A business may want to detect a specific event that a standard CCTV package does not address. Cellphone use in a designated area or vaping-related alerts are examples of requirements that need careful scoping. The first question is what can actually be observed with the available cameras, sensors and integrations—not whether a generic AI label appears on a product.
Describe the requirement precisely
“Detect phone use” could mean someone holding an object, looking at a screen or using a device while completing a particular task. Those are different observations. Describe the location, conditions and response you need. A short, clear operational requirement gives a technical team something it can assess and test, while a vague requirement can lead to an unreliable demonstration.
Separate video observations from sensor readings
Do not assume ordinary CCTV can reliably identify vaping in every environment. Visibility, occlusion and the nature of the event may require a different approach or supported sensors. If a sensor provides an alert, associated video may offer context without proving what caused the sensor reading. The proposed design should state which component detects what and where uncertainty remains.
What should a feasibility assessment cover?
Provide a general description of:
- The event and the area in which it matters.
- Existing camera positions and typical operating conditions.
- Relevant sensors or other available systems.
- What action a reviewer should take after an alert.
- Normal activities that might look similar.
- The outcome needed from a limited pilot.
Do not send staff recordings or credentials through a public enquiry form. Agree an appropriate assessment process before sharing sensitive material.
Agree acceptance criteria before the pilot
Test representative conditions and include negative examples: events that should not trigger an alert. Ask how missed events and unnecessary alerts will be recorded, how changes will be evaluated and which limitations will be documented. A successful demonstration in one camera view should not be presented as proof of performance across every site.
Keep human review in the workflow
A custom alert should prompt a review under an established workplace process. It should not, by itself, be treated as proof of misconduct. Decide who may view the information, who checks the context and how incorrect alerts are handled. Clear responsibilities make the proposed technology easier to assess and operate.
Can AI Monitor build a bot around our requirements?
AI Monitor scopes custom AI solutions according to available cameras, sensors, integrations and technical feasibility. A request is assessed before capabilities are confirmed. Explore custom AI Bots and describe the event you want to detect. Start with the business problem and expected response, and use the assessment to determine a suitable approach.
Back to Insights