The Lobbi Delivery Team
Operational Systems Engineering
Ask the team why the missed order was missed, and the answer is uncomfortable: the system flagged it. A notification fired, landed in the channel, and scrolled away under fourteen routine updates within the hour. The person responsible saw it, registered nothing, and moved on — exactly as their environment trained them to.
This is alert fatigue, and it is not a discipline problem. It is a design problem. Reliability engineers found the principle the hard way, monitoring systems at enormous scale: page a human only when human action is needed. Every violation of that rule withdraws trust from every future alert.
How Notification Debt Accumulates
No business designs a noisy environment on purpose. It accumulates one reasonable decision at a time. A new app defaults to notifying everything, and nobody changes the defaults. An automation gets a confirmation ping so we know it ran. A manager asks to be copied on a workflow temporarily, three years ago. A channel that started as urgent issues drifts into general chatter.
Each notification is individually defensible. The aggregate is an environment where a person receives 150 to 300 pings a day across email, chat, and apps — and where the genuinely critical one has no way to look different from the rest.
The cost compounds twice. First, interruption cost: attention research consistently shows that recovering focus after an interruption takes meaningful time — multiply by dozens of interruptions and you lose hours per person per week. Second, and worse, signal decay: each no-action ping raises the threshold at which anyone reacts. The team is not ignoring alerts because they are careless. They are ignoring alerts because ignoring alerts has been the correct response hundreds of times in a row.
The Sorting Rule
Everything your systems emit belongs to exactly one of three classes, and the entire fix is honoring the boundaries.
| Class | Test | Delivery |
|---|---|---|
| Alert | Someone must act now or soon | Ping, routed to the actor, escalates if unacknowledged |
| Digest | Worth knowing, no action required | Batched once or twice daily, scannable |
| Log | Worth keeping for diagnosis or audit | Stored, searchable, silent |
The test for the first class is strict on purpose. An alert must name an action and an owner. Payment failed on invoice 4127 — retry or contact the customer is an alert. Backup completed successfully is a log entry wearing an alert's clothing. FYI, the vendor updated their portal is digest material.
Run your current notification stream against the table and the result is usually lopsided: well over 80 percent of what pings today belongs in digest or log. That is the recoverable attention.
Doing the Cleanup
The redesign is a half-day exercise followed by two weeks of tuning.
- Inventory the pings. Have two or three people screenshot or tally every notification for two days — source, content, and whether anyone acted. The tally is the business case; it always surprises.
- Reclassify ruthlessly. Apply the three-class test. Default to demotion: anything without a clear action-and-owner drops to digest or log. The burden of proof sits on staying an alert.
- Route alerts to actors. An alert sent to a channel of nine people is an alert sent to no one — everyone assumes someone else owns it. Route to the role that acts, with a fallback if unacknowledged.
- Build the digest. One scheduled summary a day for the worth-knowing class. Most platforms can batch natively; where they cannot, a simple automation that collects into one message does the job.
- Re-tune in week two. A real alert will get missed because it was wrongly demoted, and a noisy one will sneak back. Adjust. The classification is a living standard, not a one-time purge.
The Standard That Keeps It Clean
Cleanups decay without a rule for new entrants, and the rule is one sentence: anything that wants to ping a human must name the action and the owner, or it ships as digest or log. Apply it to every new tool, every new automation, every well-meant request to be kept in the loop.
There is a deeper benefit waiting at the end of this work. Automated operations need trustworthy alarms — the more your business runs itself, the more the exceptions are the whole human job. A team that believes its alerts responds in minutes. A team trained by noise responds eventually. When something genuinely breaks at 2 p.m. on a Friday, that difference is the business.
If your tools send 200 notifications a day and 195 require no action, you have spent months training your team that pings are noise. They learned.
Sources
Topic clusters
Ready to see where the friction is?
The Lobbi's Operations Discovery maps your workflows, identifies your highest-impact bottlenecks, and gives you a clear picture of what's possible.