Document Processing
Occidental Petroleum needed to avoid an expensive, lengthy manual review of 1.5 million lease documents and create a scalable way to classify and extract metadata from millions of PDFs.
Reported impact
Oxy saved more than $4 million and 12 months by automating review of 1.5 million documents, and its Atlas Vector Search proof of concept showed sub-200ms retrieval across over 100GB of text.
Occidental Petroleum built an automated document-review pipeline on MongoDB Atlas that combines event-driven cloud services, OCR, and large language models to process land-lease agreements at scale. Physical documents are scanned into the enterprise content management system as PDFs, automatically moved into Amazon S3, and then processed concurrently by AWS Lambda functions. Those functions call Microsoft Azure AI Document Intelligence to extract text and metadata through OCR so downstream classification models can assign each file to one of 140 categories and pull out dates or legal terms. MongoDB Atlas started as the linking layer between physical and digital records but quickly became the system state backbone as engineers switched to stateful documents to recover gracefully from edge cases such as very large PDFs. That change let the pipeline handle thousands of Atlas requests per second while keeping classification, metadata extraction, and retries on one resilient operational store.
Scanned PDFs move from Oxy's content-management system into Amazon S3, AWS Lambda functions invoke Microsoft Azure AI Document Intelligence for OCR and metadata extraction, classification logic categorizes files and extracts legal attributes, and MongoDB Atlas stores stateful processing records so failures can resume cleanly.
The automated system that the company built using MongoDB Atlas saved Oxy $4 million and 12 months.
Using MongoDB Atlas Vector Search, they ran a proof of concept that resulted in a retrieval time of less than 200 milliseconds.
“We went from making a few requests to MongoDB to making thousands of requests a second to MongoDB, and it never really caused an issue.”
“What was amazing for us was that we were able to get something up and running in a matter of days.”
Review how this reference was scored and trace each normalized claim back to its documented source.
Scoring methodology
Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.
Traceability
Field coverage shows which facts each source supports. Dates are shown only when documented by the source.
Related
MongoDB Sample Archived · Apr 24, 2024 Eni Eni needed to make multilingual subsurface documents and extracted scientific data usable across geoscience teams, but its relational platform struggled with hundreds of thousands of documents, terabytes of unstructured… more than 30,000 employees across 69 countries MongoDB Atlas · MongoDB Atlas Charts · MongoDB Atlas Search MongoDB Sample Oct 8, 2026 Navisio AI Rapid adoption of Navisio AI's platform triggered severe latency spikes, elevated CPU and RAM consumption, and costly vertical scaling. As it prepares to scale to 1,000 by the end of the year, and eventually toward 5,000 users, the team is… MongoDB Atlas · MongoDB Atlas Vector Search 
In-house Sample Sep 22, 2026 Repsol Repsol needed to anticipate service-station store demand, reduce stock-outs and waste, and cut manual delivery-note entry so station teams could spend more time on customer service. A total of 85% of recommendations are accepted unchanged, and an additional 14% are accepted with quantity… Automatic delivery note reading · Suggested ordering 