Spot a Silently Failing Workflow: A Weekly Read
How to spot a silently failing workflow: read the counts and the real messages each week, treat zero as broken, and measure the outcome, not whether it ran.
To spot a silently failing workflow, check each week how many people entered, how many reached each step and how many failed, read a sample of the messages that went out, and treat any count that drops to zero as broken until proved otherwise. Then measure the outcome, not just whether it ran.
A workflow that fails loudly is an inconvenience. You see the error and fix it. A workflow that fails silently is a dropped handoff that looks like a working process, which is the very thing you built it to prevent. This post is about the habit that catches it.
How to spot a silently failing workflow: why it is hard
Every workflow has two versions: the one in your head and the one that runs. They drift apart as fields go stale, stages are renamed and people arrive by routes you did not anticipate. Nothing announces the drift. The rule keeps running, the dashboard looks calm and the original purpose quietly stops being served.
Silent failures tend to take a few recognisable forms.
| Form of silent failure | What it looks like |
|---|---|
| A step that runs but does nothing useful | The step shows as complete and no effect appears on the record |
| A condition that sends everyone one way | One branch is busy and the other is empty |
| A trigger that stopped firing | Entries fall away after a form was rebuilt |
| A workflow that runs and achieves nothing | Everything is green and the outcome has not moved |
The third row is the one the guide singles out: a trigger that stopped firing after a form was rebuilt. It is silent because nothing went wrong in the workflow itself. Its cause lies somewhere else.
The weekly look
The defence is a fixed, short routine. Once a week, for each workflow that matters, look at four numbers and one sample.
- How many people entered.
- How many reached each step.
- How many completed.
- How many failed, and why.
- A handful of the actual messages that went out, read as the recipient would see them.
Then compare all of this with what you expected. Where the numbers surprise you, there is almost always a reason worth finding. Surprise is the signal, and the routine exists to produce it.
Why quiet failures last so long
A loud failure gets fixed because it demands attention. A quiet one survives because everything around it looks normal. Nobody is complaining, the workflow still appears in the list and the last time anyone checked, it worked. Time passes, and the gap between what you think is happening and what is happening widens.
There is a second reason, which is ownership. A workflow without an owner is a process nobody is responsible for, so nobody has a reason to look. The weekly read works best when it belongs to a named person, and it works worst when it belongs to everyone.
Zero means broken
If you remember one habit from this post, make it this one. When a workflow's numbers drop to zero, assume it is broken until you have proved otherwise. A quiet week is possible, but it is the less likely explanation, and assuming it is the one that lets a broken rule run unnoticed for months.
The proof can be simple. Fire the trigger yourself, on a contact you control, and watch whether the workflow starts. The method is the one in how to test a workflow before switching it on, applied to a workflow that is already live.
Read the messages, not just the counts
Numbers show that something happened. They do not show what it looked like. A message can be sent, counted and still be wrong: a missing name, a broken link, wording that no longer matches the offer. Read a few real ones, in the form the recipient saw.
It also pays to look at what people say back. Replies and complaints are an early sign that your automation is making people feel processed rather than helped. If people reply asking whether a human is there, that is feedback on the workflow as well as a message to answer.
Reading the numbers against what you expected
The comparison does the work, so write your expectation down before you look. It does not need to be precise. It needs to be specific enough to be wrong.
| You expected | You found | What to suspect |
|---|---|---|
| People entering most days | No entries at all | The trigger stopped firing |
| People spread across both branches | Nearly everyone on one branch | A condition reading a field that is empty or stale |
| Most people reaching the last step | Many stopping part way | A step failing, or a wait that never ends |
| Few failures | A cluster of the same error | One cause behind many failures |
| The same message once per person | Some people appearing twice | Re-entry allowed by accident |
Where the finding matches the expectation, move on. Where it does not, do not explain it away. Open one affected record and follow it through. One concrete case usually explains the pattern.
After a change, look again
Do not wait for the usual cycle when something underneath has moved. The monthly review in the guide's checklist is meant to run after any change to forms, fields, stages or pipelines. Those are the changes that break triggers and conditions without touching the workflow itself.
| What changed | What may have broken |
|---|---|
| A form was rebuilt | The trigger that read the old form |
| A stage was renamed | Triggers and conditions that name the old stage |
| A field changed or went stale | Any fork that branches on it |
| A pipeline was reshaped | The assumption about which deals the workflow sees |
Each of these can be checked in a few minutes, provided you remember to check. A short note beside each workflow listing what it depends on makes the check quick. It also helps to know why a trigger was chosen, which is the subject of how to choose a workflow trigger.
Measure whether it worked, not whether it ran
A workflow running successfully is not the same as a workflow working. The question is whether the handoff it was built to protect is now being kept. This is the deepest silent failure of all, because every indicator can be green while nothing improves.
| You built it to | Then measure |
|---|---|
| Speed up first replies | Whether the time to a first reply is actually shorter |
| Stop deals stalling without a next step | Whether there are fewer stalled deals |
| Make sure a welcome happens | Whether new customers now receive it |
Take the measurement before you switch the workflow on and again a few weeks later. If nothing moved, the workflow may be running perfectly and achieving nothing, and it is worth knowing that before you build three more like it. A perfectly executing workflow that does not move the outcome is overhead with a progress bar.
What to do when you find one
- Do not trust it until it is fixed. If the damage could reach people, consider pausing it first.
- Find the cause. Read every failure and understand why it failed.
- Check what changed. A form, a field, a stage or a pipeline is often the answer.
- Fix it and test again. Treat the repair as a change and run the checks you would run for a new workflow.
- Decide about the people affected. Some may have missed a step, and some may have received the wrong one.
Sometimes the honest conclusion is that the workflow has outlived its purpose. If so, the decision is a different one, and when to switch off a workflow sets out how to make it.
Keeping the number of workflows reviewable
A weekly look only works if you can finish it. Every workflow you add is another one to read, and the habit collapses when the list outgrows the time. The monthly checklist includes a plain question for this: is the number of workflows you run one you can actually review? If the answer is no, the right response is fewer workflows, not less review.
Checklist for the weekly look
- I checked how many people entered, reached each step and completed.
- I read every failure and know its cause.
- I treated any workflow whose numbers dropped to zero as broken.
- I read a sample of the messages that actually went out.
- I compared the outcome with the measurement from before.
- I confirmed the trigger still means what it meant when I built it.
Frequently asked questions
How often should I look?
The guide suggests once a week for each workflow that matters, with a fuller monthly review and an extra look after any change to forms, fields, stages or pipelines.
What if the numbers look fine?
Then read some real messages and check the outcome. Healthy counts show that the workflow ran, not that it worked.
Can a step fail without being reported?
It can, and that is the danger. A step that cannot do its job should fail visibly, and where yours does not, reading the records afterwards is the only defence.
Who should do the review?
The workflow's owner. Every workflow should have one named person who can say what it is for and whether it is still doing it.
The full method, with the checklists, is in The Automation Playbook, a free 29-page playbook.