5.3 Strategies & Advancements
Overview: Strategies and Advancements
A strategy is the set of decisions you make before the money is spent, so that in year three someone can still tell whether the system worked. This lesson covers four of those decisions for an AI camera and access program. They are what you measure and when you start, how you service a system that can change between visits, what five years really costs, and what your crew has to know.
Lesson 5.3: Introduction
By the end of this lesson, you will be able to do the following.
- Take a baseline before AI features are turned on.
- Spot a statistic moved onto the wrong system.
- Plan service for a system that can change between visits.
- Include the costs a product quote leaves out.
Without a "before," there is no "after"
A customer will eventually ask whether the AI made a difference. You can only answer if you counted before it was turned on.
The rush
An apartment complex turns on person detection on 40 cameras to hit a move-in date. Nobody counts alarms first.
The question
A year later the owner asks if the analytics cut false alarms. There is no before number to compare against.
The fix
Count alarms per camera per day for two weeks with the rules off. Write down the firmware and model version the day you turn them on.
No Published Scorecard for Physical Security
A brochure can say the analytic cuts false alarms by 90 percent. Nobody in this trade has published a shared way to check that claim on your customer's site.

Where a "it works" number can come from
When a customer asks if the cameras are worth it, they want a number: fewer false alarms, faster response, operators catching more. There is no independent scorecard for that. ASIS, SIA, and the international standards catalogs do not publish a key performance indicator set for physical security operations. If the proposal still has a percentage in it, that percentage came from somewhere else.
Your job is to name the somewhere else before you repeat the number. A figure from a monitoring standard, a figure from a study with a known group of operators, and a figure from the company selling the analytic are not the same kind of evidence. The three sources below are the ones this lesson uses.
ANSI/TMA-AVS-01 (2024), alarm scoring
This is the one ANSI-accredited standard in this set. The Monitoring Association wrote it so a monitoring center can score a burglar-alarm activation and tell the dispatch center how strong the call is, from no dispatch up to a confirmed threat. It is a real published method. It scores alarm calls. It does not score whether a camera analytic works. Standard page: ANSI/TMA-AVS-01.
A peer-reviewed operator study, with the group named
Donald, "Work exposure and vigilance decrements in closed circuit television surveillance," Applied Ergonomics, 2015. Working CCTV operators watched a 90-minute video and caught about half of the target behaviors. False alarms were high. That is a citable description of the attention problem analytics are sold to fix. It is not a test of any product you install. Paper: Applied Ergonomics, 2015.
The vendor's own KPI sheet
The twelve items below show up on analytic vendor sheets. They are a fair list of what a site can count for itself. The rates printed next to them were counted by the seller, usually on a different site, with a denominator the sheet does not show. That is a sales claim, not a result for this job.
Count these on this site
Pick the items you and the customer will actually count. Count them from this site's own alarms and this site's own operator log, and start before the analytic is turned on. Do not copy the rates off the vendor sheet into the report.
A proposal line such as "90 percent fewer false alarms" is the seller's claim about their product. AVS-01 only rates how strong a burglar-alarm call looks before someone is dispatched. The 2015 study only shows that operators watching a test video missed about half the staged events. Neither document measured the analytic on this customer's cameras. Before you repeat the 90 percent, write down which alarms were counted, what counted as false, and whether the count happened on this site.
Take a Baseline Before Commissioning
Count alarms for two weeks while the cameras only record. Then turn on the detection rules, and write down the firmware and model version on the day you do it.

The analytic is the detection rule, not the camera
The analytic is the software that decides a person, a vehicle, a line cross, loitering, or a face match happened, and then creates an event. It can sit in the camera, in the recorder, or in the cloud account. Turning the camera on, or turning recording on, is not the same step. "Switched on" means those detection rules start creating events the operator has to handle.
Count the site before that switch. For two weeks, with cameras recording and the detection rules off, log alarms per camera per 24 hours and what the operator did with each one. That is the before picture. On the day the rules are enabled, write down the firmware version and the model version. A year later you compare the same counts. Without the before picture, year two is just a number, and you cannot say the detection rules caused it.
"Before we turn on person detection, vehicle detection, or the door match, we spend two weeks counting what happens now. Alarms per camera per day, and what your operator did with each one. That is the before picture. Without it, in a year we will both be guessing."
