Starting point: manual transcription with in house resources
Our customer is a company specialized in testing, inspection and certification founded in 1828. It operates in a variety of sectors, including Building & Infrastructure, Agri-food & Commodities, Marine & Offshore, Industry, Certification and Consumer Products.
We partnered with their North America office who was looking to automate a core process. For their asset management unit, they had developed a field collection tool for their assessors to capture information on site on various assets such as water heaters, kitchen units etc. They would then manually label the make, model & serial details. Other than the very slow and manual process, they had realized they had a data quality issue, with a 7% error rate in the serial number transcriptions. They started to look at ways to automate this process with the objective of reducing the error rate, freeing up their assessors time spent on a highly manual tasks and allowing them to focus on their core job, and therefore also reducing the time required for them to onboard new customers.
Scaling workflow with super.AI
After starting to look at and test various providers, they decided to go with super.AI for a Proof of Concept (POC). Our API capability allowed them to upload a significant amount of data, in contract to the quite manual alternative the other providers were offering (large Excel files). They were also attracted by the broader solutions we offered on top of our crowd labeling capabilities .
For the POC, they shared several tens of thousands of data point and wanted to extract the following attributes from the information captured by the assessors: manufacturer, model number and serial number.
As part of the super.AI solution, we provided them with the following setup:
- We built a custom data program to take photos of the labels on equipment and extract the make, model and serial number
- We worked with the customer to define instructions and identify edge cases. We were also able to easily update the project configuration when the customer realised they needed extra field ('manufacturer date')
- We automatically built a training to make sure that labelers understood the task and could perform it effectively
- Customer was able to upload the data programmatically via our API, and also fetch the results automatically
- Each photo was labeled by multiple labelers to ensure accuracy
We couldn't have handled the customer workload in house. super.AI helped us handle it , while also providing a labor rate that benefits us at an amazing accuracy. We went from something we were not able to do to something we don't have to worry about! This partnership has allowed us to significantly expand our workflow
As part of the POC process, we processed 100,000 data points for the customer at a 99,98% accuracy.
Thanks to our API solution, we were able to solve capacity, accuracy and improved labor costs for the customer.
Given the new automated workflow, they were also able to take this highly manual repetitive task off the hands of our assessors and allow them to focus on their core expertise. They were able to onboard significantly larger customer deals while also reducing customer onboarding time by 50%.
The customer is now accelerating the number of data points sent to us for processing.
Everyone loves the ideas, really easy to sell 6x improvement in accuracy at a cost savings and getting our assessors back in the field looking at buildings quicker. By the end of next year I could see us pushing 100X more images to you.