How to Add AI Theft Detection to the Cameras You Already Have

Most shops already own the hardware that AI theft detection needs: the CCTV cameras on the ceiling. What they usually lack is someone watching those cameras live. Footage gets pulled after a stock count shows a gap, by which point the person and the product are long gone.

The scale of the problem is why that gap matters. The British Retail Consortium’s 2025 Crime Survey counted more than 20 million customer theft incidents in 2023/24, costing UK retailers a record £2.2 billion.

This guide explains how to add AI theft detection to the cameras you already have: what you need, how the connection works, what the AI watches for, how to handle privacy, and a checklist to run through before you start. For the technical side of connecting a recorder, see How to Connect Your Existing Cameras to De Flow AI.

What you need before you start

• The cameras you already have. De Flow AI works with Hikvision, Dahua and AXIS cameras, with systems managed in Milestone XProtect, and with any RTSP-compatible camera. There is no new hardware to buy or mount.
• Access to your recorder or video system. The login for your NVR, DVR or video management software. A dedicated viewer account is better than sharing the admin password.
• A camera map. A simple list of which camera covers which area: tills, self-checkouts, entrances, high-theft aisles and the stockroom door.
• Your POS details, if you use Shopify POS. De Flow AI works with Shopify POS, and till-side fraud such as refunds and voids is part of what the system looks for.
• An alert rota. Who should receive alerts on each shift, and on which channel.

Step by step: from existing CCTV to live alerts

1. Map what your cameras can actually see

AI can only flag what a camera can see. Walk the shop floor with your camera list and check the views that matter most for loss: every till and self-checkout, the entrance and exit, the aisles where high-value lines sit, and the stockroom door. Note any blind spots, cameras left pointing at a wall after a refit, and views that suffer from glare or poor night-time images. Fixing a camera angle costs nothing and improves every alert that camera produces.

2. Check that your cameras are compatible

If your cameras are Hikvision, Dahua or AXIS, or your system runs on Milestone XProtect, you are covered. If your system is another brand, check whether the cameras or the recorder can provide an RTSP stream; your installer or the device manual can confirm it. De Flow AI works with RTSP-compatible cameras, so the stream a system can provide matters more than the brand on the box.

3. Connect through the browser

Because De Flow AI connects through the browser, there is no technician visit and no new hardware on site. Whoever manages your CCTV provides access to the cameras or the recorder. The camera connection guide covers what to have ready.

4. Choose what to watch for

Start where losses tend to hide: the tills and self-checkouts, then the entrance and the high-value aisles. On those cameras, De Flow AI looks for shoplifting and theft patterns, self-checkout scan-avoidance, sweethearting, and refund and void-receipt fraud. Each is explained below. Beginning with a few high-risk areas makes it easier for staff to learn the alerts before you widen coverage.

5. Decide who gets alerts, and where

Alerts can go to the De Flow AI dashboard, by mobile push notification, SMS, email or WhatsApp. Each one carries a timestamp, a short video clip and a confidence score, so the person receiving it can see what happened before deciding what to do. Route alerts by role: a self-checkout alert is most useful to the colleague supervising the self-checkout area, while a run of refund alerts belongs with the store manager.

6. Go live

Once the cameras are connected and the alert routes are set, De Flow AI is live in about 24 hours. Use the first week to review alerts with the team: which were useful, which areas need a better camera angle, and whether the right people are receiving them.

7. Agree what staff do when an alert arrives

An alert is only as good as the response to it. Write a short playbook before you go live:

• Safety first. No colleague should physically confront a suspected thief.
• Lead with service. An offer of help, or a visible presence near the self-checkouts, is a common way to deter an opportunist without accusing anyone.
• Check the clip before acting, and take the confidence score into account.
• Record what happened, so recurring patterns can be raised with the police or your security provider.

What AI theft detection on existing cameras catches

Three of these, sweethearting, refund fraud and void-receipt fraud, happen at the till rather than at the door, which is why door-based measures such as security guards and EAS gates struggle with them. Our comparison of AI loss prevention, security guards and EAS gates looks at that trade-off, including what each option costs. For the product itself, see De Flow AI Loss Prevention and real-time alerts.

It is worth being clear about the limits too. The AI only sees what your cameras see, it does not physically stop anyone, and it relies on a person acting on the alert. Good camera coverage and a clear staff playbook matter as much as the software.

Privacy and staff: doing it properly

De Flow AI does not use facial recognition. It looks for behaviour and events, such as an item bagged without a scan, rather than for who someone is.

Using CCTV still brings data protection duties. In the UK, the Information Commissioner’s Office expects a data protection impact assessment (DPIA) in most video-surveillance cases, and lists monitoring people at work among the situations that call for one. Its CCTV guidance for small organisations also asks you to record why you use CCTV and how long you keep footage. If you already run CCTV, update that record and your DPIA to cover AI analysis before you go live.

Tell your team before you go live, including that till-side checks such as sweethearting detection cover staff transactions. Our article on AI cameras and UK GDPR goes into more detail.

Pre-setup checklist

• List every camera and the area it covers.
• Confirm the brand or system: Hikvision, Dahua, AXIS, Milestone XProtect, or an RTSP-compatible camera.
• Check the views of tills, self-checkouts, entrances and high-value aisles, and fix any blind spots.
• Create a dedicated viewer login for the recorder or video system.
• If you use Shopify POS, have its details ready.
• Decide who receives alerts on each shift, and on which channel: dashboard, push, SMS, email or WhatsApp.
• Write a short alert playbook that puts safety first.
• Update your DPIA, your CCTV records and your staff briefing to cover AI analysis.
• Book time in the first week to review alerts with the team.

Frequently asked questions

Do I need new cameras to add AI theft detection?

No. De Flow AI connects to the CCTV and IP cameras a shop already has, including Hikvision, Dahua and AXIS cameras, systems managed in Milestone XProtect, and RTSP-compatible cameras. There is no new hardware to install.

How long does it take to set up?

De Flow AI is live in about 24 hours. The connection is made through the browser, so there is no technician visit.

What if my cameras are a different brand?

Check whether your cameras or your recorder can provide an RTSP stream; your installer or the device manual can confirm it. De Flow AI works with RTSP-compatible cameras, so the stream matters more than the brand.

How will I receive alerts?

On the De Flow AI dashboard, by mobile push notification, SMS, email or WhatsApp. Each alert includes a timestamp, a video clip and a confidence score.

Does De Flow AI use facial recognition?

No. It looks for behaviour and events, such as an item bagged at a self-checkout without a scan or a suspicious refund, not for who someone is.

How much does it cost?

Pricing is quoted per store. Book a demo to get a quote for your shops.