Customers rarely refuse this. It costs almost nothing, and it is the only record that can show the detection rules changed anything.
Year two has a number and no before picture
If the detection rules go on with no count behind them, the honest answer in year two is "we cannot tell." That is a hard sentence to say to someone who paid for the analytic.
False Camera Events Are Counted Here
A false burglar-alarm rate is not a false person-detection rate. Count the events this camera and this door actually produce.

What a false event is on this job
On a camera, a false event is a detection rule firing when the thing it was set to catch is not there: headlights scored as a person, rain scored as motion, a shadow scored as a loiter. On a door, a false event is a face match or other rule granting or alerting when the person was not the one the rule was built for. Neither of those is a burglar-alarm call, and neither one has a national percentage you can paste into a proposal.
Use the two-week baseline from the previous section. Before the detection rules are turned on, count what the operator already handles. After the rules are on, count events per camera per day and mark each one real or not, from the clip. The improvement is the change on these cameras, not a figure from a police study of alarm calls.
Do not quote a burglar-alarm false rate as proof the analytic works. If the proposal needs a number, it has to be this site's events, with the rule name and the model version written on the count.
A Real Percentage Moved Onto the Wrong System
A camera-alert rate and a door-match rate are not the same number. Here is how a real figure gets pasted onto the wrong one.
A vendor page makes three claims about a camera analytic. No camera count, no site, no time period, and no outside source is given for any of them.
More than 98%
"Physical security alerts face a more than 98 percent false alarm rate."
86%
"86 percent of users see a return on video analytics within one year."
More than 80%
The product "resolves more than 80 percent of alerts in under a minute."
The 98 percent is not a camera number
Search that 98 percent and you can find a real source. The source is police responses to burglar-alarm calls around 2002, when about 94 to 98 percent of those calls were false. That is a different system from the one you install.
A false event on this job is a person rule firing on headlights, or a face match releasing the wrong door. Different device, different rule, different site, different decade. The number was not invented. It was lifted off burglar-alarm calls and pasted onto camera alerts. Once it moves, it no longer describes your customer's cameras or doors.
Where the 98 percent came from
- Burglar-alarm calls
- That police responded to
- Around 2002
- Not a camera rule and not a door match
Where the page put it
- Camera alerts in a security office
- Person, vehicle, and loiter rules
- No site named
- Treated as proof the analytic is needed
The number was not made up. It was moved. A search will find a real source. Read what that source counted. If it was not these cameras, or these doors, do not put it in the proposal.
Who counted it, how many, percent of what
Who counted it? A vendor test of its own camera, a vendor-paid writeup, and a trade award are not an outside test of this door or this parking lot.
How many? How many cameras, how many doors, how many days. If none of the three is on the page, the percentage is a claim.
Percent of what? Alerts from a person rule, grants at a door, or something else entirely. If the denominator is not the same device you are installing, the rate does not transfer.
Where a Noisy Camera or Door Rule Gets Switched Off
A false person alert or a bad door match does not, by itself, pull police off a site. It gets ignored, then disabled.
What this job actually loses
City alarm permits and police false-alarm fines apply to a monitored burglar-alarm dispatch. That is a monitoring function, and it is not this course. A camera rule or a door match is not on that count unless someone has separately tied the event to a monitored alarm call. Do not tell a customer that a noisy person rule will cost them a police response.
What does happen is simpler. The operator gets too many false camera events, or the door match grants the wrong people, and someone turns the rule off. The camera still records. The door still locks. The analytic you installed is no longer in the path. Nobody calls you when they do it.
- Count false camera events and bad door grants before the rules go on, then again after.
- Fix the scene causes you can see: headlights, a moved camera, a dark vestibule, vines in the zone.
- Write the rule name and the firmware version on the count, so a later jump can be tied to a change.
- Ask the customer at each service visit which rules are still enabled. A rule that was on at handover and off at the next visit is the finding.
A rule nobody checks becomes a rule nobody trusts, and a rule nobody trusts gets switched off, usually without anyone telling you.
The baseline and the service chart exist so the camera rule and the door match are still on, and still doing what you handed over, a year later.
The rest of the plan, decision by decision
The cards above cover the first decision, what to measure and when to start. Here are the other three, service, five-year cost, and your crew.
