Clivet clears customs faster with generative AI on Amazon Bedrock
99.97% Exact match precision and 90% less time on every customs document.
Customer
Clivet, part of the Midea Group, designs and manufactures chillers, heat pumps, rooftop units and air treatment systems, from single homes to large shopping centres, in more than 90 countries.
Challenge
Clivet’s production relies on components imported from all over the world, and every shipment must clear customs. Extracting data from about 100 complex customs documents a month was a semi-manual job: about 30 minutes per document, around 50 hours of work a month and a 15-20% human error rate. Traditional OCR could read the text, but not understand it.
- about 30 minutes per document,
- around 50 hours of work a month
- 15-20% human error rate
Traditional OCR could read the text, but not understand it.
Solution
Omnys built an AI service on AWS that reads each document with Amazon Textract and lets a Claude model on Amazon Bedrock understand it as an expert would. Codes, descriptions, values and references are extracted even from complex, multi-page layouts, with no templates to configure, and flow straight into Clivet’s systems through simple APIs.
Results
Clivet wanted more precise data and faster processing than its manual process allowed. The results:
- • 99.97% exact match precision, compared with a 15-20% human error rate.
- • 90% less time per document: from about 30 minutes to about 3.
• About 45 hours of manual work saved every month.
Customs clearance is now faster and more reliable, and the team no longer spends time on repetitive data entry.
