Patrol Robots and Repeat Victimisation
At a glance: Most security robot business cases are written on labour substitution and are rejected, because a guard post is rarely deleted. The case that survives is built on repeat victimisation: the well-documented finding that a small share of sites and a small share of time windows account for a large share of incidents. This guide sets out the concentration arithmetic, the detection and deterrence evidence a patrol unit actually produces, and the incident-data baseline a buyer needs before the first unit goes out on route.

Repeat Victimisation: The Security Fact That Changes the Buying Case
Repeat victimisation is the finding that crime concentrates: a small proportion of sites and of time windows accounts for a large proportion of incidents, and a site that has been victimised once has a materially higher probability of being victimised again in the following weeks. It has been documented across residential burglary, commercial burglary, retail theft, vehicle crime and school and transit settings, and it is the reason targeted patrol is more effective per hour than uniform coverage.
It is also the reason a patrol robot can be justified commercially when a headcount argument cannot. A guard post is a fixed, visible, continuous presence. A patrol unit is an intermittent, scheduled, record-producing presence. Intermittency looks like a weakness until it is aimed at concentration, at which point it becomes a cost advantage: the same number of patrol hours can be spent on the windows and locations where incidents already cluster, and the record produced at each visit is the evidence an insurer and a client both want.
The trap is that most buyers have no incident baseline. Without one, the robot is dispatched to cover everything at low frequency, which is the least effective use of intermittent patrol and produces no measurable improvement. Establishing the baseline is a half-day exercise and it is the first thing to do.

Building the Concentration Baseline
Twelve months of incident data, even from a paper log, is enough to find the pattern. The table shows the shape the analysis takes on a typical mixed estate of one office block, one retail unit, one car park and a service yard.
| Analysis | What to compute | Typical finding |
|---|---|---|
| Location concentration | Share of incidents in the top 20 percent of named locations | 55 to 75 percent |
| Time concentration | Share of incidents in the top 3 hours of the week | 30 to 45 percent |
| Repeat rate | Share of locations with two or more incidents in 12 months | 20 to 35 percent |
| Repeat interval | Median days between first and second incident at the same location | 18 to 45 days |
| Type concentration | Share belonging to the two most frequent incident types | 60 to 80 percent |
Two numbers from that table drive the patrol design. The repeat interval sets the revisit frequency: if the median second incident arrives 26 days after the first, a location that has just had an incident needs elevated presence for roughly four weeks, not for the rest of the year. The time concentration sets the roster: patrol hours outside the top windows have much lower expected value per hour.
The baseline also settles the argument about what the unit is for. If 65 percent of incidents sit in six locations, the specification is not "patrol the estate"; it is "revisit these six locations on this schedule, record what changed, and escalate on exception". That is a specifiable, testable, and priceable scope.
Deterrence: What a Visible Unit Actually Changes
Deterrence claims are where security technology business cases usually overreach. The honest position is narrower and more useful.
A patrol unit changes the perceived risk of an offender in three ways: it makes presence predictable in some windows and unpredictable in others, it records, and it produces a response. The recording is the largest measurable effect in most commercial settings, because an incident that is captured with a timestamp and a location is far easier to investigate, claim on and prosecute than one discovered the next morning. The recording value is also the easiest to show to an insurer, since it converts an unverifiable narrative into a document.
What a patrol unit does not reliably do is replace the deterrent effect of a person who can intervene. Buyers should separate three outcome categories in the business case and not blend them: incidents prevented (hard to attribute), incidents detected earlier (measurable, via time-to-discovery), and incidents better evidenced (measurable, via claim and recovery outcomes). Only the last two survive scrutiny, and in most deployments they are enough.
The most defensible public-sector formulation is the one used in crime-prevention-through-environmental-design practice: surveillance increases the perceived risk and the effort required, and its measured effect is strongest when paired with access control and lighting rather than used alone. A buyer writing an internal case should say exactly that, and should avoid claiming an arrest-rate improvement the data cannot support.

