Driving Adoption of Facial Recognition Attendance

How giving users control over an AI feature drove full feature adoption.

Company

SYNAPSIS

Year

Q2 2025

Role

Product Designer


Team

Business analyst, mobile engineer, me

Key Result

ADOPTION

Legacy fingerprint attendance fully replaced

After the improvements were released, facial recognition fully replaced the old fingerprint system.

OPERATIONS

Zero HR support tickets

Three months after release, there were no HR support tickets about facial recognition problems.

WORKFLOW

No more manual HR work

With one data source, HR doesn’t need to merge attendance data every month.

Background

We were building an attendance management app for a client. At some point, the team added a facial recognition model to the existing check-in and check-out flow without a designer involved. By the time I was brought in, it was already live. The first signal that something was wrong came from the HR report.

  • The model wasn't sensitive enough to detect users' faces.
  • Users gave up on checking in through the app and went back to fingerprint.
  • HR had to merge attendance data from the app and the fingerprint machines by hand every month.
Face recognition UI before improvement
Face recognition UI before improvement

My Role

I worked with a business analyst and a mobile engineer. My part was:

  • Finding out why users dropped the feature, from the HR report and talks with the HR department
  • Running a design workshop with the BA to agree on the problem and the direction
  • Designing the capture flow, the system status, the countdown ring and the haptic pattern
  • Handing off the whole interaction to the engineer as a Protopie video

Insight That Sparked The Changes

From our talks with the HR department, we found two user behaviours that came from how facial recognition was implemented.

  • Users kept moving their phone around to get their face detected, because the app never told them what it was seeing.
  • Users struggled to time the shutter button after their face was detected. Tapping the button moved the phone a little, and even a slight movement reset the detection.

So the model wasn't the only problem. Users had no feedback to work with, and the manual shutter made things harder.

Constraint That Focused The Thinking

At that time, we couldn't change the facial recognition model because of limited resources. Instead of asking how to make detection better, we asked how to help users work with the detection we already had.

Leveraging user control through system feedback

Exploring The Ideas

I ran a design workshop with the BA using diverge to converge thinking. We listed as many ideas as we could to fix detection and shutter timing, then narrowed them down based on what the model could actually support.

One concept we liked was a progress bar that moves based on how well the face is recognized. We dropped it because the face recognition endpoint didn't return any recognition level, so there was nothing to drive the bar. The idea of showing progress stayed with us though, and it came back later as the countdown ring.

In the end we went with automatic capture, because it removes the shutter timing problem completely. For that to work, users need to trust what the system is doing, so feedback became the main focus of the design.

Solution Overview

Clear system status

We designed the experience around Visibility of System Status, so users always know what the model is doing. The screen shows three states: Detecting, Not detected and Detected. Cases like glasses and masks were already handled by the existing feature, so we could focus on the feedback.

Face recognition UI after improvement
Face recognition UI after improvement

Countdown ring from detection to capture

Once the face is detected, a rounded progress ring counts down to the photo capture. Users know how long they need to hold still, and they don't have to guess when to tap anymore.

This is the progress bar idea from the workshop. Since the endpoint can't tell us the recognition level, the ring follows a fixed 1 second countdown instead.

Automatic photo capture with haptic feedback

We adopted the haptic feedback used by QRIS* to confirm a successful QR scan, and extended it into two stages. The first one tells users their face is detected, and the second one confirms the photo is taken. People already know this pattern from paying with QRIS, so it didn't need any explanation.

*QRIS is Indonesia's national QR payment standard, used by millions of people every day.

Haptic feedback interaction
Haptic feedback interaction

Testing With IT and HR

Before the full release, we tested it internally and then ran a pilot with the IT and HR departments. Pilot users said holding still for 2 seconds felt too long, so we cut the countdown to 1 second.

See It In Action

What I'd Do Differently

All of our insight came through HR, not from employees directly, and we didn't have any numbers from before the redesign. Next time I'd talk to a few employees myself and track detection success from the start, so the impact can be measured.

Better Days Are Yet To Come.

© 2026 Sofyan Hidayat•Built with Fumadocs and Base UI