Service+
The decision. How often to come back, and what to check.
What the evidence says. No published standard sets a service interval for cameras or AI rules. Seasonal re-checks and re-running the held-back test clips are good practice, but nothing requires them. A control chart tracks nuisance alarms per camera per day, with updates marked. A sudden jump on the day of an update points to a model problem, called a regression. A slow climb points to the scene changing, such as vines or a new light.
What you do. Write your interval into the service agreement as your own, not a code interval. After a sudden jump, re-run the test clips and call the manufacturer.
Five-year cost+
The decision. What the customer will really spend over five years.
What the evidence says. Re-testing the AI after each update is labor that no standard requires, no vendor pays for, and almost no contract prices. Most vendors do not publish AI prices, but city purchases are public record. One city contract for 90 license plate cameras came to about $3,000 per camera per year, a pure subscription.
What you do. Put re-verification labor in the budget as its own line, and use public contracts to understand how these systems are sold.
Your crew+
The decision. What your technicians need to know, and how to show it.
What the evidence says. A large yearly AI skills study counts information-security staff, not physical security technicians. No study with a stated method measures an AI skills gap in this trade. A company certificate to the AI management-system standard, or a strong vendor ranking in government face recognition testing, says nothing about how an installed system performs on a site.
What you do. Do not borrow cybersecurity figures for your crew, and prove site performance with a pilot that has a passing score stated in advance.
What to tell the customer
A hotel general manager is reviewing your service proposal.
Go deeper (optional)What to Measure, and What the Denominator Is+
What to Measure, and What the Denominator Is
The right-hand column tells you what you can cite when someone pushes back.
Here is the set.
| Metric | Definition and denominator | Where it comes from |
|---|---|---|
| Probability of detection | True positives divided by ground-truth events, per scenario and per condition, with a stated confidence interval | Named in the federal criteria document; framed by the national lab test manual; sized by the metrology technical note |
| Recall | Detections divided by ground-truth alarm events | Standard detection math (see Lesson 1.4) |
| Precision | True alarms divided by all alarms the system raised | Same source |
| F1 with declared recall bias | A single score that weights recall against precision by a factor you declare from what the alarm is used for (live alerting or event recording) | Same source |
| Nuisance alarm rate | Alarms from a real but non-threat stimulus, divided by camera-days, stating whether that is calendar days or armed hours | Named distinctly in the federal criteria |
| False alarm rate | Alarms with no external cause at all, divided by camera-days | Same, and it is a different metric from the line above |
| Missed events | Events known from an independent channel (guard log, police report, insurance claim, HR case) that the analytic did not alarm on, divided by all such events | Good practice rather than a standard. No standard requires it, and it is the metric a customer cares about most |
| Alarm volume per operator per shift | Raw count. This is the workload denominator | Vendor taxonomy |
| Alarm level distribution | Share of dispatched calls at each validation level, and the share escalated on machine evidence alone | The ANSI-accredited alarm scoring standard |
| Quality assurance sample compliance | At least 50 events per 90 days; corrective action required below 80 percent | Same standard |
| Time to acknowledge and time to resolve | Alert to acknowledgment; acknowledgment to closure | Vendor taxonomy |
Two rows where practitioners lose arguments
Nuisance alarm
A real cause that is not a threat, such as a cat, a flag, or headlights on wet pavement.
You fix it with geometry, scheduling and classification.
False alarm
No external cause at all. It is equipment or software.
You fix it by replacing something.
Count Missed Events From Another Source
It requires an independent channel. You have to learn about events some other way and check whether the system caught them. Use guard logs, police reports, insurance claims, and HR cases. It is the only metric that shows the failure the customer actually fears, and no standard requires it.
Tune for What the Alarm Is Used For
This is the most useful tuning rule the industry has. The same analytic should be tuned differently depending on what a human does with the output.
Live operator alerting
If an operator has to look at every alarm right now, precision matters enormously. A noisy system burns a person out and gets switched off.
Event recording for later search
If the output is only used to mark recordings for later search, recall matters far more and you can tolerate a great deal of noise, because nobody is being paged.
Same camera, same model, different correct setting. Ask the vendor which of these two uses the default setting is tuned for.
