Cash regulations
Violations in the sequence of actions, cash handling and banknote-detector use are hard to identify online.
AI platform for operational control of bank branches
A unified video analytics layer for controlling cash processes, queues, SLA, security, ATMs, staff, compliance and KPIs across the branch network.
On-Premprocessing, events and archive inside the bank infrastructure
RTSP / ONVIFconnection to existing VMS, NVR and CCTV systems
FX operationscontrol of foreign-exchange and cash operations
SLAqueues, waiting time and service-window load
ATMincidents at ATMs and self-service zones
BI / KPIbranch ratings, reports and executive dashboard
Background
Bank branches need a new level of operational control
Cameras in branches, cash areas and ATM locations are already installed, but they usually work as an archive for post-incident review. AI turns CCTV into a source of events, KPIs, notifications and evidence for security, operations and service-quality teams.
Queues and SLA
Queue length, waiting time and service-window load are often measured manually or with a delay.
Physical risks
Suspicious behavior, unattended items, ATM incidents and access to staff-only zones depend on operator attention.
Branch network
Unified standards are difficult to scale: management needs objective KPIs by branch, region and shift.
What changes with AI
From separate cameras to a bank network management system
The platform brings together video, events, rules, notifications, investigations and management reporting. A violation is no longer just a recording in the archive, but an event card with a frame, video fragment, status and owner.
Video as data
Video streams are transformed into events, metrics, timings, zones and sequences of actions.
Online response
The cashier, controller or security team receives a near real-time signal instead of learning about the issue after a complaint.
Evidence base
Events are linked to video segments, processing statuses, comments and exportable reports.
Executive dashboard
Branch ratings, SLA, operational losses, security and efficiency are visible at network level.
7 modulescash desk, hall, security, staff, speech, ATM, compliance
1 layerevents, roles, reports, video and BI
10-30streams for a pilot in 1-2 branches
4-6 weeksscenario and reporting stabilization
Before
Video surveillance works as an archive
- Critical operations are checked selectively and after the fact.
- Queues, SLA and branch load are estimated approximately.
- Security incidents depend on manual monitoring.
- There is no unified branch rating by quality and risk.
After
AI records events and manages response
- Cash actions, queues, ATMs and staff-only zones are controlled automatically.
- Events are logged with a frame, video segment and status.
- The manager sees KPIs for branches, regions, cashiers and shifts.
- New scenarios are added without replacing cameras.
Platform capabilities
Core AI modules for bank branches
Scenarios can be launched step by step: first quick modules for queues, SLA and security, then more complex process cases such as foreign-exchange operation control.
Cash deskCash operations
Cashier actions, cash drawer, banknote detector, cash, documents and process order.
HallService hall
Queues, waiting time, window load, visitor counting, heat maps and SLA.
SecuritySecurity
Suspicious behavior, unattended items, conflicts, crowding and staff-only zones.
StaffStaff and standards
Presence, uniform, badge, idle time, shift load, patrols and discipline.
SpeechSpeech analytics
Dialogue quality, required phrases, complaints, conflict topics, cross-sell and script assessment.
ATMATMs and self-service
Long dwell time, camera covering, vandalism, ATM queue and night incidents.
1banknote intake
2authenticity check
3core banking entry
4cash drawer
5cash payout
6operation closure
Authenticity check skippedcash desk 2 · high importance · 11:42
notificationPayout before completionlink to video fragment · status open
FX operationsReference operationall steps completed · no violations
ok
Cash process
Control of FX operations and cashier actions against the reference business process
The system compares actual video events with the reference operation scheme: banknote intake, authenticity check, cash-drawer actions, registration, payout and closure.
- recognition of hand actions, cash, documents and the cash drawer
- control of banknote checks through DOCASH / detector
- detection of skipped steps, reordered steps and prohibited actions
- signal to the cashier and control units
- violation log with status, comment and video segment
Service hall
Queues, SLA, window load and customer flow
Video analytics provides objective KPIs for branch load: how many clients came in, where a queue formed, how long the wait takes and when an additional window should be opened.
- queue length, average and peak waiting time
- visitors per hour, peak intervals and zone heat maps
- load of service windows, cash desks and workstations
- lost clients and unresolved visits
- comparison of branches, regions and shifts by SLA
08:42average wait
97%network SLA
35%high load
Service hallsuspicious behavior, conflict, crowding, fall, unattended item
ATM / self-servicelong dwell time, camera covering, vandalism, night incident
Staff-only zonesunauthorized passage, open door, outsider, patrol violation
Perimeterentrance area, queue, abnormal movement, high-importance event
Security
Control of branches, ATMs and staff-only zones
AI helps the operator avoid missing an event and quickly open the right video fragment. The system records physical risks in the hall, at ATMs, at the entrance and in staff rooms.
- suspicious behavior, long dwell time and repeated approaches
- unattended items, conflicts, falls and crowding
- entry to staff-only zones and unauthorized passage
- camera covering, vandalism and abnormal events at ATMs
- event cards for security and investigations
ATM and self-service
Control scenarios for ATMs and self-service zones
ATMs and self-service terminals operate outside the normal branch mode, so early events matter: long dwell time, attempts to cover the camera, vandalism, queues and night activity.
