PVID AR24: Improving a high-assurance remote identity verification flow

Before proposing a new version, we wanted to understand the existing experience from every angle rather than redesign it from assumptions.

We started with a detailed audit of the live journey, reviewing the flow screen by screen and identifying what worked, what created friction and where the experience could be improved. We then presented that diagnosis back to the Product team to establish a shared understanding of the problems before moving into design.

We complemented the audit with user testing on the existing product. Watching people go through the real journey allowed us to observe hesitation, instructions they overlooked, actions they misunderstood and the moments where the experience became difficult or stressful.

In parallel, we benchmarked other remote identity verification experiences, including providers that were also adapting their products to the new PVID framework. This helped us understand how an emerging market was approaching similar security and usability constraints.

Combining the expert audit, user behaviour, direct feedback and benchmark gave us the foundation we needed before starting the redesign.

One step stood out with the highest failure rate in our tests: more than 20% of users failed when they had to record themselves turning their head in the requested direction.

The instruction seemed simple on paper, but in practice many users turned the wrong way. The issue became even more noticeable when people were filming themselves, as the mirrored camera view made left and right easier to confuse.

First iteration: make the direction clearer.
Initially, the interface explicitly asked users to turn their head to the right or left. Testing showed that many participants still confused the two directions, so simply making the instruction more visible was not enough.

Second iteration: show the movement instead of describing it.
We removed the notion of left and right altogether and replaced it with a large animated arrow showing the direction to follow, alongside a face illustration performing the expected movement.

The interaction changed from asking users to interpret a direction to letting them mirror a demonstrated movement. The following tests showed a further improvement.

Some interactions still showed meaningful failure rates, but in a standard session with a small group of participants that could represent only one or two failed attempts.

That was enough to identify a potential problem, but not always enough to understand its cause.

I therefore adapted the research approach and ran targeted guerrilla tests around specific gestures, focusing each session on the interactions we still needed to understand better.

We also met AR24 operators in Strasbourg. Because they reviewed failed submissions every day, they had visibility over recurring execution errors that were much harder for us to encounter repeatedly during occasional usability sessions.

Instead of relying on one research method, I adapted the method depending on what we still needed to understand.

Once an iteration was ready, my involvement continued through implementation. I prepared the screens, states and interaction specifications, worked directly with the developer, and ran Design and functional QA across multiple devices.

Once live, the product became a new source of evidence. We used Hotjar to identify remaining friction and continued iterating based on real user behaviour.