Go deeper (optional)Service Intervals for Cameras and Access+
Service Intervals for Cameras and Access
No standard sets how often to recheck a person rule or a face match. The scene and the model can both change between visits.

What you can cite, and what you cannot
No published standard has been identified that sets a maintenance interval for video surveillance or for a detection rule. The international video standard series covers planning, installation, testing, commissioning, and maintenance. It does not tell every business to come back every six months. Annual service on a commercial camera system, and a shorter visit on a higher-risk door, is trade practice. No standard is behind that interval.
That gap matters on this work because a person rule or a face match can change without a truck roll. A firmware push can change what the camera calls a person. A vine, a new light, or a two-degree pan can move the zone. A reader firmware change can change who the door match accepts. Write the interval into the service agreement, and tell the customer it is your interval, not a code interval.
A camera that still records is not proof the person rule still works. A door that still beeps is not proof the face match still grants the right people. Recheck the rule against the live scene, not just the cable and the lens.
What the 2025 video application guidelines add
The second edition of the international video application-guidelines standard brings maintenance into scope. Trade summaries, including one by the project lead, say it supplies the four items below. Buy the standard and read it before you treat any of this as a required interval. The summaries do not confirm a numeric service period.
Three kinds of maintenance
Preventive, corrective, and improvement work, written as separate tasks.
Checklists
Visual inspections, functional checks, and service checks.
A lifetime operations log
Faults, modifications, maintenance, and changes, kept for the life of the system.
Three skill levels
Instructed person, competent person, and competent system engineer.
What a defensible camera and door program contains
Each item is labeled. If no standard requires it, say so in the agreement. That is what makes the program defensible.
Camera scene checks (no published interval)
- Clean the dome, lens, and infrared window. Clear webs, nests, and vines. Check the housing seal.
- Confirm the camera still points where the as-built says. A small pan moves the detection zone and changes every object the rule sees.
- Check focus, illuminator output, and glass. Note anything new in the scene: a light, a sign, a tree.
Recheck the camera rules (no standard requires it)
- Walk every zone, line, direction, and object filter against the current scene.
- Check that a person at the far edge of the zone is still large enough for the rule the design assumed. If the camera was zoomed or moved, the rule is outside that design.
- Run the same test clips you held back at commissioning.
- Write the firmware and analytic version in the log. If the model version is not shown, write that you could not get it.
Recheck the door match (no standard requires it)
- Confirm a known enrolled person is granted, and a person who should not pass is not.
- Check the reader, the camera aimed at the door, and the light on that face. A lobby light change is enough to break a match that worked in June.
- Write the reader firmware and the match-rule version in the same log as the cameras.
A second visit for outdoor cameras (no standard requires it)
A June visit does not prove the person rule still works in December, in low sun, with headlights and wet pavement. Two seasonal checks for outdoor cameras is defensible. Tell the customer no standard requires it and no vendor guarantees it.
Count false camera events per camera per day, and false door grants, on a simple chart. Mark the firmware or model version on the date it changed.
A jump on a firmware date is a model change. A slow climb with no version change is the scene: dirt, vines, a new light, a moved camera. Those are different fixes. You cannot tell them apart without the chart and the version note.
Schematic only. These are the shapes to look for, not real data.
Put the interval in the service agreement. No camera or access-control standard sets it. The chart is how you show a later visit whether the rule changed or the scene did.
Go deeper (optional)Five-Year Cost, Including the Lines Nobody Quotes+
Five-Year Cost, Including the Lines Nobody Quotes
Vendors in this industry rarely publish list prices for AI analytics, and some say so outright.

Two major video management vendors describe their support and upgrade programs in detail on their own sites, one naming a 15-minute phone first-response service level and nine languages, and neither publishes a price. A cloud vendor's price-change announcement gave its actual percentages only inside an image, not as readable text.
The best verifiable public prices in this field come from government contracts, because municipal purchases are public record.
One public contract, read closely
A city council in September 2026 approved a renewal, on a five-to-three vote, of an automated license plate reader contract. The same department announced it would reduce standard data retention from 30 days to seven, with longer preservation where information is evidence. No performance metrics were disclosed.
This one contract shows the pricing model more clearly than any market chart. It is a pure subscription, per device, per year, on a one-year term, with no hardware asset on the customer's books, no performance guarantee, and a retention period the customer changed by policy rather than by engineering. That is how most of these systems are now sold. We can see it only because the customer was a city.
