Early Warnings

Everyone has already been warned too much

The instinct, once a building is instrumented, is to send everything to somebody. It feels responsible. It produces a channel that people mute within three weeks, and after that the system is worse than the clipboard it replaced — because now there is a record that everyone was informed, and a shared understanding that nobody reads it.

We take the opposite starting position: a warning has to earn an interruption. Most of what risk detection surfaces should quietly become a task on a schedule, seen by the person who will do it and nobody else.

Four audiences, four different thresholds

The cleaner on shift gets what changes the next hour of their work: an area that moved up the list, a supply run that needs to happen first. Delivered through CleanPilot by voice, because someone in gloves is not going to read a push notification.

The supervisor gets what needs a decision: a risk their account is carrying, a task that missed its window, an inspection that scored badly. This is the role most warnings are aimed at, because it is the role that can actually reallocate people.

The manager gets pattern, not incident. The same failure recurring across accounts, or a site whose scores are trending the wrong way — things no single warning shows.

You get what you asked to see. For most clients that means conditions they would otherwise discover through a complaint, plus anything the operational chain failed to close on time. It is deliberately the narrowest tier.

Escalation on a timer

The failure mode of every notification system is the warning that is delivered correctly and then acted on by nobody. It is a hard failure to see, because the log looks perfect.

So acknowledgement is timed. Each severity has a window; if the window passes without someone taking it, the warning climbs to the next role and the delay is attached to the record. The client tier exists partly for this reason — one route to your inbox is the escalation path of last resort, which means a warning our own operation dropped ends up visible to you rather than buried.

That is uncomfortable by design. A system that only surfaces its successes to the client is marketing.

Quiet hours, and the exception

Non-urgent warnings respect the hours a site actually keeps. A dispenser projected to run out by Tuesday does not need to wake anyone on Saturday night.

Urgent ones ignore all of that, and route to whoever is genuinely on duty — the operations line runs 24/7 precisely so overnight and weekend conditions have a real destination rather than a queue that opens Monday. A spill in a facility running third shift is not a Monday problem.

Warnings that went nowhere

We keep the ones nobody acted on. Over a quarter, that number says more about whether this is working than any delivery statistic: if warnings are routinely sent, escalated, and still closed late, the threshold is wrong or the tier is wrong, and both are adjustable.

It is the same principle as everywhere else in the proof of service — the record has to be able to show the operation failing, or it cannot be used to show it working.

Where they come from and where they land

Warnings are generated by risk detection and delivered through smart alerts in the platform. Whatever they trigger becomes either a scheduled visit through predictive cleaning or an immediate proactive response.