Key takeaways
- Every exception type needs a named owner, a maximum response clock, and a documented handoff when that clock expires.
- Criticality beats fleet size: a stalled meal tray run and a paused overnight scrubber should not share the same SLA.
- Alert quality matters as much as speed; false positives train night staff to ignore the dashboard.
- Remote triage should clear simple faults before you dispatch a field engineer across town.
- Log unresolved minutes, not only rescue counts, or you will never see the true cost of slow escalation.
What should response time mean for a blocked robot?
Response time is the elapsed minutes from the moment the fleet software records a hard stop until a trained human acknowledges the fault and begins recovery. It is not mean time to repair, and it is not vendor travel time. Operators who blur those clocks end up arguing about dashboards while guests still wait on a tray or a lobby stays wet.
A useful target always names three things: who is on point for this shift, how they get paged, and what they are allowed to do without calling a supervisor. Blocked paths, localization loss, spill detections, and manual handoffs each deserve their own row in that matrix because the safe recovery steps differ.
If you cannot explain the target to a night janitor in one breath, it is too vague. Write it on the same run sheet as dock assignments and fire-watch rounds so robot exceptions feel like facilities work, not IT tickets.
How do you rank exceptions by task criticality?
Start with customer or patient impact, then physical risk, then revenue throughput. A delivery robot holding a medication tote in a clinic corridor outranks a scrubber waiting behind a movable display in a closed big-box store. A patrol unit stuck in a live vehicle lane outranks one paused on an empty gravel pad.
Use four tiers. Tier one covers people-facing moves and any stop in a fire or egress path. Tier two covers production or kitchen supply loops that will stall a line within the hour. Tier three covers overnight cleaning or inventory drones where the finish time slides but quality still matters. Tier four covers non-urgent inspection reruns you can defer to the next scheduled round.
Document the tier on the work order template your integrator imports into the fleet portal. When a new site joins a multi-building program, tiers travel with the building name so a shared remote desk does not guess.
- Tier one: acknowledge within five minutes, on-site or remote recovery begun within fifteen.
- Tier two: acknowledge within ten minutes, recovery plan within thirty.
- Tier three: acknowledge within thirty minutes, recovery before the published finish window ends.
- Tier four: batch into the next business day unless safety flags escalate it.
Who should be on call for each shift?
Match the responder to the building, not to the brand logo on the bumper. Day shifts can rely on a facilities lead who already knows which elevator needs a escort. Overnight may lean on a security officer plus a remote operations desk that can reboot maps or release a virtual zone.
Union and contract sites should spell out whether a robot rescue counts as facilities, security, or production labor before go-live. Ambiguity shows up as thirty-minute debates while a scrubber blocks a freezer threshold.
Service Robot Co. customers often pair a local on-site champion with our nationwide remote triage so simple faults clear before a van rolls. The contract should list which fault codes stay with remote support and which automatically open a field dispatch.
How should alerts be filtered before they page a human?
Raw fault feeds overwhelm small teams. Group alerts by exception class, suppress duplicates within a ten-minute window, and require a location label a human recognizes, such as Dock 3 north aisle, not only a map coordinate.
Spill and obstacle detections need photo or short video attachments when the network allows. Without context, a guard treats every ping as a drill. With context, they can decide between a real mop job and a mylar balloon.
Escalation should step through two channels before it hits a director at home. Text the primary, call the backup, then open a ticket with the integrator if both miss the clock. Publish that ladder on the wall next to the robot dock.
What belongs in remote triage versus a truck roll?
Remote teams should own map reloads, stuck virtual doors, cleared no-go zones, and battery swaps when spares are staged on site. They should not guess at brake hardware or fluid leaks over a phone.
Give remote operators a scripted checklist tied to fault codes your pilot recorded. If three steps fail, the system auto-escalates to field service with the log attached so the tech arrives with the right squeegee or lidar cover.
Track how many exceptions close remotely each month. A rising remote closure rate usually means your playbooks are maturing. A flat line after six months means alerts are still too noisy or staff still fear touching the console.
How do spills and localization loss differ on the clock?
Spill stops are often correct behavior. The response target should include how fast a human reaches the scene to verify, cordon, and release the robot, not how fast someone overrides safety. Set a maximum wet-floor dwell time if slip risk is high.
Localization loss in a remodel zone is a mapping problem. Clock the minutes until someone updates the zone or walks the unit through a re-localization path. If the same corner fails twice in a week, pause autonomy there until the map owner signs an update.
Manual handoffs for carts or trays need a staffed station within the response window. Promising a five-minute tray recovery while the nurse station sits empty is a staffing lie, not a robotics failure.
What does unresolved exception time cost?
Most fleets track rescues per week but not minutes idle. Minutes are where labor hides. A scrubber that sits for forty-five minutes in a grocery aisle may force a full overtime pass to finish before opening.
The Bureau of Labor Statistics projects about 351,300 openings per year for janitors and building cleaners from 2024 to 2034, which is a reminder that backfilling a missed overnight clean still falls on a thin pool of workers when robots stop early.
Multiply idle minutes by the shift wage band you already publish for porters or runners, then add guest complaints or late shipments. That product is the honest penalty for a vague escalation chart, even before you invoice emergency vendor hours.
How do you bake targets into contracts and pilots?
Put tier definitions and acknowledgment clocks in the service schedule attachment, not only in a slide deck. Tie spare units or loaner coverage to tier-one misses if uptime guarantees matter to the deal.
During a commercial robot demo or month to month pilot, run two planned exception drills: one spill scenario and one localization loss. Time them with a stopwatch and adjust targets before you sign a full rental term.
Review targets quarterly with building ops, security, and the integrator on one call. Rename owners when shifts change. Robots fail in the real world; programs fail when nobody owns the clock.
Where does a vendor-neutral integrator fit?
OEM portals rarely speak the same alert language across scrubbers, runners, and patrol units. A vendor-neutral integrator normalizes fault codes, staffs remote triage, and dispatches field engineers under one number.
Service Robot Co. writes exception matrices during site assessment mapping so finance, facilities, and security agree before the first autonomous shift. That document becomes the appendix your night supervisor actually opens when a phone buzzes at 2 a.m.
