Design for A.I.-powered Security Notifications: Tailgating Events
UX Designer. Working closely with a peer UI designer, backend, frontend developers and AI engineer
~2 months
Achievement
Launch Tailgating event. User get instant notification when there’s a tailgating event on their platform and phone with 97% accuracy.
Contribution
- Participated the technical discussion and co-design the event logic
- Created prototypes and lead stakeholders reviews.
Impact
- Introducing contextual thinking within the technical team to ensure that the product retains 'user agency' even in automated processes
- Generate the framework that makes onboarding the following security events easier
Background
Prior to addressing Tailgating, Umbo had already earned the trust of customers with several security alarm features, boasting a 99.5% accuracy rate in pixel-level person intrusion AI detection. These features included "Someone is here," "Scaling the wall," and "Loitering."
During an interview with an enterprise user, we received a new request to detect "Tailgaters" using our AI technology.This request elevated our event detection to a new level.
The system not only needs to determine if they are human but also need to ascertain "is this person an employee? (who they are)" and "does this person try to enter the door (what they are doing)" from the video.
Technology Feasibility
As there are numerous identification and behavior detection algorithms, our challenge is to provide accurate alerts that can genuinely assist the security center in these complex scenarios.
Our exceptional AI engineering team proposed a technical solution design that combines our person segmentation AI model with the integration of a card ID identification system capable of determining passengers' identities
UX Challenges
Users already had an existing mental model based on our existing AI features. However, Tailgating conditions are much more complex.
The challenge is to set the right user expectations by effectively communicating the limitations and possibilities through features and clear interactions.
Interview to collect the real scenarios
Through collaboration with users, we've gathered various real-world user scenarios related to Tailgating events. These scenario-based discussions have enabled us to re-evaluate our technical solutions and jointly define the system logic, taking into account user needs and technical feasibility.
Develop the possible “errors”
Just like the relationship between humans, communicate the “errors” could not only make the other one feel respectful but also help them calibrate their expectations.
With the above scenarios, we could presume the possible errors, design and evaluate the experiences we would like to deliver to the users.
The visual timeline created by designers bring the team on the same page while discussing this complex technical logic.
For instance, during busy periods, there may be numerous entries and card swipes. Below is a visualized timeline that aids in the discussion of how the system determines an event based on various conditions of person entries and card swiping.
This visual structure enables us to brainstorm various possible and extreme scenarios and design the logic accordingly.
Outcome
Throughout the feature development process, we forged strong relationships with the customers who requested these features. We engaged in co-designing the feature and successfully delivered it to our customers within the established timeline.
Over 60 cameras have been installed and are actively utilizing this feature to enhance security by transmitting Tailgating alerts to the security team.
Most importantly, our customers are enthusiastic about providing feedback on our platform to collaborate on improving the AI models.
