How to choose an AI CCTV system for your site: a UK buyer's checklist
Last reviewed
Short answer: choose the AI CCTV system that works with the cameras you already have, detects the specific events on your risk list, and proves it on your own site before you commit. Then compare how each option handles false alarms, the shape of its detection zones, whether it learns your site, privacy, support and the trial on offer. The 11-point checklist below, with a scorecard you can save, covers each question.
Buyer's checklist11 questions to askCameras, edge or cloud, zones, false alarms, privacy, support, trial
Vendor-neutralSaveable
Alert description · this clip“Smoke rising from a parked car at the junction.”Sent by app with the clip, and by email
Ask your cameras“Show me smoke near the junction this afternoon”Find the clip, then download it
Real footage from a camera running Woodcock AI: smoke from a car at a road junction, labelled as it rises.
The checklist
What should I look for in an AI CCTV system?
AI CCTV, also called video analytics, is software that watches camera pictures and raises an alert when it recognises something that matters, such as smoke, flame or a person in a yard at night. If the term is new, start with what is AI CCTV. If you already know you want it, these are the eleven questions that separate the options.
The list is vendor-neutral: the same questions we would want any buyer to ask, including of us. Each links to a section below, and the scorecard lets you score two systems side by side.
Polygon detection areas. Zones drawn with as many corners as you need, round the exact stockpile, walkway or fence line, rather than rectangles that take in roads and plant.
Plain English. Ask your cameras questions about past footage, and describe changes to your detection design in plain English instead of working through settings.
Self-learning, site-built models. Each camera's model is fitted to its own scene, uses quiet footage and your feedback as training signal, and is backed by synthesised training data and on-site smoke tests.
End to end, and supported 24/7. Site survey, design, supply, installation, commissioning, go-live and ongoing support from one team, with support available around the clock.
Question 1
Will it work with the cameras I already have?
It should. Replacing working cameras is the most expensive way to add AI, and for most sites it is unnecessary. The right question is not whether a system is compatible in general, but which of your camera views are suitable for which detections.
Most IP cameras and network video recorders made in the last decade share their video using two open standards. ONVIF is the industry standard that lets equipment from different manufacturers find and talk to each other. RTSP is the streaming protocol that carries the live picture. AI software that connects over ONVIF or RTSP can usually work with cameras and recorders from a mix of brands. Older analogue cameras can often be brought in through an encoder or a hybrid recorder.
Compatibility is only the first step. A camera that streams perfectly may still be too far away, pointed at the wrong area, washed out by a floodlight or blocked by a stack of pallets. Ask any supplier to assess your cameras one by one, and to tell you which views support smoke and fire, intrusion or PPE detection before you sign anything.
Woodcock AI connects to existing cameras over ONVIF or RTSP, and every camera is checked in a free site assessment before anything is switched on. The guide for installers and resellers explains the same process from the installer's side.
Mixed cameras
A typical mix of existing cameras on one wall: a dome, a newer bullet and an older bullet. Select a numbered point to highlight it.
1Dome cameraDomes often cover wide areas. Fine for intrusion across a yard, though distant objects take up few pixels.
2Newer bullet cameraA modern IP camera that streams over ONVIF or RTSP usually connects directly.
3Older bullet cameraOlder or analogue cameras can often join through an encoder or a hybrid recorder.
4Junction boxCabling stays as it is. The AI connects to the recorder or the network, not to each camera.
Question 2
Should the AI run on site or in the cloud?
There are two places the analysis can happen. Both work; the right one depends on your connection, your number of cameras and how your IT team likes to run things.
Edge processing means a small server on site analyses the camera streams locally. Video does not have to leave the site to be analysed, only the alerts and clips go out, and detection keeps working through a weak or interrupted internet connection. The trade-off is a box to house, power and keep secure.
Cloud processing means camera streams are sent to remote servers for analysis. There is nothing extra to install on site and the system is easy to scale, but it relies on a steady upload connection, which is worth checking on rural yards and temporary sites.
Be wary of a supplier who offers only one and insists it suits everyone. Woodcock AI offers both, an on-site edge server or the Woodcock AI Cloud, and the choice is made per site in the free site assessment. The picker further down this page gives a quick starting point.
An edge server in a comms cabinet, beside the network switch and recorder.
Edge
What an on-site edge server looks like
An edge server is a compact box that sits in the same cabinet as your network switch and recorder. It reads the camera streams over the local network and sends out only alerts and clips.
If the internet drops, analysis on site carries on. If space or power in the cabinet is tight, cloud processing may suit better.
Question 3
Which detections do I actually need?
Start from your risks, not from a feature list. A long list of detection types is less useful than a short list that works reliably on your cameras.
Write down the events that would hurt your business most: a fire starting in a stockpile overnight, a break-in at the compound, someone walking past plant without hi-vis. Then ask each supplier which of those events it detects, on which of your cameras, and what it needs to see to do so. Ask to see real footage of each detection, not only animations.
