Choosing a video analytics and AI solution for a single pilot site is relatively simple. Choosing a partner capable of deploying the solution across 50, 200, or 500 sites is an entirely different matter. Between proof-of-concepts that don’t scale, vendors that disappear after the sale, and architectures incompatible with existing infrastructure, decision-makers in retail and logistics face real risks.
This guide outlines the 10 criteria to evaluate before selecting your video AI solution for large-scale deployment.
A large-scale deployment cannot require the systematic replacement of cameras. The solution must be compatible with standard RTSP video streams and work with CCTV equipment from most major vendors (Axis, Hikvision, Bosch, etc.).
Question to ask the vendor: What is the list of certified compatible camera models? What is the procedure in case of incompatibility?
Some sites have limited connectivity. Certain data is too sensitive to be transmitted to the cloud. The solution must offer an edge computing (local processing) or hybrid architecture, not just a reliance on the central cloud.
For retail and logistics, local processing reduces latency, limits bandwidth costs, and enhances GDPR compliance.
In France and Europe, any video analytics solution that processes personal data is subject to the GDPR. Non-biometric computer vision —which analyzes silhouettes and foot traffic without identifying individuals—is the only approach compatible with consumer-facing retail deployments that do not require individual consent.
Verify that the provider has comprehensive GDPR documentation, a clear data retention policy, and does not use facial recognition or biometric data.
The performance of a computer vision model in a lab does not predict its performance in a store with obstructions, variable lighting, suboptimal camera angles, and dense crowds.
Require precision and recall metrics measured on real-world datasets, not just on standardized benchmarks.
CORE by XXII benchmark: 98.6% accuracy in group detection across real-world retail deployments.
A large-scale deployment requires process standardization: standardized onboarding, centralized configuration, remote update management, and availability monitoring by site.
Ask to see the deployment roadmap for an existing multi-site customer. How long does it take from contract signing to go-live at the 10th site? At the 50th?
An incident affecting 200 sites simultaneously cannot be managed in the same way as an incident on a single pilot site. The provider must offer explicit SLAs (response time, resolution time), access to Level 2 support, and ideally proactive site monitoring.
Reputable technology partners publish their contractual SLAs and adhere to them.
Video analytics is only valuable if its data feeds into your decision-making systems. The solution must expose REST APIs or webhooks to enable integration with your ERP, logistics WMS, or business intelligence tool.
Verify the availability of comprehensive API documentation and the existence of native connectors or pre-existing integrations with your tools.
The price per site should decrease as volume increases. Be wary of fixed-cost models per camera or per site, which become prohibitively expensive at scale. French software vendors specializing in video AI generally offer models based on the number of active sites with tiered volume levels.
Negotiate the terms for scaling up from the outset and ensure they are contractually agreed upon.
A 3- to 5-year deployment means your partner must remain in business, be well-funded, and committed to product development. Assess the vendor’s financial stability, its track record of fundraising or profitability, and its functional roadmap for the next 18 months.
For large enterprises, being part of a French or European technology ecosystem is sometimes a sovereignty criterion worth considering.
A successful POC does not guarantee a successful multi-site deployment. Ask for client references with a significant number of deployed sites, and take the time to contact them directly.
French software vendors such as XXII have deployed CORE for organizations like SNCF, Nhood, Klepierre, and Iris Galeries, with multi-site deployments involving complex infrastructure, compliance, and SLA requirements.
| Criterion | What to Check |
|---|---|
| 1. Camera compatibility | List of certified devices |
| 2. Edge/cloud architecture | Local processing available |
| 3. GDPR / Non-biometric | Comprehensive documentation; no facial recognition |
| 4. Accuracy in real-world conditions | Metrics from real-world deployments |
| 5. Multi-site deployment | Standardized onboarding process |
| 6. Support SLA | Contractually defined and adhered to |
| 7. API integrations | Documentation available |
| 8. Volume-based pricing model | Decreasing price tiers |
| 9. Vendor sustainability | Funding, roadmap, history |
| 10. Verifiable references | Multi-site clients available for contact |
What is the difference between a video analytics solution and a basic video surveillance system? Video surveillance records footage and allows for playback. AI video analytics processes the video stream in real time to extract behavioral metrics (traffic, wait times, conversion rates) without requiring continuous human intervention.
Do you need to be an AI expert to deploy a video analytics solution? No. Solutions like CORE by XXII are designed to be integrated by retail IT teams without AI expertise. The vendor handles algorithm configuration during onboarding.
Can data collected by AI video be used in court? No, in the case of non-biometric solutions: the processed data is aggregated and anonymous, with no recording of identifiable footage. It cannot identify individuals and does not constitute admissible evidence.
How can you evaluate the accuracy of a computer vision model before purchasing it? Request a proof of concept (POC) at your own location, using your own cameras, and measured against a representative sample of one week’s worth of data. Compare the results to manual counts (ground truth).
Choosing a video analytics and AI solution for large-scale deployment means choosing an industry partner—not just a tool. Technical criteria matter just as much as operational robustness, regulatory compliance, and the ability to scale with your organization.
→ Request a project assessment with the XXII team.