Scheduling Patrol Hours Against Concentration
Once the baseline exists, patrol hours become an allocation problem with a simple arithmetic. Take a 7-day week, a 10-hour night window and a unit that can cover 4 km of route per hour at survey speed. That is 70 hours and roughly 280 km of route per week. The question is where those kilometres should go.
| Allocation | Share of patrol hours | Route choice | Expected effect |
|---|---|---|---|
| Surveillance of repeat locations | 40 percent | Six named locations, two visits per night in the top windows | Highest value per hour: revisits inside the repeat interval |
| Perimeter and access points | 25 percent | Gates, doors, shutters, service yard | Detects forced entry and tailgating evidence |
| Car park and open areas | 20 percent | Bay aisles on a varied sequence | Vehicle crime coverage, unpredictable timing |
| Rest of estate, low frequency | 10 percent | Full route sweep once per night | Evidence of presence, minimal detection value |
| Reserve and exception response | 5 percent | Diverted to alarms and escalated events | Keeps the schedule honest when incidents occur |
The sequencing matters as much as the allocation. A predictable route invites timing. A unit that visits the top-window locations at a varied but bounded interval keeps the revisit frequency inside the repeat interval while removing the "wait until the robot has passed" behaviour that a fixed schedule permits. Route randomness should be bounded rather than uniform: each repeat location visited at least twice per night, with visit order and offset varied.
The roster also has to respect the guard handover. Detection without response has limited value, so the patrol windows should be aligned to the periods where a response is actually available, whether that is an on-site officer or a remote monitoring desk. The handover and escalation design is set out in the patrol robot security rounds guide, which also covers the night, perimeter and car-park duty split in more detail.
Data Quality: The Part That Decides Whether the Case Survives
A patrol programme is only as good as its incident records, and robot-generated records are easy to inflate. Three quality rules keep the data usable.
- One incident, one record. A single forced door detected on four patrol passes is one incident with four observations, not four incidents. Systems that log every detection as a new event make the prevention effect look large and destroy trust in the report.
- False-positive accounting. Every detection category needs a measured false-positive rate, expressed per hundred observations. A lighting-change alarm that fires eleven times a night will be ignored by month two, and the operator will believe the system does not work even where it does.
- Time to discovery, not counts alone. The metric that captures value is the interval between incident occurrence and human awareness. A programme that cuts median time to discovery from seven hours to forty minutes has demonstrably changed outcomes even if the incident count is unchanged.
The same discipline applies to the evidence itself: retention window, chain of custody, and who can export a clip. Those are contract questions and they are set out with the insurance view in the insurance evidence pack guide and the insurance claims evidence guide. Privacy obligations for recorded footage, including when a data protection impact assessment is triggered, are covered in the GDPR DPIA walkthrough.

The Commercial Case in Five Lines
A security budget case that survives finance needs five numbers, all derived from the baseline rather than from a vendor claim.
| Line | How to derive it | Worked example |
|---|---|---|
| Incident cost baseline | Incidents per year x average cost per incident, including excess, loss and staff time | 34 incidents x 780 = 26,520 per year |
| Addressable share | Share of incidents at repeat locations inside the patrol windows | 58 percent = 15,380 per year |
| Assumed reduction | Conservative, stated as a range, applied only to time to discovery and evidence quality | 20 to 30 percent of addressable = 3,076 to 4,614 |
| Programme cost | Unit, docking, monitoring integration, service contract, staff time to run the schedule | 9,800 per year over three years |
| Net position | Reduction minus programme cost, plus the insurance effect if the insurer credits verified patrol records | Negative alone; positive only with the premium effect or the avoided excess |
Read the last line honestly. On incident reduction alone, a single patrol unit rarely pays for itself at small estate scale. It pays when the programme also reduces insurance exposure, avoids a claim excess on one or two events, or protects a contractual service level owed to a tenant. Those are the lines that belong in the case, and the ones that are usually left out because they sit in a different budget owner's spreadsheet.
For multi-site operators the arithmetic changes because one unit can be shared across sites with the schedule designed around each site's own concentration pattern. The multi-site deployment strategy guide covers the sharing model and the utilisation assumptions behind it.

What to Measure in the First Ninety Days
A pilot that runs for a quarter should be judged on four measures, agreed before it starts and taken from the same baseline as the business case.
- Revisit compliance: share of scheduled visits to repeat locations actually completed, with the completion timestamp.
- Time to discovery: median and 90th percentile interval between occurrence and human awareness, for all incidents in the pilot area.
- False-positive rate: per alert category, per night, tracked weekly so that category tuning is visible.
- Evidence usefulness: share of incidents where the patrol record was used in an insurance notification, a recovery or a disciplinary process.
These four are chosen because they are attributable to the programme rather than to the season, the estate's footfall or the local crime rate. If they improve while the incident count stays flat, the case is still defensible on evidence quality and response; if they do not improve, the schedule is wrong and no amount of additional units will fix it.
How AOMAN Supports Patrol Deployments
AOMAN, whose core platforms are the D1 delivery, C1 and C2 Pro cleaning, and G1 reception units, works with security integrators on patrol deployments rather than selling a patrol appliance directly. That means the schedule design, the escalation path and the incident-record schema are built with the integrator who already holds the guarding contract, and the platform exposes the telemetry and event data the monitoring desk needs. The security and surveillance in commercial facilities guide covers the wider deployment context, and the technology page describes the navigation and event-recording architecture the deployment depends on.