For fire, the buyer's question is often framed as finding the best AI fire detection software. Video smoke and flame detection works on cameras already watching your yard or building and can see smoke in open areas before it reaches a ceiling detector, if there is one at all. It is an extra layer alongside your fire alarm, fire prevention plan and any thermal cameras, not a replacement. The video AI vs thermal vs on-camera analytics comparison sets out where each fits.
01
Smoke and fire
Video smoke and flame detection over stockpiles, yards, plant and buildings, on cameras you already have.
02
Intrusion
People and vehicles in zones you draw, during the hours you set, with lines across gates and fence lines.
03
PPE
Hard hats and hi-vis vests or jackets, checked in the zones where your risk assessment calls for them.
04
Vehicles
Vehicle detection in yards and entrances, where supported on your cameras.
Woodcock AI covers smoke and fire, intrusion, PPE (hard hats and hi-vis) and vehicles where supported. Which of those suit each of your cameras is confirmed in the site assessment, camera by camera.
Question 4
How does it handle false alarms?
False alarms decide whether a system is still switched on in six months. If alerts arrive for every headlight and rain shower, people stop looking, and the one alert that matters gets missed.
Older video motion detection reacts to any change in the picture: swaying branches, rain, insects near an infrared lamp, shadows, headlights sweeping across a fence. AI detection is different because it recognises objects, such as a person, a vehicle, smoke or flame, and ignores the rest. Combined with zones and schedules, that removes most of the noise.
No system is perfect, so ask what happens when it gets something wrong. Can the person receiving the alert mark it as wrong? Does that feedback change anything for that camera, or does it disappear into a support queue? How are false alerts reviewed during the trial? A supplier that answers these questions clearly is one that expects to be measured.
A night view full of motion but nothing to act on: a branch, rain, insects and a headlight, all ignored.
Question 5
Can I draw zones that match my site, or only boxes and lines?
Detection zones tell the system where a rule applies. Their shape matters more than most buyers expect, because real hazards are rarely rectangular.
Many camera systems and analytics products offer two tools: a rectangle and a line. A rectangle drawn round a row of stockpiles also takes in the apron in front of them, the loading shovel parked beside them and part of the building behind. Every one of those extra areas is a source of alerts you did not want.
Polygon detection zones let you draw a shape with as many corners as you need, following the edge of a stockpile, a walkway, a fence line or a compound. The rule applies inside the shape and nowhere else. One camera can carry several polygons, each with its own detection type and schedule, such as smoke and fire over the bays around the clock and intrusion on the apron out of hours.
Polygon zones are also a privacy tool. Drawing round the area you need, and leaving out neighbouring property, roads and welfare areas, is easier to justify in a DPIA than watching a whole view. Polygon zones are one of the clearest differences between Woodcock AI and many alternatives, so ask every supplier to show you how its zones are drawn.
Polygon vs box
The same stockpiles zoned two ways: a polygon that follows the bays, and a box that takes in the apron and the loader too.
Check your own cameras before you choose
A free site assessment reviews the cameras you already have, confirms which views are suitable and for which detections, and recommends edge or cloud processing for your site, before anything is switched on.
Every site looks different: different cladding, different lighting, steam from a vent, a flag that moves in the wind. A model trained only on other people's footage meets your site for the first time on day one.
Ask whether the system is tuned to each camera. Woodcock AI is self-learning: models are hyperfitted to each camera's own scene, so the camera over your loading yard learns that yard. The ordinary footage where nothing is wrong is used as positive training signal, teaching the model what normal looks like at that camera.
Because real fires are rare, Woodcock AI also synthesises training data, including smoke, and runs controlled smoke tests on site to confirm each camera can see what it needs to. Every alert has a feedback button; marking an alert correct or wrong feeds back into retraining for that camera.
Camera 4 · loading yardQuiet footage is training signalHours where nothing happens teach the model what normal looks like at this camera.
Camera 11 · north fenceAlert marked wrongSteam from a vent mistaken for smoke. Fed back into retraining for this camera.
On-site checkSmoke test on the real viewSynthesised smoke for training, then a controlled smoke test on site to prove the camera sees it.
Example of how one site's own footage and feedback shape the model for each of its cameras.
Quiet footage
A quiet morning at a loading yard: footage like this teaches the camera what normal looks like.
Question 7
Can I search footage and change the set-up in plain English?
Ask your cameras All cameras
You asked
“Was anyone at the north fence after midnight?”
North fence · camera 11 Last night · clip
Rear gate · camera 12 Last night · clip
Change the design
“Extend the stockpile zone on camera 3 to cover the new bay.”
Download clipReview change
Example plain-English question and design change in the app, with matching clips ready to download.
