Clivet speeds up customs clearance of components with generative AI on Amazon Bedrock
Omnys automated data extraction from customs documents with a serverless pipeline on AWS: 99.97% precision and 90% shorter processing times.
Customer
Clivet is a European center of excellence in solutions for total comfort and energy efficiency. It designs and manufactures chillers, heat pumps, rooftop units and integrated air treatment systems, from single homes to large shopping centers. Clivet operates in more than 90 countries with 35 agencies in Italy, 7 branches in Europe and 160 service centers, employs more than 1,000 people and exceeded EUR 328 million in net sales in 2023. Since 2016 it has been part of the Midea Group.
Challenge
Clivet’s production also relies on goods and components coming from foreign countries, and every shipment must clear customs. Data from customs documents was extracted through a semi-manual process: long processing times, difficulty in keeping uniform quality standards, a high risk of human error and limited ability to scale with volumes.
Clivet needed a solution able to interpret, recognize, extract and store in the corporate data warehouse, automatically and with maximum precision, the data contained in customs documents. There were four key requirements:
- Complex documents: many data points to recognize, variable content and critical entities to identify precisely.
- Reliability and control: automate without losing quality and traceability of information.
- Integration: feed the extracted data into existing business systems consistently and automatically.
- Cost control: balance efficiency and budget with an approach as serverless as possible.
Solution
Omnys built on AWS a set of APIs that integrates Amazon Bedrock into an asynchronous processing pipeline based on AWS Lambda. When a new document arrives, the system automatically invokes the generative model, which extracts and identifies all relevant entities, such as codes, descriptions, values and references, even from complex or unstructured layouts and from information spread across several pages.
Unlike traditional OCR, which reads text mechanically, the solution understands its meaning and context. Omnys configured the models with advanced prompt engineering and, where needed, targeted fine-tuning on the customer’s data. Extracted data is returned in a structured format through REST APIs and feeds the corporate data warehouse directly.
Key advantages include:
- Beyond traditional OCR: higher precision and flexibility on complex formats, with no templates to configure for each document type.
- Easy integration: structured output ready for business workflows.
- Scalability and flexibility: a solution that easily adapts to evolving business needs and growing workloads;
- Cost efficiency: serverless, pay-per-use architecture.
- Scalability: the solution grows with volumes and extends to other documents, such as invoices, delivery notes and certificates.
Results
- 99.97% precision in entity extraction and recognition, a level hardly achievable with the technologies used before.
- -90% process time, moving from a semi-manual to a fully automated approach.
Customs clearance of components is now faster and more reliable, and staff no longer spend time on repetitive data entry.