Transport Monitoring System — how a “small” feature sped up incident processing by 34%

This is a case study about a transport monitoring system: a network of AI cameras managed by operators. The system helps prevent road incidents, minimize consequences and catalogue cases for analysis — making the city safer.

The Anthracite team and I spent a year working on the interface with the client. After the MVP release the database grew, the AI model got smarter — and we kept adding new features. Here are two stories from that year: how we designed icons through research, and how one “small” feature unexpectedly delivered a big result.

Part 1. Camera icons: does research pay off?

Camera icons are the key way of notifying operators about incidents detected on roads and streets. There are many camera types, each specialized: one analyzes faces in footage, another catches traffic violations. To report incidents quickly and dispatch emergency services, an operator must instantly read the camera type, category, function and status on the map.

The task: design an icon set operators could recognize at a glance. There were plenty of challenges: 15+ camera types, main categories, AI module presence, size and orientation variations, statuses and incident priorities. We also needed a simple process for adding new cameras to existing screens and managing updates.

We started with research: a good chunk of time went into studying how cameras look in real life. That gave us the foundational elements for depicting each type.

Even though the system runs on AI cameras, it’s managed by people. So we ran several iterations of usability testing with operators and refined the designs for those who rely on them daily.

We set >80% recognition as the threshold: chasing 100% made no sense — in real scenarios operators work with several cameras at once rather than identifying them one at a time. With this benchmark we finalized the new icon designs.

The icons became part of the design system: a well-organized component set that sped up the assembly of new screens.

In parallel we documented design principles, guidelines and rules for the design team and developers — keeping the design and its implementation consistent.

And to simplify implementation, we built prototypes showing exactly how new camera features should behave in the product:

Extra time spent on research and integrating icons into the design system may look like an unnecessary investment — but the long-term payoff is undeniable: more confidence in decisions and noticeably faster updates, modifications and new layouts.

Part 2. A “small” feature worth a big 34%

Imagine a car accident captured by ten cameras from different angles. Until recently, operators manually merged those streams into a single incident report. Without it the system clogs up with unclassified incidents and efficiency suffers. Even once the AI learned to find and merge related incidents, validation still fell to the operator.

The task: cut merging time and minimize errors. But the main challenge was for the feature to fit seamlessly into the workflow: the system runs 24/7, and we had no right to break the existing efficiency of keeping order in the city.

Discovery

We observed operators through their workday and ran in-depth interviews — giving us the problem from several angles. Key findings from observation:

• operators abandon tasks midway when something urgent comes in;
• finding the right incident takes far too long;
• merging errors create backlog congestion.

And from the interviews:

• unfinished tasks cause anxiety;
• chasing KPIs sometimes leads to errors and frustration;
• full incident documentation eats a lot of time;
• some related footage simply gets overlooked.

It became clear: operators are already overloaded with on-screen events — especially at rush hour, when incidents peak. Adding cognitive load would only make things worse. So we based our null hypothesis on mental model theory: familiar patterns from mass-market apps shape expectations and behavior in similar interfaces.

Null hypothesis: familiar interaction patterns from widely used apps will help operators seamlessly integrate the new feature into their workflow.

Operators are ordinary people with ordinary apps on their phones. A quick survey showed which ones — and that steered the hypotheses for our prototypes.

Prototypes

As an alternative we also considered a manual scenario — in case operators preferred to steer the process themselves:

We then consulted the developers and weighed each hypothesis by implementation cost — only the ones promising impact without excess expense went into prototyping.

Tests

We ran moderated usability tests on the prototypes, measuring effectiveness with several metrics: task success rate, task time, and learnability — success rate over time.

Learnability was critical for the client: operators had to get used to the feature as fast as possible, and the transition to the new workflow had to go unnoticed. So we gave learnability results double weight. And although prototype V1 showed a better average time, V3 won on learnability — and became the final solution.

The outcome

Testing confirmed the central hypothesis: familiar patterns outperformed the “manual” alternatives — even though those were closer to the operators’ existing workflow. The production version used a red-bubble-style notification with a photo stack: it clearly signals merged incidents and captures the operator’s attention.

After release, operator workflow efficiency grew by 34%. The feature proved especially valuable in corner cases where operators used to lose 5–9 minutes per task. Post-release observation also showed extra work — like cleaning accidental omissions out of the system — dropped by about 2 hours a week. That’s 8 hours saved per month per operator, and noticeably higher job satisfaction.

Takeaway

Two episodes — one approach: research before design. Icons built on real cameras and operator testing gave the system a visual language you read instantly. And a “small” feature grounded in mental models and validated with metrics delivered +34% faster incident processing.

Art director: Sydorov Alex · design and text: Lisa Furina · Anthracite Studio