Finding a moment in recorded footage usually means scrubbing through hours of video. Plain-English search changes that.
With ask your cameras, you type a question the way you would ask a colleague, such as whether anyone was at the north fence after midnight, and Woodcock AI lists the matching clips to watch and download. It is useful after an incident, for an insurer's question or before a toolbox talk.
The same approach applies to the set-up itself. Instead of working through settings menus, you can describe a change to your detection design in plain English, such as extending a stockpile zone to cover a new bay, and review it before it goes live.
When comparing systems, ask whether search covers all cameras at once, how far back it goes, and whether design changes need a support ticket or can be made by you.
Question 8
How will alerts reach me, and who responds?
An alert is only useful if it reaches someone who can act on it, with enough information to decide what to do.
Ask how alerts arrive, and whether each one carries the clip. A short clip and a plain-English description of what the camera saw let the person receiving the alert judge it in moments, without logging in to a recorder.
Woodcock AI sends alerts by the app, with the clip, and by email, to the people you choose for each zone and shift. Alerts to a monitoring centre (ARC) and phone-call alerts are available as optional add-ons. Those add-ons suit sites that are empty at night or need a guaranteed response.
Smoke alert · exampleSmoke seen in the stockpile zone
Clip attached
“Grey smoke rising from the middle bay, drifting towards the building.”
View clipCorrectWrong
Email copySame alert, same clipSent to the duty manager and the site lead
Optional add-onMonitoring centre or phone callIf your site needs a 24/7 response
Example smoke alert in the app, with the clip, a plain-English description and feedback buttons.
Question 9
What about GDPR and a DPIA?
AI CCTV processes images of people, so UK GDPR and the Data Protection Act 2018 apply. The ICO expects a data protection impact assessment before processing that is likely to result in high risk, and AI analysis of CCTV often falls into that category.
A good supplier helps you write it. Ask whether it will help you describe the purpose of each detection, the zones and hours covered, how long clips are kept, who can see them and how a person reviews alerts before anyone acts. Ask where video is processed and stored, which is another reason the edge or cloud choice matters.
Be cautious of any supplier that describes its product as compliant on its own. Compliance depends on how you use the system: proportionate zones, clear signs and privacy information, sensible retention and human review. The AI CCTV and GDPR guide and the CCTV compliance hub cover the detail.
Side by side
How do the main approaches compare?
Most buyers end up choosing between four broad approaches. Products vary within each, so treat this as a map of the questions to ask, not a verdict on any one product.
How the main approaches to AI CCTV compare on six buying questions
Approach
Existing cameras
Zone shapes
Learns your site
Footage search
Alerts and response
Install and support
On-camera analyticsbuilt into some cameras
Only cameras with the feature; mixed estates need matching models
Usually boxes and lines
Rarely: fixed models in the camera
Varies; often by event type only
Via the recorder or VMS you already use
Your installer, camera by camera
Generic cloud video AIsoftware as a service
Often, over ONVIF or RTSP
Varies; many offer boxes and lines
Varies: often one model for every customer
Varies by product
App or email, varies by product
Often self-install, remote support
Remote monitoring centrepeople watching
Yes, alongside their own alarm triggers
Depends on the triggers feeding it
Operators learn your site over time
Operators review on request
A person decides and calls out
Monitoring company, alongside your installer
Woodcock AIvideo AI on existing CCTV
Yes, over ONVIF or RTSP; suitability confirmed per camera
Polygon zones with as many corners as you need, plus lines
Yes: models fitted to each camera's scene, retrained from your feedback
Plain English: ask your cameras, and describe design changes
App with the clip, and email; ARC and phone-call alerts as add-ons
End to end: survey, design, supply, install, support 24/7
Software is only half the job. Someone has to survey the site, decide which cameras cover what, fix poor views, install any extra hardware, commission the system and look after it.
Ask who does each of those steps, and what happens when you have a problem at 3am on a Sunday. Systems sold as software only leave the design and camera work to you or your installer, which can work well if you have a strong installer, and less well if you do not.
Woodcock AI offers an end-to-end design, supply and install service: site survey, design, supply, installation, commissioning, go-live and ongoing support, with support available 24/7. If you already have a trusted installer, Woodcock AI works through installer and reseller partners too.
Site survey
Survey kit on a van tailgate: the site plan, camera positions and the tools used to check each view.
Question 11
How should I run a trial?
A trial on your own cameras, on your own site, is the most reliable test there is. Demonstrations on someone else's footage tell you what a system can do somewhere else.
Agree what good looks like first. Write down which cameras are included, which detections are being tested, who reviews the alerts and how you will judge the result, for example how many alerts needed action and how many were noise.
Cover the conditions that matter. Run long enough to include day and night, a spell of bad weather and both busy and quiet periods. For smoke detection, ask whether a controlled smoke test can be run on site.
