EFL, a leasing company operating in a highly competitive and regulated financial environment, wanted to automate the extraction of unstructured information from leasing documents. Until now, monthly document monitoring required manual review of around 100 PDF’s, consuming many hours of expert time, limiting how frequently they could be reviewed. The process was lengthy, inefficient and prone to human error.

Softwise.AI was asked to design and implement a system that could automatically extract key financial data from leasing documents and deliver standardized outputs, while complying with strict banking and RODO requirements.

Key Challenges

• Manual, slow competitive monitoring

Monthly analysis of around 100 documents was performed manually, which was time‑consuming and error‑prone, and made it hard to scale the process or run more frequent analyses.

• Heterogeneous, messy input data

Documents from more than 40 sources came in very different formats, layouts, and terminologies, often as scanned or photographed documents, making traditional template‑based OCR brittle and expensive to maintain.

• Strict security and compliance requirements

As part of a banking group, EFL needed the solution to be aligned with internal AI, cloud, and security committees, as well as RODO, including anonymization of personal data and controlled access to cloud resources.

• Need for production‑grade robustness

The solution had to move beyond prototype level and provide controlled retries, quality monitoring, predictable processing times, and clear operational procedures for business users and IT.

Solution

Softwise AI designed and implemented an extraction and analysis pipeline for leasing documents, built on Azure and large language models.

Business workflow

• The business user uploads documents to a dedicated inbox container in Azure Blob Storage.

• From an authorized email address, the user sends a trigger email to start processing the current batch.

• The system runs automated extraction.

• Once processing completes, the user receives an email with:

  • JSON files: a short version containing extraction results and a full version containing confidence levels and metadata.

  • An Excel file with formatted, human‑friendly tables for quick verification.

This makes it possible for a business analyst to turn a batch of raw documents into a complete monthly report in a few minutes, without manual copy‑paste or scripting.

Architecture

OCR technology and model inference together extract all required fields using prompts tuned to leasing terminology. The system then validates the data, enforcing mandatory field checks and confidence scoring. Instead of template‑based rules, the solution relies on LLM extraction, which has been iteratively refined on real examples to handle widely varying layouts and formats.

Processing is deliberately batch‑oriented rather than real‑time. The solution is optimized for predictable monthly runs where EFL uploads a set of documents and receives a complete report when everything is done. To keep the MVP efficient and cost‑effective, the system focuses on first‑page extraction, which captures the vast majority of relevant information in typical leasing documents. Under the hood, a retry mechanism automatically reprocesses problematic files up to four times and selects the best result based on confidence scores across attempts. 

The pipeline comfortably scales to production volumes, successfully processing 188 PDFs in 13 minutes - about 40 seconds per file when processed in parallel by ten workers and remains stable even above the expected monthly load.

Results

• Extraction accuracy

Target: 85%

Achieved: 94% correct extraction of key data fields across test documents

• Processing time for a 100‑file batch

Target: under 15 minutes

Achieved: 10 minutes

• Rejected file rate 

Target: under 15%

Achieved: 10.5% 

• System availability

Target: >95%

Achieved: 99%

Business impact

For EFL, the system delivers tangible benefits:

• Automation of manual work

Eliminates manual analysis of roughly 100 documents per month, saving an estimated 98,75% specialist time.

• Faster market insights

Monthly reports can be generated within minutes, enabling more frequent and timely reviews of market conditions.

• Standardized, analysis‑ready data

All documents are transformed into a consistent JSON format, making downstream analytics and integration with the data warehouse straightforward.

• Scalability by design

The pipeline is able to handle at least three times the planned volume without infrastructure changes, giving EFL room to expand scope.

• Higher confidence in decisions

With 94% extraction accuracy and explicit confidence scores per field, EFL can rely on the extracted data as a solid foundation for strategic and pricing decisions.

Challenges and how they were addressed

Diverse document formats - radically different layouts and wording.

Response: 

• Adoption of LLM‑based extraction instead of rigid templates.

• Iterative prompt tuning using real document examples.

• Introduction of a retry mechanism and confidence‑based selection of the best extraction attempt.

• Conscious decision to exclude extremely exotic cases where handling cost would outweigh business value.

Performance on larger batches - single‑file processing was too slow for monthly production workloads.

Response:

• Parallelization (up to 10 files in parallel).

• Kafka‑based orchestration and resource tuning resulting in an 89% reduction in processing time.

Compliance hurdles - stringent bank requirements and multiple governance bodies (AI, cloud, security).

Response:

• Early involvement of IT and security teams.

• Parallel handling of approval processes.

• Choice of a local anonymization solution and detailed documentation to support audits.

Summary

Softwise.AI delivered a production‑ready system for automated data extraction and analysis of leasing documents.

The solution combines Azure cloud services, large language models, and domain‑specific validation rules to deliver fast, accurate, and compliant insights.

With 94% extraction accuracy, 10 minutes processing time for 100 files, full automation of the monthly reporting workflow, and alignment with regulatory and Credit Agricole Group compliance, EFL is now positioned to scale its competitive intelligence capabilities and base strategic decisions on high‑quality, standardized data.

We assess our cooperation with Softwise.ai very positively. The company has demonstrated strong technological expertise, good work organization, and a responsible approach to task execution. The solution not only delivers business value (98%+ shorter extraction time compared to the initial process) but at the same time is fully compliant with our strict regulatory requirements, compliance and data privacy policies.

Krzysztof Polcyn, Director of Innovation, EFL