It is the night of the 5th, the eve of commission payday. The payroll analyst needs the closed sales ranking to check the payroll at 8 a.m.
The dataflow feeding the model failed at 10 p.m., after running for six minutes. Refresh history shows the word Error, a Request ID and an Activity ID. Not one line says which table, which source or what went wrong.
The BI analyst who published the model gets the failure email, as Power BI does by default. However, the alert does not say what to do, and the dashboard stays open with yesterday's number, without saying it is from yesterday.
In this article we explain what Power BI notifies, why the error arrives opaque, what a failure alert must carry and when it is worth stopping the retries.
What Power BI notifies when a refresh fails
Power BI does notify, and it does so by default. According to Microsoft's documentation, a failed scheduled refresh emails the semantic model owner, and pushes a notification to the mobile app if the owner uses it.
- Additional contacts. The Email these contacts when the refresh fails box accepts other addresses, including a team alias. A group alias gets the email but not the mobile push.
- Shutdown after repeated failures. After four consecutive failures, the service disables the schedule and sends an email saying so. That threshold cannot be changed.
- Errors that will not fix themselves. An expired or invalid credential disables the schedule right away, without waiting for the fourth failure.
- Manual refresh does not notify. On-demand refreshes and those triggered through Power Automate do not send failure emails.
The design is reasonable. The trouble sits in two places it does not cover: who owns the model today, and what the message says.
The alert arrives. What it lacks is content.
Why the error arrives as a Request ID
On the Power BI forum, on September 30, a user described exactly the scene above. The dataflow stopped after six minutes, with an error message made only of a Request ID and an Activity ID. The dataflow had worked for months and, that morning, it simply stopped.
A Request ID is not an explanation. It is the number Microsoft's support uses to find the run in the service logs. For whoever opens the history, it says something failed, but not whether it was the source, the gateway, the credential or the capacity.
And when the failure goes away on its own, the diagnosis goes with it. The analyst does not know whether to change something or just wait, and the next failure starts from the same place. The message does not even help decide whether to open a ticket.
An identifier helps support. It does not help whoever is on call.
Five things a failure alert needs to carry
The alert exists so someone can act before another person opens the dashboard. For that, it has to answer the questions the analyst would ask first.
- Where it failed. Which table, which step, which source. "The sales model failed" means opening five screens.
- The source error in plain text. The message the database or API returned, not only the internal code of the service that called it.
- A direct link to the run. The shortest path between the email and the technical detail is worth more than any explanatory paragraph.
- Since when the data is stale. The last successful run tells how much of the dashboard still holds.
- What happens next. Whether there will be another attempt, when, and how many are left before the schedule is switched off.
None of these items needs a new tool. They need the failure to be recorded where it happens, not only where it is noticed.
A well-described failure is half solved.
Retry or stop: what each choice costs
Transient failures and configuration failures call for opposite responses. A transient one, such as a network blip, a restarting gateway or a busy source, usually clears on the next attempt. A configuration one, like a changed password, fails every time.
Retrying without a limit has a cost. Every attempt loads the source, the capacity and, in usage-billed tools, the quota. Stopping too early costs too: the dashboard goes stale over a failure that would have cleared on its own.
A concrete example
On Power BI Pro, which runs on shared capacity, a model can have up to eight refresh slots a day. Refreshing every three hours, the fourth failure in a row comes nine hours after the first. If the first one is at 10 p.m. on the 5th, the schedule is only disabled at 7 a.m. on the 6th, one hour before the payroll check.
The stop rule solves nothing by itself. It only helps if the shutdown email reaches someone who can act that night.
Stopping is good. Stopping unnoticed is not.
Who should receive the failure alert
The model owner is whoever last published or took over the model. On a team of one or two people, that changes during holidays, when someone leaves, and the day someone republishes the file from their own laptop.
That is why the additional contacts box matters. A team alias makes sure the email arrives even when the owner is away. The on-call phone has to be listed individually, because the app push does not go to an alias.
It is also worth alerting whoever depends on the number. The payroll analyst will not fix the dataflow, but can decide to wait instead of checking payroll against yesterday's ranking.
An alert that reaches the wrong person is silence.
Where Januss comes in
In Januss, the failure is recorded in the pipeline, where it happens. The failure email carries the error in plain text and says which consecutive failure it is, and when the pipeline succeeds again another email arrives, a recovery one. If the Runner installed in your network stops responding, that triggers an alert too.
The emails go to every user on the account, not only to whoever created the pipeline. That is Januss's answer to the question in the previous section, at the cost of no per-person choice.
On the home screen, the Needs attention panel lists the pipelines whose last run failed, with the way to the technical details. The history shows, per run, status, trigger, duration, failed steps and rows.
After three scheduled failures in a row, the pipeline switches itself off, and the email says so. As in Power BI, a manual run sends no alert. Januss alerts arrive by email.
Januss alerts on the pipeline, not on the Power BI refresh. If the data reached the destination and the gateway failed afterwards, Power BI is still the one that tells you.
Want to see what a real failure alert looks like? The 14-day trial needs no card. Create your Januss trial workspace.
Sources
Notification behaviour comes from Microsoft's documentation and the official Power BI forum:
- Microsoft Learn — Data refresh in Power BI
- Microsoft Learn — Configure scheduled refresh
- Microsoft Fabric forum — One of my Dataflow Gen 1 stopped working (09/30/2026)
Create your workspace in minutes.
Point at your source and watch the data reach the database the same day.