Use the feedback. Mark alerts correct or wrong throughout. A system that learns your site should get better during the trial, and you should be able to see that it has.
Review honestly. At the end, look at missed events as well as false ones, check the DPIA still matches what you are doing, and decide which cameras to add. Woodcock AI trials start with the free site assessment.
Scorecard
How do the systems on my shortlist score?
Use the eleven questions to score up to two systems side by side. Yes scores two, Partly one and No nothing, for a maximum of 22. Your scores are saved in this browser, so you can come back after each supplier call.
Score each system Yes, Partly or No on every question. Scores are saved in this browser only, nothing is sent.
AI CCTV buyer's scorecard: eleven questions, scored for two systems
Answer four quick questions for a suggested starting set-up. It is a starting point for a conversation, not a design: the free site assessment confirms what each of your cameras can do.
Waste or recycling
Smoke and fire detection with polygon zones round each stockpile and bay, plus intrusion out of hours. Alerts by app with the clip and email; consider a monitoring centre add-on if nobody is on site at night.
Industrial or warehouse
Smoke and fire detection over storage and plant areas, intrusion on the perimeter and doors, and PPE detection (hard hats and hi-vis) where your risk assessment calls for it.
Construction
Intrusion zones over the compound and laydown areas and a line across the gate out of hours, smoke and fire detection near hot works and stores, and hard hat and hi-vis checks at the gate during site hours.
Vehicle or storage yard
Intrusion zones round parked vehicles and stock out of hours, smoke and fire detection, and vehicle detection where supported.
Weak or mobile-only connection
An on-site edge server usually suits best, because video is analysed on site and only alerts and clips are sent out.
Reliable broadband
The Woodcock AI Cloud can work well with nothing extra to install; an edge server is still an option. The choice is made per site in the free site assessment.
There is no single best system for every site. The best AI CCTV system for you is the one that works with your existing cameras, detects the events on your risk list, keeps false alarms low on your views, and comes with a proper trial and support. Use the 11-point checklist on this page to compare the options side by side.
Can AI CCTV work with my existing cameras?
Usually, yes. Most modern IP cameras and recorders share their video over ONVIF or RTSP, which is what AI software connects to. Older analogue systems may need an encoder. Ask any supplier to confirm, camera by camera, which views are suitable before you commit; Woodcock AI does this in a free site assessment.
Is edge or cloud AI CCTV better?
Neither is better everywhere. Edge processing runs on a box on site, so it keeps working through a weak internet connection and sends out only alerts and clips. Cloud processing needs nothing extra on site but relies on the connection. Woodcock AI offers both, an on-site edge server or the Woodcock AI Cloud, chosen per site.
What should an AI CCTV buyer's checklist include?
At minimum: compatibility with your existing cameras, edge or cloud processing, the detections you need, false-alarm handling, the shape of detection zones, whether the system learns your site, footage search, alert channels, GDPR and DPIA support, install and support, and a trial on your own site.
How do I choose AI fire detection software?
Look for video smoke and flame detection that runs on cameras already watching your stockpiles, yards or buildings, uses detection zones drawn round the areas at risk, sends alerts with the clip, and is tested on your own site. Treat it as an extra layer alongside your fire alarm, fire prevention plan and any thermal cameras, not a replacement.
Why do polygon detection zones matter?
Real hazards are rarely rectangular. A polygon follows the edge of a stockpile, a walkway or a fence line, so the rule applies only where you need it. A rectangle takes in roads, neighbouring buildings and plant, which adds alerts you did not want.
How do AI CCTV systems reduce false alarms?
Good systems recognise objects such as people, vehicles, smoke and flame rather than reacting to any pixel change, and only alert inside the zones and hours you set. The strongest also learn from each site, using feedback on wrong alerts to retrain the model for that camera.
Do I need a DPIA for AI CCTV?
The ICO expects a data protection impact assessment before processing that is likely to result in high risk, and AI analysis of CCTV often falls into that category. In practice, complete one before you switch AI detection on, and choose a supplier that helps you describe zones, retention and human review.
How long should an AI CCTV trial last?
Long enough to cover the conditions that matter on your site: day and night, a spell of bad weather, and the busy and quiet periods of your week. Agree up front what a good result looks like, which cameras are included and who reviews the alerts.
How are Woodcock AI alerts sent?
Alerts go out by the Woodcock AI app, with the clip, and by email, to the people you choose for each zone and shift. Alerts to a monitoring centre (ARC) and phone-call alerts are available as optional add-ons.
Next step
Put your shortlist to the test on your own cameras
A free site assessment reviews the cameras you already have, confirms which views suit smoke, fire, intrusion and PPE detection, recommends edge or cloud processing, and sets up a trial you can judge for yourself. Alerts to a monitoring centre (ARC) and phone-call alerts are available as optional add-ons.
