The Client needed to train an AI model that would accelerate vehicle insurance claims and make gains with respect to costs and accuracy.
Classifying this dataset of 5000 images correctly was vital.
Qualitest assessed the image dataset sent by the Client using its substantial expertise in data annotation.
Our global delivery solution used expertise at a low cost to classify these images quickly and accurately.
Our one-time fixed-price engagement that classified 5000 images met aggressive deadlines and exceeded the acceptance criteria set by the Client.
The client is a prominent American financial institution. It offers financial services such as online banking, corporate lending, vehicle insurance, mortgage and auto loans as well as installment sale and lease agreements to its customers. It believes in putting ‘people first’ through its offerings, imparting financial education and delivering excellent customer service.
The Client wanted to speed up auto insurance claims processing by training an AI model to categorize images on its machine learning platform. For this, a dataset of vehicle images had to be categorized accurately by a team of data analysts first. Based on the categorized dataset, the platform would ‘learn’ how to classify thousands of similar images. This would not only cut costs as well as the labor and time spent on manually processing insurance claims but also maintain accuracy.
Without in-house expertise, the Client needed a partner that had substantial experience with data annotation. However, if Qualitest was to successfully finish the project, a dataset of 100 images had to be processed within a two-week timeline with a first-time acceptance rate that was greater than 95%. After which, 4900 images had to be categorized within 5 weeks, with the same accuracy. More importantly, the team had to be scaled up to maximum capacity within a week, with training and infrastructure in place.
The Client selected Qualitest due to our roster of proven quality assurance specialists working at select low-cost locations. Our ability to quickly allocate the necessary quality assurance expertise and infrastructure also played a significant role when being selected for the project.
Qualitest partnered with the Client to understand the requirements of the image categorization task. All images had to be labeled, based on three options provided by the Client. These three options included no damage, a single occurrence of damage or total loss. Labeling these images accurately was vital as the results would be used to train an AI model that would, in turn, process auto insurance claims in a similar fashion but with greater velocity and accuracy.
Along with a one-time fixed price engagement model, Qualitest leveraged its global delivery offerings where managed project offshore teams at low-cost locations would work on this task. With 5000 images to be categorized, a team of data analysts were selected to complete the work in two stages with outputs provided daily for the Client to review.
Before commencing image categorization, the validation of standard workflows and image processing guidelines took place against a sample dataset of images, resulting in the development of a process template. This ‘factory model’ approach was adopted by all data analysts to increase productivity by leaps and bounds while maintaining accuracy levels.
Qualitest wrapped up the image categorization project, offering benefits such as:
*The image chosen for this case study features our Data Annotation team at Qualitest’s offshore offices in Madagascar.
