The evolution of Verizon's AI dashcamfrom a hackathon entry to a life-saving system drivers trust

1. 1

   One video safety product

   Replace third-party dashcams, map replay and generic safety scores with one product.

2. 2

   AI that managers trust

   Earn driver and fleet-manager trust in how the AI classifies events.

3. 3

   A second business pillar

   Build a product line big enough to stand beside GPS tracking.

The two screens below show the change in the product during my time as Head of the global UX Design, Content and User Research team.

Before

Legacy telematics tracking

After

Modern AI and IoT product

The idea came from a hackathon. I developed the core concept with Alessandro Lori, then Head of AI and now CTO of Verizon Connect. I judged and sponsored the team before and after the event while the work was still a concept.

A few weeks after the hackathon, we assessed the idea using the Dan Olsen Lean Product Playbook: identify the customer and need, define why we should solve it, prototype the product, validate it with customers, test the business case, then decide whether to proceed.

We framed the opportunity through the Kano model: reliable video evidence and an explainable safety score as the must-haves, speed and classification accuracy as the performance needs, and proactive alerts that stopped risky behaviour before it became an incident as the delighter worth chasing.

The customer need was concrete. Fleet managers were facing steadily rising insurance costs, needed evidence of liability when incidents happened, and were struggling with unsafe driving they had no reliable way to catch. What they had access to at the time, simple 3rd party dash cameras, separate map replay and generic safety-score reports, was meeting that need poorly or not at all. We set out to unify that experience. The opportunity was a multi-hundred-million-dollar market with a clear revenue target. It justified a significant multi-year investment. The hackathon idea became a funded roadmap item.

The Integrated Dashcam MVP launched with a video browser, map replay, a mobile app and alerts. It worked, but customers did not yet trust it.

We rebuilt the alert language and severity scale. We tested three ways of describing an event with 30 fleet managers and 7 internal reviewers. The bulleted version scored 89% on accuracy and 92% on readability, against single digits for the paragraph version. We shipped it with the reason behind each classification and a speed graph from our own telematics data, so customers could check the context themselves. Customer ratings then trained the video analytics engine, and positive ratings rose across every classification.

The ease of use, the graph to visually see what's happening... just wonderful. Not to mention all the times the camera has saved our business.

That trust became the foundation for a driver-facing AI dashcam with real-time early warning and in-cab coaching, built with our Data Science and ML teams. The alerts had to be instantaneous and understandable, so we designed distinct audio and voice cues for each type of risk. Matthew O'Sullivan, Design Manager, led this work: defining when a tone, a spoken word, or a combination of both was the right call, and how loud, how long, and how urgent each cue should feel to a driver already focused on the road.

This introduced audio UX and its associated user testing to the design team.

In its first week, a customer contacted us after the system detected her drowsiness, slow blinks and dropping head, and alerted her before an incident occurred.

While shipping the driver-facing dashcam, I worked with Product and Engineering leadership to define Reveal AI, a provocative North Star vision built to stretch thinking and guide debate. It was never a committed roadmap. It used a recurring cast, a safety manager, a dispatcher, a fleet manager, living through one continuous day, rather than a feature list, and it shaped how our research team thought about immediate versus long-range questions: Cheryl Abellanoza, PhD, our UX Research Manager at the time, wrote her own account of what that meant for a research practice, and credits the discipline on our team to this period of leadership.

The company I joined as a telematics tracking business has moved well beyond that description. This product line contributed to that change. Verizon Connect's current AI dashcam line now runs dual-facing cameras, real-time driver monitoring for phone use, smoking and fatigue, tailgating and pedestrian-proximity alerts, and up to four peripheral cameras for full 360 degree coverage. This is a significant advance from the video browser and map replay of the original MVP.

I was there through the driver-facing AI dashcam launch and the start of the peripheral camera work that would become the 360 degree coverage. After that I moved on to Contentful and then Canto, before the full peripheral system and the awards that followed shipped. The central point remains: the system had to earn trust before it could develop further. The early work gave the video-intelligence product the foundation it needed to become a second pillar of the business rather than an addition to GPS tracking.

Matthew O'Sullivan

Design Manager. Carried the AI Dashcam design work day to day.

Lorenzo Moschi

Lead Designer on the dashcam experience across its redesign and driver-facing release.

Rachel Ryan

Lead Researcher. Ran the trust study and the alert-language testing the redesign was built on.

Alessandro Lori

Head of AI at the time, now CTO of Verizon Connect. Collaborator on the original concept.

Squad designers joined the work throughout the product's life, and their contribution was essential.