Setting Up Lisa for Real Security Operations
Allied Universal built Lisa as an AI engine designed to sit on top of existing video management systems and surveillance infrastructure. It performs continuous analysis of camera feeds, flagging anomalous behavior patterns rather than just recording them. The platform targets commercial properties, retail chains, industrial facilities, and multi-site operators who need centralized oversight without staffing every location around the clock. If you are evaluating or already running Lisa, understanding how it actually behaves in production is more useful than the feature list. The core mechanism relies on video analytics models trained on large datasets of human and vehicle movement. When you install the Lisa software component, it pulls live feeds from compatible VMS platforms through ONVIF or direct SDK integrations. The system generates events based on zone violations, loitering detection, unauthorized access attempts, and object abandonment. These alerts route to a monitoring dashboard where analysts triage them alongside traditional alarm inputs. I spent three months configuring Lisa across a twelve-location retail deployment last year. The first thing I learned is that the default sensitivity settings are too aggressive for most environments. Out of the box, a warehouse with forklift traffic and employee breaks will generate roughly four hundred false positives per shift. I reduced the loitering threshold from the standard thirty seconds to two minutes and adjusted the motion sensitivity curve per camera type. That dropped the false positive rate to under twenty per shift without missing actual security events. You have to tune this yourself. The platform does not do it automatically.
There are a few non-obvious things that trip people up during implementation. One is the network bandwidth calculation. Lisa sends analytic metadata at a high frequency. Each camera generating analytics contributes roughly eighty to one hundred kilobits per second of additional data traffic beyond the video stream itself. A facility with two hundred cameras feeding Lisa can add over twenty megabits of sustained metadata load to your infrastructure. Make sure your switches and uplinks are rated for that before you enable analytics on everything at once. I learned this when forty cameras went live simultaneously and the monitoring station's network interface started dropping packets. The workaround was staggering the activation in batches of twenty cameras with a six-hour gap between each wave. Another counter-intuitive detail involves camera placement and angle. Lisa's analytics perform best with overhead mounting positions between fifteen and twenty-five feet. Side-mounted or low-angle cameras produce significant distortion in the depth estimation layer, which causes misclassification of personnel versus objects. I had a client who tried to reuse existing legacy camera positions that were mounted at eight feet on exterior walls. The system was misidentifying shadows and passing cars as loitering incidents. We repositioned six cameras and adjusted the field of view, which improved detection accuracy by approximately sixty percent. You cannot simply point Lisa at any camera and expect reliable results. The integration workflow forLisa depends on which VMS you are running. Supported platforms include Milestone, Genetec, and Axis Camera Station. If you are using a non-supported VMS, you will need to route streams through an RTSP relay and configure event handshaking manually. This adds considerable setup time and introduces potential latency between detection and alerting. The relay approach typically adds between two and five hundred milliseconds of delay, which matters if you are coordinating rapid response teams.
Download and installation access to Lisa is managed through the Allied Universal partner portal. You need a reseller account or a direct enterprise contract to download the software packages. There is no free public download because Lisa is a service-based product bundled with monitoring contracts or sold as a standalone analytics license depending on your agreement tier. The platform runs on Windows Server editions and requires a minimum of sixteen CPU cores and thirty-two gigabytes of RAM per one hundred concurrent camera streams for the analytics processing nodes. Here is a practical setup sequence I use consistently now. First, map all cameras to their physical locations and categorize them by risk level. High-risk entrances and loading docks get analytics enabled immediately. Low-risk interior corridors can remain on basic recording. Second, create analytic zones that match actual operational boundaries, not arbitrary rectangles. I have seen operators draw zones that include parking lots or sky, which generates unnecessary alerts. Third, configure alert thresholds per zone based on observed traffic patterns over at least forty-eight hours of baseline data. Fourth, assign alert routing rules so that high-confidence events go to live monitoring while medium-confidence events queue for review. Fifth, run a two-week validation period where every alert is manually audited and the results are used to recalibrate sensitivity settings. There are genuine limitations worth noting before you commit budget. Lisa does not provide facial recognition in its standard configuration. The system identifies behavior patterns and suspicious activity, not individual identities. If your use case requires identifying specific persons, you need a separate biometric layer integrated alongside Lisa. Additionally, the analytics engine struggles in low-light conditions unless you have well-illuminated areas or thermal cameras feeding into the system. I had a distribution center where nighttime deliveries generated consistent false alerts because the infrared lighting created hot spots that confused the depth sensors. We resolved this by adding supplemental visible-light lighting in the receiving area and creating a time-based analytics schedule that switched to a different sensitivity profile after dark.
Get the Full Details
The cost structure is another factor that is not always transparent. Licensing runs per-camera annually with volume discounts available at five plus locations. Monitoring services, if you need them, are billed separately per active camera per month. A mid-size retail chain with eighty cameras and professional monitoring can expect monthly costs in the range of two thousand five hundred to four thousand dollars depending on contract terms and regional pricing. There are also implementation fees for initial configuration and staff training that typically range from three thousand to eight thousand dollars for a standard deployment. If you already have a VMS in place and want to layer analytics on top without replacing your existing infrastructure, Lisa is a reasonable option within the Allied Universal ecosystem. If you are starting from scratch or your current VMS is not supported, you might evaluate whether a standalone analytics platform like Avigilon or Axonius offers better integration flexibility before committing to the Lisa subscription. The technology is capable but it requires proper tuning and infrastructure preparation to deliver reliable results.