Market analyst figures
The principal market database for this sector is a paywalled subscription product with no public statement of method. It gives no sample frame, no revenue attribution rules, no unit definitions, and no statement of how "analytics-capable" is defined. Do not put such a number in course material or a proposal unless you hold the license and can read and cite the methodology note.
Note the difference that matters most. "percentage of cameras shipped with AI capability" and "percentage of cameras with analytics actually enabled and tuned" are wildly different numbers, and nobody publishes the second one.
The budget itself, in two lists
The first list is what appears in a quote. The second is what does not.
Recurring by contract
- Per-camera or per-device annual subscription.
- Software maintenance or support agreement, which is usually the gate for version upgrades and new capabilities.
- Cloud storage and egress where retention is cloud-side.
- AI inference cost, now explicitly passed through. One vendor named AI, memory and storage costs in a 2026 list increase.
- Per-feature or per-analytic licenses on top of the base device license. Structure varies and is not publicly comparable across vendors.
Recurring by necessity, usually unbudgeted
- Regression re-verification after every model or firmware update. No standard requires it, no vendor funds it, almost no contract prices it.
- Seasonal re-commissioning. Two visits a year for outdoor analytics, required by no video standard.
- Cleaning and optical preventive maintenance, at an interval no standard specifies.
- Scene-change reconfiguration driven by the customer's own operations, such as new fencing, new parking layout, new lighting, a new loading bay. Not a fault, and it invalidates a commissioned zone.
- Operator time on nuisance alarms. The cost the AI is supposed to reduce, which is why the baseline matters.
- End-of-support-driven replacement. Where a vendor publishes lifecycle windows, check whether five years crosses them. Where a vendor publishes nothing, the prudent assumption is worse, not better.
- Replacement forced by a rule change. A federal rule, such as NDAA Section 889 on federally funded work, can force a camera brand out. Contingent, unquantifiable, and real.
How to Build a Five-Year Budget in Five Steps
Separate the two lists
Separate the contractual costs from the necessary-but-unpriced ones, using the two columns above.
Get end-of-support in writing
Get end-of-support in years, in writing, per product line, and check whether the five-year window crosses it.
Price re-verification labor
Price the re-verification labor as an explicit line item, using a trial count derived from the target detection probability and confidence level. That is what makes it defensible rather than a guess.
Cap the escalation
Assume subscription list prices can rise mid-estate. Get the escalation cap in writing. Where a vendor commits that previously purchased licenses will not increase during their term, that is an argument for longer terms.
Say there is no benchmark
Tell the customer that no published, methodology-bearing total-cost benchmark for AI physical security exists. Any five-year model is a build, not a lookup. Saying so is what makes your build credible.
Go deeper (optional)The AI skills gap nobody has measured+
The AI skills gap nobody has measured
No published data with a stated method has been found on an AI skills gap in the physical security industry.

The manufacturers' association publishes a compensation and benefits study conducted by an outside firm for a talent foundation in partnership with several trade groups, a careers guide, a hiring guide, and several workforce programs. Its workforce development material contains no figures on labor or skills gaps with documented methodology.
What does exist is cybersecurity workforce data. A large annual study had a 2025 theme that skills matter more than headcount in an AI era. That is an information-security workforce study. It is not a study of physical security technicians, and its numbers must never be presented as physical-security figures. Watch for slides that do this. They are common.
The honest version is stronger than the borrowed statistic. This industry has compensation data, careers guides and hiring guides, and no published measurement of whether its technicians can commission, tune, re-verify or govern a machine-learning system. The lack of any baseline is the best argument there is for training. Say plainly that the data does not exist.
AI certifications that exist
Two AI governance credentials exist and neither is physical-security-specific.
- One is marketed as the first AI-centric security management certification, with three domains covering governance and program management, technologies and controls, and risk management. It requires an existing information-security management or information-systems-security credential as a prerequisite, which effectively excludes the field technician population.
- The other covers foundational AI knowledge, applicable laws and frameworks, the lifecycle and responsible governance, and emerging issues.
Both are management credentials. Lead implementer and lead auditor courses against the AI management-system standard are training-provider certificates.
What the Crew Needs to Be Able to Do
Each item fills a gap covered in this course.