Long dwell time
Notification if a person stays at the ATM longer than the configured time without a normal operation.
Camera covering
Detection of an attempt to block the view, turn the camera or degrade video quality.
Suspicious behavior
Non-standard actions at the terminal, repeated approaches, inspecting the case or abnormal activity.
ATM queue
People counting, waiting time and self-service zone load during peak hours.
Vandalism
Hit, damage, attempted break-in, aggressive behavior and night incidents.
More than one person
Event for scenarios where several people in the zone increase risk.
Camera offline
Monitoring stream availability and image quality to maintain archive continuity.
Service access
Control of scheduled work, patrols, access and device service time.
Staff and standards
Presence, discipline and shift load
The platform shows staff presence at workplaces, shift load, idle time, compliance with uniform, badge, patrols and branch operating standards.
- employee presence at the service window and cash zone
- idle time, absences, shift overload and uneven load
- uniform, badge, phone, discipline and front-office regulations
- control of patrols, staff-only zones and security posts
- KPIs for employees, shifts, branches and regions
Compliance and investigations
Violation log and action audit
Each event is saved as a card with evidence: frame, video fragment, zone, time, status, owner, comment and processing history.
logviolations, incidents, statuses and comments
videolink to fragment and confirming frame
reportsexport for quality control and security
ratingunits, cashiers, shifts and regions
Speech analytics
Dialogue quality and speech-standard control
Video shows the fact of an action, while audio explains the context of the conversation. Speech analytics helps check scripts, required phrases, consultation correctness, complaints, conflict topics and service quality.
- greeting, request identification and required phrases
- consultation correctness, dialogue closure and stop words
- cross-sell / upsell, objection handling and product offering
- aggression, raised tone, conflict topics and complaints
- escalation of disputed dialogues to quality control
Required phrase missedoperator 04 · consultation · 10:18
scriptConflict topicraised tone · sent to quality control
riskProduct offeringscenario completed · positive dynamics
ok
Interfaces
Executive dashboard, event log and investigation cards
The interface separates operational work from management analytics: the branch processes events, security investigates incidents, and the central office sees network KPIs.
97.2%network SLA
23violations
12incidents
Executive dashboard
Key network KPIs: SLA, average wait, operational losses, security, incidents and branch ratings.
Cash-process violationbranch #12 · cash desk 2 · video segment
newQueue above thresholdservice hall · 6 clients · SLA risk
SLALong dwell time at ATMself-service zone · camera 4 · high risk
riskEvent log
Filters by branch, region, camera, zone, event type, status, employee, shift and time.
fragmentstep 2 skipped · frame attached
statuscontroller assigned · comment open
Event card
Confirming frame, video fragment, process step, owner, processing history and export.
#12best SLA
#7queue growth
#3ATM incidents
Branch rating
Comparison of sites by SLA, violations, service, security, cash processes and KPI dynamics.
Implementation
Pilot inside the bank environment and scaling to the network
A rational start is 1-2 branches, 10-30 video streams and quick scenarios with clear metrics. The complex FX case can be shown as target process analytics, but the first pilot is better started with queues, SLA, security and ATM events.
Pilot
We start with measurable scenarios and real video streams
For the first stage, video quality, visibility of hands and zones, stream access, information-security requirements, user roles and clear success criteria are important.
queues, waiting time and service-window load
service hall, entrance area and ATM zone
security incidents and links to video
convenient event log for the security team
assessment of false positives and trust in signals
economics of network rollout
01
Survey
Cameras, angles, processes, cash-zone visibility, information security and storage requirements.
02
MVP
Stream connection, basic events, zones, log and first response rules.
03
Pilot
Fine-tuning, test violations, metric evaluation, stabilization and user roles.
04
Rollout
Integrations, reports, ratings, delivery of models to branches and the control center.
On-Prem
Core bank scenario: processing, events, archive, roles and reports inside the bank infrastructure.
Hybrid
Local processing of sensitive video plus centralized analytics and model updates.
Cloud
Possible for less sensitive environments, pilots or centralized reporting when allowed by the bank policy.
AI Factory
Developing bank scenarios without starting each project from scratch
The platform supports a model library, fine-tuning for bank regulations, VLM/GenAI for new scenarios, no-code rules, versioning and delivery of models to branches.
labelingfine-tuningVLM/LLMmarketplacerollout
Integrations
Connection with internal bank systems
Events and KPIs can be sent to VMS/NVR/CCTV, core banking, cash software, CRM, Service Desk, SIEM/SOC, BI/DWH, LDAP/SSO and internal APIs.
VMSCore bankingCRMService DeskSIEMBISSO
Want to test the platform in 1-2 branches?
We will agree on a fast pilot scenario: queues and SLA, security, ATM, staff or cash process. We will validate the value on real video streams inside the bank environment.