Commercial Bank Video Cloud and AIOps

Challenge

Video conferencing, remote-teller and training streams ran on three incompatible systems with a dozen terminal models; branches bought their own circuits, so quality varied and stutter complaints stayed high. Regulators required all audio-video data and AI analysis to remain inside the bank.

GSS scope of responsibility

GSS owned private platform deployment, branch network QoS upgrade, nationwide terminal installation and centralised operations, plus the in-house infrastructure for AI capabilities. Platform software came from the vendor; activation and review of the AI meeting assistant and compliance alerts were decided by the bank's compliance function.

Solution

In-house private video cloud, nationwide outlet circuit QoS upgrade and wave-based installation by GSS field teams, plus an on-premises AI meeting assistant (live captions, minutes, compliance keyword alerts) and AIOps stutter prediction built on circuit and terminal telemetry, all operated 7×24 by the GSS NOC.

Background

The bank ran three mutually incompatible audio-video systems — conferencing, remote teller and training live-streams — across 31 provincial branches and more than 1,200 outlets, on over ten terminal models. Branches procured their own access circuits, so bandwidth and QoS were inconsistent. Head-office IT received hundreds of “frozen picture, broken audio” tickets each month and could not tell whether the fault was the terminal, the outlet circuit or the platform.

Regulators added two hard constraints: recordings and any AI-derived output must never leave the bank’s environment, and all meeting content had to be searchable and auditable.

Solution

Network and platform layer (traditional network business)

• Private video cloud deployed active-active across the bank’s two data centres, with media servers placed per provincial branch for local ingress.
• Outlet circuit QoS upgrade: GSS re-procured and re-engineered access circuits for 1,200+ outlets, adding a dedicated queue for media with end-to-end DSCP marking, in 20M / 50M / 100M tiers by outlet class.
• Terminal standardisation down to two models, installed, commissioned and old units recovered by GSS field engineers in 11 provincial waves, with average outlet downtime held under 40 minutes.

AI layer (AI business)

• AI meeting assistant on the bank’s own GPU pool: live captions with Chinese–English translation, automatic minutes and action-item extraction. For remote-teller sessions a compliance keyword monitor flags regulated sales language in real time and bookmarks the recording segment for compliance review.
• AIOps stutter prediction: the GSS NOC consolidated circuit latency / loss, terminal logs and platform media-quality metrics from all outlets into one time-series store and trained an outlet-level degradation model. It raises alerts 15–30 minutes before users notice, letting the NOC adjust bandwidth or restart a terminal pre-emptively.
• All models and data stay inside the bank; GSS operates exclusively through the bastion host and never touches business data.

Results

Metric	Before	After
Outlet terminals	~900 across three systems	1,200+ on one platform
Platform availability	Not measured	99.98% (12-month rolling)
Stutter-related tickets	~320 / month	~90 / month (−72%)
Minutes preparation	~45 min per meeting	Auto-generated, 5 min review
Compliance review coverage	~3% sampled	100% of keyword hits flagged

The programme started in 2023 and completed the final outlet cut-over in June 2024. GSS now provides ongoing managed operations with quarterly network-quality and AI-assistant usage reports.