Reading a rate with its denominator
Because the industry routinely detaches a real figure and reattaches it to a different population. This is the most valuable skill on the list.
Sample size and confidence
Because the published tables give the answer and nobody in the trade uses them. A salesperson who can say "your demo of twelve supports no claim" is worth money.
Writing acceptance criteria before the pilot
Threshold, confidence, denominator, recall bias, pass arithmetic.
Pixel density on target as a design constraint
Re-verified at every service visit.
Recording versions
Firmware, analytics application, and model where obtainable, in a lifetime operations log, with the gap documented when it is not obtainable.
Regression re-verification
Against a held-back blind set after every update. Nothing requires it; nothing else detects a silent regression.
Reading end-of-support dates
Reading a lifecycle policy in years and mapping it onto a five-year budget.
Alarm scoring levels
The levels, machine-evidence escalation, the 12-month retention, the 90-day and 50-event sample, the 80 percent threshold.
Knowing what a certificate means
An AI management-system certificate attests governance, not detection performance. An algorithm ranking in a government evaluation ranks a submitted algorithm, not your installed system. Neither is a product approval.
Knowing what the law limits
Some features, such as face matching, carry legal limits that differ by state and city. Ask the client's counsel before you sell them.
Key Terms
Click any term to reveal its definition.
Baseline+
The measurement taken before the analytic is enabled. It is typically alarms per camera per 24 hours with operator disposition, over about two weeks. Without it no later change can be attributed to anything.
Probability of detection+
True positives divided by ground-truth events, stated per scenario and condition with a confidence interval. This is the main performance metric to use.
Nuisance alarm rate+
Alarms caused by a real but non-threat stimulus, per camera-day. Fixed by geometry, scheduling and classification.
False alarm rate+
Alarms with no external cause at all (equipment or software), per camera-day. Fixed by replacing something. A different metric from nuisance alarm rate.
Recall and precision+
Recall is detections over ground-truth events; precision is true alarms over all alarms raised. They trade against each other, and which one to favor depends on what a human does with the output.
Recall bias+
A declared weighting factor applied when combining recall and precision into a single score. Same analytic, different correct tuning.
Missed events+
Ground-truth events learned from an independent channel (guard log, police report, claim, HR case) that the analytic did not alarm on. The metric customers care about most and no standard requires.
Analytical Data Confirmation+
In the alarm validation scoring standard, an automated determination that an event warranting a call for service is probably occurring. It explicitly contemplates analytics output that no operator ever saw.
Verified response+
A policy requiring confirmation of an alarm before police dispatch. In peer-reviewed analysis it cut police alarm responses by roughly 87 percent a year and was associated with a 26 percent annual reduction in burglaries.
Alarm permit+
A city or county registration for an alarm system, kept with a current contact list. Under a local alarm ordinance, false alarms are counted against it. Too many bring fines and can cost the site its permit and police response.
Enhanced Call Verification (ECV)+
The monitoring center makes two or more calls, first to the premises and then to another contact, before it asks police to respond. The monitoring association standard TMA CS-V-01 calls it enhanced confirmation. Hold-up alarms are not verified first.
ASAP+
The Automated Secure Alarm Protocol. It sends alarm data from the monitoring center straight into the 911 center's dispatch computer, and it can carry the alarm level with the call.
Control chart with version annotation+
Tracking nuisance alarms per camera per 24 hours over time with firmware and model versions marked. A step change on an update date is a regression; a slow trend is scene or optical drift.
System operations log+
A lifetime record of faults, modifications, maintenance and changes for a video system, introduced by the 2025 edition of the international video application-guidelines standard.
Regression re-verification+
Re-running a held-back blind test set after a model or firmware update. Required by no standard, funded by no vendor, and the only way to detect a silent performance loss.
Necessary-but-unpriced cost+
A recurring cost that appears in no quote, such as re-verification labor, seasonal re-commissioning, optical maintenance, scene-change reconfiguration, and end-of-support replacement.
End-of-support window+
The published period, in years, for which a vendor will supply software and security fixes after a product is discontinued. Where nothing is published, the prudent planning assumption is worse, not better.
