Automation maintenance is the regular checking and fixing that keeps an automation doing its job after the day it was built. Automations break because the systems around them change: a login expires, a form field is renamed, a connected tool updates, or a customer types something nobody planned for. Keeping them running takes three things: an alert when a run fails, a retry for temporary faults, and one named person who owns the result.
Why automations break
An automation is a chain of small agreements between systems. Your website form agrees to send a name, an email and a message. Your CRM agrees to accept them in certain fields. Your email tool agrees to send from a certain address. Break any agreement and the chain stops, often with no sign on the surface. These are the usual causes.
- An expired connection. Many tools link to each other with a sign-in that expires, is revoked when a password changes or when a staff member leaves. The automation stops until someone signs in again.
- A changed field. Someone renames a field in the CRM, adds a required field, or edits the website form. The automation still sends the old name and the new system rejects or drops it.
- An update on the other side. The software you connect to changes how it works, retires a feature or changes limits. You may not be told.
- Limits. Many services cap how many requests or messages you can send in a period. A busy week or a bulk import can hit the cap, and later runs fail.
- Unexpected data. A name with an apostrophe, a phone number in a different format, an empty field, an attachment that is too big. The workflow was built for typical cases.
- Duplicate triggers. A form is submitted twice, or two automations act on the same record, and the customer receives two messages or two records are created.
- People change. The person who built it leaves. The person who was meant to receive the alerts moves roles. Nobody knows how it works.
Why a silent failure costs the most
A loud failure is easy to deal with. An error appears, someone fixes it, and you lose an hour. The expensive failure is the quiet one. A form stops reaching the CRM on a Friday, and on Monday nobody has noticed that three enquiries never arrived. By the time you find out, those people have called someone else.
For that reason, the first rule of automation maintenance is that every automation must tell a person when it fails. If you have an automation that has run for months and you cannot say how you would know if it stopped, treat that as a fault.
Alerts, retries and an owner
These three pieces cover most of what goes wrong.
Alerts
Set up a message that goes to a named person, by email and ideally by text, whenever a run fails. Make the message say which automation failed, on which record and at which step. An alert that says “error” with nothing else gets ignored. Also set up an alert for the opposite problem: no runs at all. If your enquiry automation normally runs several times a week and has not run for three days, something is probably wrong upstream.
Retries
Many faults are brief. The other system was slow, or down for a minute. A good build retries a few times, with a pause between attempts, before it raises an alert. Retrying too fast, or forever, can cause its own damage, such as sending a message twice. A careful build makes sure a retry does not repeat a step that already worked.
An owner
Someone must be responsible for each automation. Not a team, not “the office”. The owner reads the alerts, does the monthly check, and decides what to do. They need a backup for holidays. Put the owner’s name next to the automation in the list described below.
An automation maintenance routine you can keep up
A routine only works if it is small. This one takes under an hour a month for a handful of automations.
- Weekly, five minutes. Open the run history in your automation tool or your CRM and look at failed runs. Clear them, or note why they happened.
- Monthly, fifteen minutes per automation. Run a test through the real front door. Submit your own form, book your own test appointment, create a test invoice. Confirm that each step happened and that the customer-facing message reads correctly.
- Monthly, five minutes. Check that connections are still active and that the alert still reaches a person. Send a test alert.
- Quarterly. Ask what changed. Did the website form change? Did anyone add fields to the CRM? Did the team change? Did a connected tool announce changes? Update the automation to match.
- After any change to a connected system. Test straight away, not at the next monthly check.
What a useful alert looks like
Compare two alerts for the same fault. The first reads “Workflow failed”. The second reads “New enquiry from J. Smith, received 2:14pm, was not added to the CRM because the phone field was empty. Nothing was sent to the customer. Add the contact by hand and reply today.” The second one can be acted on in a minute by anyone, including a person who did not build the automation.
Good alerts say what happened, to whom, what the customer has or has not received, and what to do next. If you are paying for a build, ask to see a sample alert before you sign it off.
Keep a list of what you have
You cannot look after what you cannot list. Keep a single sheet with a row for each automation:
- What it does, in one sentence.
- What starts it.
- Which systems it touches.
- Who owns it, and who is the backup.
- Where the sign-ins are held, and who can renew them.
- The date of the last test.
This is dull and it matters more than anything else on this page. It is also what lets a new person take over without starting from scratch, and what a specialist needs to fix something quickly. Our guide to workflow automation examples lists a check for each pattern.
Have a manual fallback
Decide in advance what happens when an automation is down. For enquiries, that may be a shared inbox that someone looks at twice a day, so the website’s contact form also sends a plain email. For reminders, it may be a list printed each morning. Write the fallback down. Without it, an outage becomes a scramble.
The same thinking applies to tools you rely on for replying quickly. If your missed call text back stops, calls go unanswered without anyone knowing. A weekly comparison of missed calls to texts sent takes a minute and catches it.
Who should look after it
Three options exist, and they suit different businesses.
- You or a staff member. This suits a small number of simple automations in a tool you know. You carry the time and the risk.
- The tool vendor’s support. Support helps with the vendor’s own product, but not usually with how your systems fit together.
- The person or business that built it. They know the design and can fix it quickly. Ask what is included, how fast they respond, and whether you can see the workflows.
CentriWeb builds and looks after automations for clients as part of automation work, starting with the part that costs the business most. If you already have automations built by someone who is no longer around, a review and a list is a sensible first job.
Nothing is set and forget
Treat an automation as you would a vehicle or a piece of equipment. It does a job every day, and it needs a small amount of regular attention to keep doing it. Before you build the next one, work out who will check it and how often. If you cannot name a person, the project is not ready. To weigh whether upkeep time fits the numbers, see automation ROI.