In-Play redesign

Research and testing

Research focused on how golfers corrected a round while still playing. We used unmoderated, quantitative usability testing in Maze, and I interviewed members of our roughly 200-person beta group throughout the redesign.

The team also tested on courses in Connecticut and Florida. Golfers were using the app in sunlight, with a glove, and while keeping pace with other players. Those conditions made map visibility and access to common corrections central to the design.

Shot and putt controls

I compared controls around the map with a narrow bottom panel that showed only distance and weather. The smaller panel left more of the hole visible, but hid the putt counter. Feedback favored keeping shot and putt controls in view.

I grouped those controls at the bottom, accepting that they would use some map space. In testing, players tried to dismiss a sheet that covered too much of the hole. I lowered its resting height and let it expand for detailed edits, keeping common corrections available without covering as much of the map.

Shot editing

The old app put corrections in a separate Edit Shots mode. I moved editing onto the current hole: select a shot, change its details or position, then return to play. Keeping the hole in view lets golfers check the change against where they actually played.

Putts were harder for the sensors to track and needed more frequent corrections. Quick putt counts stay on the main screen. I gave detailed corrections their own green view, with enough room to adjust both the count and each putt’s position.

PGA TOUR Quality Shots

Arccos had recently raised an investment round that included the PGA TOUR and become its Official Game Tracker. PGA TOUR Quality Shots brought that partnership into the product, letting golfers compare their approach shots with TOUR pros using official ShotLink data.

I connected the TOUR comparison to the golfer’s own shot, showing what made it stand out and marking it on the course map. That made the partnership part of the playing experience rather than a separate promotional feature.

MagicEdit score correction

Some golfers knew their correct score but could not identify which shots were missing or extra. I designed a score-review flow that let them submit the total they knew. Engineering’s model used that total and shot-confidence data to adjust the record, while direct editing remained available for details that still needed correction.

AI Caddie

AI Caddie uses computer vision and machine learning to identify trees, estimate their heights, and map course shapes. That analysis informs its club, target, and strategy recommendations.

My role was to make that analysis understandable during a round. I brought recommended clubs and targets onto the course map, supported by shot patterns and conditions, so golfers could judge the advice in context. A dedicated planning view kept that detail separate from everyday distances and corrections. Engineering owned the recommendation model.

Breakfast Ball DS

I created Breakfast Ball DS around the patterns shared by shot editing, putt corrections, the shot list, and score review. Consistent bottom panels, distance displays, and count controls let golfers carry what they learned from one task into the next. The components became a foundation for the wider Arccos app and remain in use.

Results

After the combined detection and interface update, unusable rounds fell by 40%. Retention rose from 45% to 65% over about 18 months. These were outcomes of the team’s release, with improvements to detection and the experience working together.

Detection would never be perfect. Making both shot editing and score correction part of the core experience gave golfers a way to recover when tracking went wrong.