MTailor Achieves Industry-Leading Fit Accuracy with an AI-Powered Body Measurement App

Challenge

, the custom clothing startup that uses smartphone cameras to take precise body measurements, had built an impressive proof of concept — but accuracy dropped significantly in variable lighting conditions and for body types underrepresented in their training data. Return rates driven by fit issues were eroding margins and customer trust.

The core measurement pipeline relied on a computer vision model that had been trained on a limited dataset, producing inconsistent results for customers outside a narrow demographic range. The team needed a partner who could both improve the ML model and refine the mobile capture experience to guide users toward better scans.

"Fit is everything in custom clothing. A single centimetre of error destroys trust," said MTailor's CTO. "We needed someone who understood both the machine learning and the user experience sides of the problem equally well."

Solution

Winklix augmented MTailor's existing team with ML engineers and mobile developers to rebuild the measurement pipeline from the ground up, expanding the training dataset and redesigning the in-app capture flow. The engagement ran over 8 months.

Dataset Expansion & Model Retraining

Winklix built a data collection pipeline to source and annotate over 50,000 new training samples representing a wider demographic range. The TensorFlow model was retrained with augmented data and improved preprocessing, increasing accuracy by 34% in controlled tests.

Computer Vision Pipeline Optimisation

The OpenCV-based image preprocessing layer was overhauled to handle variable lighting, background clutter, and clothing occlusion more robustly. Real-time feedback cues in the app guided users to optimal scan conditions before capture.

Mobile UX Redesign

Winklix redesigned the in-app measurement flow with step-by-step visual guidance and instant quality scores for each scan. Users could retake poor-quality scans immediately, reducing the volume of low-confidence measurements reaching the production pipeline.

Outcome

Within four months of the updated app reaching production:

• Fit-related return rate dropped by 42%.
• Measurement accuracy improved to within 0.8 cm on average.
• Scan quality scores improved by 55% across all user demographics.
• Customer repeat purchase rate increased by 31%.
• App Store rating improved from 3.9 to 4.7 stars.