Find it in one query, not hours of tape.

Questions buyers ask us

FAQ

What does 'AI search' on video actually mean?

Finding a specific person, vehicle, behavior, or event across hours of recorded footage in seconds, without scrubbing the timeline by hand. Modern platforms index frames as they're recorded, so any future query runs against a structured database, not raw video. The manual incident-review scrub becomes a single filtered query.

Does AI search work on cameras we already have?

On vendor-tied cameras, search runs natively if the cameras support it. On older or mixed fleets, camera-agnostic overlays ingest existing feeds via ONVIF or RTSP and build the search index on top. Older cameras need to clear a basic resolution and framerate threshold (typically 1080p at 15 fps) for object-recognition accuracy.

Is using cameras for productivity monitoring legal?

Cameras can capture employee activity in most states, but legal limits vary. Federal law allows video monitoring with notice. Connecticut, Delaware, and New York require written notice. Audio recording is restricted in two-party-consent states. Illinois BIPA requires written informed consent for biometric identifiers like face geometry. Tec-Tel scopes deployments to the regime your sites operate under and documents the consent flow as a deliverable. Not legal advice.

What does Illinois BIPA actually require?

Illinois BIPA (the Biometric Information Privacy Act) requires written, informed consent before a private entity collects biometric identifiers such as face geometry, fingerprints, or retina scans, plus a published retention and destruction schedule. For productivity monitoring in Illinois, Tec-Tel defaults to badge or PIN credentials over facial recognition unless the customer has a written, BIPA-compliant consent program. Class-action settlements under BIPA have run into the hundreds of millions, so this is not a corner to cut.

Who's the right buyer for AI search?

Loss prevention teams running multi-store incident investigations. HR-supported safety teams reviewing slip-and-falls and near-misses. Audit teams pulling PCI, HIPAA, FERPA, or CMMC retrieval. Operations leaders running line-balancing or speed-of-service analytics. The common thread: people whose job involves answering 'what happened on camera' and who spend hours doing it manually.

What about training data privacy and footage retention?

Most vendors pre-train object-detection models on internal labeled datasets plus public benchmarks. On-customer fine-tuning is typically opt-in. Footage retention, region of storage, and whether the vendor uses customer footage for model improvement live in the data processing agreement. Read the DPA before signing, especially for healthcare, education, and EU-touching deployments, where GDPR adds DPIA requirements for facial recognition.

What's a realistic accuracy rate?

Vendor specs typically claim under 5% false-positive rate for object detection in tuned environments. Person-search accuracy depends heavily on lighting, camera angle, and query specificity (a red jacket in a crowd of red jackets is harder than a unique plate). Real-world false-positive numbers start higher and drop as zones, masks, and thresholds get tuned on site. Plan for a tuning window before relying on it for legal-defensible retrieval.

How much does AI search cost?

Vendor-tied platforms bundle search into the camera or VMS license. Camera-agnostic overlays price per camera per month, scoping spend to the cameras that need search rather than the whole fleet. The labor-cost offset is usually the clearest part of the math: every incident review that once took hours of manual scrubbing collapses to a filtered query, and that time adds up fast across a multi-site loss-prevention team.