault that at least two people can open. In Power Automate, add a co-owner on top.

What happens to automations during an internet outage or the builder's holiday

A holiday on its own breaks nothing. Flows keep firing while their builder is away. The risk is the first error that turns up while they are away. If that alert goes to their personal inbox, nobody acts on it. Power Automate adds a deadline: Microsoft switches off a cloud flow that has failed without a break for 14 days, and a three-week absence leaves plenty of room for that.

An outage at the vendor or at a connected service is different, because runs then actually fail. Power Automate and Zapier both retry, on different terms. In Power Automate you set a retry policy on each action, with fixed or exponential intervals. Zapier's autoreplay works to its own schedule: it tries a failed step again up to five times, the last attempt roughly ten and a half hours after the first error, and only sends the error email once that final attempt has failed. A Zap that breaks at nine in the morning can stay quiet until the evening.

Your own office line going down matters least of the three. A flow that moves data between two cloud services is unaffected, since your building was never part of the route. The trouble sits with automations that run on hardware in the building: desktop flows and scheduled jobs on a local server stop with the line. So does a cloud flow that reads from the ERP in your server room through Microsoft's on-premises data gateway. That gateway only ever dials out to Microsoft's cloud, so while the line is down the flow cannot reach it.

Start the preparation with people. Before the next holiday, give each important automation a named deputy who can open it and has permission to restart it. The runbook page should tell that deputy how to do the job manually in the meantime and how to replay a run once the retries have run out, because after that nothing restarts it on its own. Then make sure a failure reaches more than one person: in Power Automate, a step whose "Run after" setting is "has failed" can post to the team address whenever the action before it breaks. Last, switch on retries in Power Automate or autoreplay in Zapier, so short outages fix themselves.

The quarterly check

Once a quarter, each owner takes their automations and holds the runbook up against what actually happens. It rarely takes long, and it is the only thing that keeps the list honest.

Check that it still runs. The log shows when it last ran and whether anything failed. An automation that has done nothing for weeks is either no longer needed or broken, and you want to know which.

Hold the runbook against reality. Has the trigger changed, the target system, the folder, the address? Then the owner updates the runbook now, not next quarter.

Check the alert address. Does the shared mailbox still exist, and do the right people still read it? After any change of staff in the team, this is the first thing to look at.

Check the manual route. Does the form still exist, the login, the knowledge? A fallback nobody can use any more is not a fallback.

The date of the check goes on the list. Anything without an owner, or without a check in the last six months, becomes a candidate for switching off.

An automation that one person understands belongs to that person, and it runs for as long as they have the time, the patience and an employment contract. The list, the runbooks and the quarterly check cost a few hours a year. That is a small price for turning a colleague's favour into something the company owns.

Topics

  • Automation
  • Operations
  • Power Automate
  • Documentation
  • Mittelstand
All articles

Keep reading

  • 7 min read

    Measure before you build: the baseline nobody takes

    Most automations are called a success because the work feels faster afterwards. Nobody can say how much faster, because nobody measured before. This is how to take a two-week baseline and write the one-line before and after statement a managing director can check.

  • 7 min read

    Human in the loop: when AI agents need approval

    Companies tend to give an AI agent either no freedom at all or far too much, and both go wrong in predictable places. The fix is to judge each action on its own, by whether it can be undone and what a mistake costs. This article gives you a five-step autonomy ladder you can apply to your own processes this week.

  • 8 min read

    AI assistant for company documents: fix files first

    Most companies planning an AI assistant for their own documents spend their time choosing a model. The real work is in the file share, with its outdated versions, documents nobody owns and access rights nobody has checked in years. Here is the order to do it in, from one bounded question area to a test with 30 real questions before anyone else gets access.

End of call

You can hang up now.

Or book 15 minutes. I'll ask you the questions I'd ask on the phone, and tell you at the end whether it fits. If it doesn't, we've both lost a quarter of an hour instead of a month.

Or call. Cold, if you like. My AI assistant picks up and I call you back. +49 5136 9414999

Analytics

With your consent I use Google Analytics to see which pages get read. Without it, no analytics code loads. Your answer is stored locally in your browser.

Privacy notice