What's the difference between a threshold alert and an anomaly alert?

A threshold alert fires when a value crosses a predefined fixed limit (e.g., overdue receivables above 8%). An anomaly alert compares the current value against the expected historical pattern for that day and context, and notifies when the deviation is statistically unusual — catching problems a fixed threshold wouldn't.

Through which channels can automatic alerts be sent?

The most common are email, WhatsApp (via the WhatsApp Business API), Slack and Microsoft Teams. Best practice is to assign the channel by severity: instant messaging for urgent operational issues, email for summaries and indicators under watch.

How do you keep alerts from becoming spam?

Three rules: every alert must have a concrete action and an owner; every new alert is calibrated in silent mode before going live; and persistent conditions escalate to the next level instead of repeating to the same recipient. If an alert fires constantly, the problem is the process or the threshold, not the tool.

Dashboards

Smart Alerts: Let Your Data Notify You, Instead of Going to Check It

Okun Data Team · May 13, 2026 · 5 min read

The most complete dashboard in the world has a structural limitation: someone has to go look at it. And between meetings, emails and fires to put out, nobody watches dashboards all day. The result is familiar: the problem was on the dashboard since Tuesday, but it was discovered on Friday.

Alerts flip the logic: instead of people watching the data, the data watches for people and speaks up only when there's something to do. Well designed, they're the most cost-effective complement to any dashboard. Poorly designed, they're spam everyone learns to ignore.

The three types of alert worth having

Threshold: the classic. "Stock of X fell below reorder point", "overdue receivables exceeded 8%", "tickets unassigned for more than 2 hours". Simple to define and understand.
Anomaly: they fire when something deviates from the expected pattern, without a fixed threshold. "Today's sales are 40% below normal for a Tuesday in May". They catch problems no fixed threshold would detect — especially useful with seasonality.
Absence: the most underrated. "The bank file didn't arrive", "the North branch didn't record sales today", "the nightly job didn't run". They detect silence — often the first sign of a technical or operational problem.

The golden rules for avoiding spam

Every alert must have an associated action. If nobody has something concrete to do upon receiving it, it's not an alert: it's noise. The design question is always "who does what when this arrives?".

Calibrate before switching on. Every new alert runs silently for two weeks to measure how often it would have fired. If it's 30 times a week, either the threshold is wrong or the problem is chronic (which makes it a project, not an alert).

Escalate, don't repeat. If the condition persists, the alert isn't re-sent hourly to the same person: it escalates — to the owner, then the manager. Repetition trains people to ignore; escalation trains them to resolve.

The channel matters as much as the content

An urgent operational alert goes to WhatsApp or Slack; a summary of yellow-flag indicators goes by email in the morning; and whatever can wait for the weekly meeting doesn't need an alert at all. Mixing severities in one channel is the fastest way to kill the system: when everything is urgent, nothing is.

Technically, alerts run on the same foundation as your dashboards — which is why they're the natural next step after a business intelligence project. If the data is already integrated and modeled, adding the monitoring layer takes days, not months.

Conclusion

A good alert system is a tireless employee who watches every number all the time and only interrupts when appropriate. Start with five well-chosen alerts — the ones that would have prevented your last three scares — rather than fifty generic ones. The measure of success is simple: every message received should trigger an action.

Finding out too late about what happens in your business?

We'll design your alert system: what to watch, with which thresholds, on which channel. Running in two weeks.

Request a demo

Frequently asked questions

What's the difference between a threshold alert and an anomaly alert?: A threshold alert fires when a value crosses a predefined fixed limit (e.g., overdue receivables above 8%). An anomaly alert compares the current value against the expected historical pattern for that day and context, and notifies when the deviation is statistically unusual — catching problems a fixed threshold wouldn't.
Through which channels can automatic alerts be sent?: The most common are email, WhatsApp (via the WhatsApp Business API), Slack and Microsoft Teams. Best practice is to assign the channel by severity: instant messaging for urgent operational issues, email for summaries and indicators under watch.
How do you keep alerts from becoming spam?: Three rules: every alert must have a concrete action and an owner; every new alert is calibrated in silent mode before going live; and persistent conditions escalate to the next level instead of repeating to the same recipient. If an alert fires constantly, the problem is the process or the threshold, not the tool.