Regression+
References
+
- Aragon, A., Baum, G., Mack, T., Russell, J. R., & Stromberg, B. (2021). Security technology testing and evaluation manual (SAND2021-14561 TR). Sandia National Laboratories. https://www.sandia.gov/
- Blackstone, E. A., Hakim, S., & Meehan, B. (2020). Burglary reduction and improved police performance through private alarm response. International Review of Law and Economics, 63, 105930. https://doi.org/10.1016/j.irle.2020.105930
- City of Gulf Shores. (n.d.). Alarm permit [Alarm permit packet, Ordinance No. 356]. https://www.gulfshoresal.gov/DocumentCenter/View/71
- City of Montgomery Police Department. (n.d.). False alarm reduction program. https://montgomeryal.gov/government/city-government/city-departments/police/false-alarm-reduction-program
- False Alarm Reduction Association. (2024, September 20). FARA guide to alarm ordinances. https://faraonline.org/wp-content/uploads/2025/01/FARA-Guide-to-Alarm-Ordinances-9-20-24.pdf
- International Electrotechnical Commission. (2025). Video surveillance systems for use in security applications: Part 4: Application guidelines (IEC 62676-4:2025, Ed. 2.0). https://webstore.iec.ch/
- International Organization for Standardization. (2023). Information technology: Artificial intelligence: Management system (ISO/IEC 42001:2023), Clause 9.1. https://www.iso.org/standard/81230.html
- ISC2. (2025). 2025 ISC2 cybersecurity workforce study. https://www.isc2.org/research
- Leber, D. D., Pibida, L., & Enders, A. L. (2019). Confirming a performance threshold with a binary experimental response (NIST Technical Note 2045). National Institute of Standards and Technology. https://doi.org/10.6028/NIST.TN.2045
- National Fire Protection Association. (2026). Standard for the installation of premises security systems (NFPA 731). https://www.nfpa.org/codes-and-standards
- National Institute of Standards and Technology. (2026, September 3). Face Recognition Technology Evaluation (FRTE) 1:N identification. https://pages.nist.gov/frvt/html/frvt1N.html
- Sampson, R. (2007). False burglar alarms (2nd ed., Problem-Specific Guides Series No. 5). U.S. Department of Justice, Office of Community Oriented Policing Services. https://popcenter.asu.edu/
- Security Industry Association. (2024). 2024 security industry compensation and benefits study [Conducted by Vault Consulting for the Foundation for Advancing Security Talent]. https://www.securityindustry.org/
- The Monitoring Association. (2022). Alarm confirmation, verification and notification procedures (TMA CS-V-01-2022, Version April 01, 2022). https://tma.us/standards/tma-cs-v-01-alarm-confirmation-verification-and-notification-procedures/
- The Monitoring Association. (2023). Guideline for alarm validation scoring (AVS-01) integration within Automated Secure Alarm Protocol (ASAP) (1st ed.). https://tma.us/wp-content/uploads/2024/05/TMA-AVS-ASAP-01-2023-Guide-Rev_13.pdf
- The Monitoring Association. (2024). ANSI/TMA-AVS-01 alarm validation scoring standard. https://tma.us/standards/tma-avs-01-alarm-validation-standard/
- U.S. Department of Defense. (2025, July 28). Unified Facilities Criteria 4-021-02: Electronic security systems. Whole Building Design Guide. https://www.wbdg.org/ffc/dod/unified-facilities-criteria-ufc/ufc-4-021-02
- UL. (2016, October 4). UL issues first managed video monitoring services certification for remote guarding through use of command and control [UL 827B]. https://www.ul.com/news/ul-issues-first-managed-video-monitoring-services-certification-remote-guarding-command-and
Additional resources from outside Alabama
Background only. This ordinance does not apply in Alabama.
- City of Winston-Salem. (2003). An ordinance amending Chapter 38 of the Winston-Salem City Code entitled "Miscellaneous offenses and provisions" to add a new article to regulate police alarm systems/police false alarms [Effective April 1, 2003]. https://www.cityofws.org/DocumentCenter/View/7012/Police-Alarm-Ordinance-PDF
Explore & Complete
Complete the Knowledge Check below. Navigate through each slide and answer every question. You have unlimited attempts to answer each question correctly.
Next Steps
After you have earned a perfect score on the Explore & Complete activity, click the Mark Complete button to complete this lesson. You can then move to the next lesson.

