How GNS AI-managed service works: minimising downtime from SMEs to very large enterprises
GNS Engineering Team 4 min read
For any business that depends on its network, failure is not a rare event — it is an operational risk that will eventually materialise. Offices rely on wireless, cameras, door access and intercom at the same time; when one element fails, front-line operations can stall. Traditional maintenance starts with a fault report: the problem has already happened and the time has already been spent on recovery rather than prevention.
GNS AI-managed service (GNS CARE) changes that starting point. We take device activity records, compare them against a long-term baseline for each device, and step in before hardware actually fails.
Why traditional maintenance struggles to prevent downtime
The three constraints below have nothing to do with a provider's technical skill — they are structural limits of a response-based model:
- It only starts after a failure. The trigger is the customer's call; before that call, the provider knows nothing about device health.
- It only sees "this moment". An on-site check is a snapshot: if it powers on and connects, it passes. Disk read errors, voltage fluctuations and roaming failures never appear in that snapshot.
- Maintenance and engineering are separated. Hardware is covered by one supplier and the network by another, with records scattered in between; when a problem crosses both, time goes into coordination rather than repair.
The five-layer architecture behind GNS CARE
The architecture deliberately avoids placing third-party software inside a customer network, and each layer has a clear role.
Layer 1 — Official API telemetry. UniFi's official APIs provide device activity records, alerts and status. No agent is installed and no inbound port is opened.
Layer 2 — Encrypted transport. Records travel to our cloud platform over encrypted channels, encrypted in transit and at rest, viewable only by authorised engineers.
Layer 3 — Our own cloud platform. Tenants are isolated from one another, and history is retained so that trends can be compared.
Layer 4 — SIEM and anomaly detection. SIEM correlates events across devices while anomaly detection compares each device against its own baseline; only when both agree does a signal become actionable.
Layer 5 — Our own AI platform. The AI platform combines correlation results with baselines to identify devices heading for failure, sequence replacements and recommend configuration or placement changes.
The data we actually read
All telemetry fields come from official developer documentation rather than a self-built agent or guesswork. Typical examples:
- Access points: CPU and memory utilisation, uptime, load averages, channel and width, firmware version, heartbeat timing.
- Gateways and firewalls: device state, firmware updatability, firewall rule sources and destinations, ISP latency, packet loss and throughput.
- Switches: per-port link state and negotiated speed, PoE state (with the ability to power-cycle a single port), stacking and aggregation settings.
- Cameras and NVRs: recording state, arm mode, smart-detection events (motion, line crossing, loitering, audio), disk health and offline events.
When a disk, power supply, link or wireless coverage starts to drift from normal, we receive the signal before the customer notices.
The same principle from SMEs to very large enterprises
AI-managed service is not reserved for large enterprises — the difference is coverage and response tier.
- SMEs (single office): start with Basic — 24/7 remote monitoring and alerts, firmware and security updates, configuration backup, unlimited remote support and monthly reports.
- Growing businesses and multi-department offices: move to Full — a 1-hour response, 24-hour device replacement, emergency site visits, quarterly reviews and camera health monitoring.
- Schools, clinics and retail: connectivity-critical sites where we recommend dual-gateway high availability (Shadow Mode) alongside offline alerts and PoE monitoring.
- Large enterprises and multi-site organisations: every site in one managed list and one platform, with telemetry and event records available through APIs for existing ITSM or monitoring tools, and reporting by department or site.
Six practical benefits
- No agent to install and no port to open. Your existing architecture does not change.
- The moment of discovery moves from after failure to before it. Replacement decisions are based on trends rather than a fault report.
- Cross-device correlation. Anomalies across switches, gateways and cameras are connected instead of handled in isolation.
- Auditable records. Events and how they were handled are retained, which suits organisations that need compliance evidence.
- A written monthly report. Management sees trends, risks and improvement recommendations.
- Engineering and managed service from one team. From survey and installation to long-term maintenance, interface coordination disappears.
Case reference: railway train monitoring
We already apply the same method in railway operations. Onboard equipment feeds an EN50155 data logger, and our AI platform analyses trends and anomalies — hardware-related incidents fell by more than 30%, and real-time alerts shortened response time. The parallel with network devices is clear: failures will happen; what matters is discovering them before they cause an outage.
Cost and return
Managed service is priced by device count and tier: Basic from HK$800 per month and Full from HK$2,000 per month. Against the operational loss, staff hours and customer impact of a single unplanned outage, proactive management usually costs a fraction of one incident.
FAQ
How is AI-managed service different from ordinary remote monitoring?
Ordinary remote monitoring compares instantaneous values against thresholds and acts after an alert fires. GNS CARE reads device activity through official APIs, keeps the history, and builds a baseline for each device; when values drift, it raises a signal before the hardware actually fails and recommends replacement order and configuration improvements.
Do you install software on our network or open inbound ports?
No. We read device records directly through official UniFi APIs — there is no third-party agent on your network and no inbound port to open, so your existing architecture is unchanged.
What device data do you actually read?
Access points, gateways and firewalls, switches, cameras and NVRs. Typical fields include CPU and memory utilisation, uptime, load averages, firmware versions, switch port speed and PoE state, ISP latency, packet loss and throughput, camera and recording state, plus offline and smart-detection events.
Is it limited to UniFi devices?
The telemetry and prediction layer focuses on UniFi devices because the official APIs provide complete and stable data. Where other brands are in use, we assess which sources can be integrated during solution design and present them in the same report.
Is it suitable for SMEs, and how is the monthly fee calculated?
Yes. GNS CARE is priced by device count and tier: Basic from HK$800 per month and Full from HK$2,000 per month. Smaller companies can start with Basic and upgrade to Full as their estate grows or as their tolerance for downtime falls.
How does it work for large or multi-site organisations?
Multi-site organisations can bring every office's devices into one managed list and view them on a single platform. Telemetry and event records can also be fed into an existing ITSM or monitoring system through APIs, with reports by department or site. The Full tier adds a 1-hour response and 24-hour device replacement.
Can failure prediction produce false alarms?
A signal is only raised when SIEM event correlation and baseline deviation both apply, which reduces noise from single-value fluctuations. Every signal is reviewed by an engineer, and the reasoning is documented in the report.
How long does onboarding take?
A free site assessment and device inventory typically completes within a week. Once the managed list and tier are confirmed, telemetry and monitoring usually go live within a few working days, with monthly reporting thereafter.
