The company, a global leader in the supply chain of building materials, faced a significant operational challenge due to the vast diversity and volume of its stock-keeping units (SKUs), which exceed 100,000. With vendors across 35 countries, the complexity of their inventory management was compounded by frequent inquiries from customers who sent photographs of home furnishing SKUs, asking about availability and lead times. These photographs often feature intricate designs (imagine different wood patterns on table tops or doors), including specific veneer line patterns and colors that require precise matching.
Their process to match each customer-submitted image with the correct SKU was time-consuming and prone to errors. The errors were further exacerbated by variations in image quality — pictures were often taken by phone cameras of varying resolutions, unsuitable lighting conditions, and different camera angles. Images were often blurred, not showing the intricate line patterns. Staff members spent days trying to identify matches manually, a task that not only slowed down response times but often led to the procurement of incorrect designs or missing out on the right options. Dissatisfied customers were common, impacting the company’s reputation and operational efficiency.
The solution needed to stem the errors and save the huge operational effort the company was spending on this activity.
Artificial Intelligence was needed to play a role here. However, the technology had its limitations. It was not an ordinary problem where an approximate image match would have sufficed. A major re-engineering effort was needed since we were tasked with finding exact matches after comparing a low-quality image against 100,000+ SKUs, and we aimed to do it in under 3 minutes. Moreover, the operational costs had to be kept low.
And that is exactly what our team delivered — a cloud-based solution that could scale using microservices, leveraged open-source computer vision models, combined with proprietary machine learning approaches. The first step was to get around the image quality problem - here, we combined elaborate image engineering methods with fine-tuned convolutional neural nets to process images from varying resolutions, under different light conditions, and in some cases, even images with filters on them. To keep costs low, the entire solution was designed to run as a secure, serverless solution providing matches on-demand.
The customer has gained over 99% operational efficiency in the pattern-matching process. The pattern matches done by our solution were found to be 70% more accurate during our benchmarking tests. Their operational resources are optimised and their turn around time to customer queries has reduced drastically, resulting in better customer satisfaction.
Ready to transform your operations with AI? Reach out to us today and discover how our tailored AI solutions can elevate your business to the next level.
At itmtb, we are reliable provider of cloud, virtual reality, technology consulting and software and IT development services to businesses and startups across the globe. We have the right expertise to help your business with the right technologies. Write to us at hello@itmtb.com if you think we can help.
Stay tuned for more technology updates.