How We Automated 30,000 PDF Files a Month for Huawei Indonesia and Saved IDR 750 Million
In 2021, Huawei Indonesia's billing team was manually processing 30,000 PDF billing files every month. The process consumed between 5,000 and 15,000 man-hours annually. Today, the same job takes 3 minutes.
Here is exactly what we built, how we built it, and what it cost versus what it saved.
What problem was Huawei Indonesia trying to solve?
Huawei Indonesia's billing department received thousands of PDF invoices and billing statements every month from multiple sources. Each document had to be opened, read, extracted, cleaned, and consolidated into structured data, manually, by a team of people, every single month. The volume was 30,000 files. The cost in human hours was enormous.
The initial brief was to build an OCR (Optical Character Recognition) system to read the PDFs automatically. This seemed logical, the files were PDFs, OCR reads PDFs, problem solved.
Except it wasn't that simple.
Why was OCR the wrong solution?
OCR would have been the wrong solution because the real problem wasn't reading the text inside the PDFs — it was extracting the right fields, cleaning inconsistent formatting, and consolidating everything into a usable output.
After two weeks of structured requirement gathering across four departments, it became clear that different billing documents had different formats, different field names, and different levels of data quality. A pure OCR approach would have produced raw text dumps that still needed significant manual cleaning. The time savings would have been minimal.
The problem wasn't reading the documents. It was the entire pipeline from raw PDF to clean, consolidated, actionable data.
How did we build the Billing Convert Tool?
We built a standalone Python application with a Tkinter desktop interface — no server required, no internet dependency, deployable on any Windows machine. The tool runs entirely locally, which was critical for Huawei's data security requirements.
The core logic:
- Ingestion: The user points the tool at a folder. All PDFs are loaded automatically.
- Extraction: Custom parsing logic, not OCR, it extracts structured fields from each document based on known document templates.
- Cleaning: Inconsistencies, formatting errors, and missing fields are flagged automatically.
- Consolidation: All extracted data is merged into a single structured Excel file with anomaly flags visible at a glance.
- Output: One clean file. Ready for finance. No manual intervention required.
The entire process 30,000 files completes in under 3 minutes.
What were the results?
The Billing Convert Tool eliminated between 5,000 and 15,000 man-hours of manual work per year. At Huawei Indonesia's internal labor cost, that translated to IDR 750 million saved per month.
The tool has been in active use since 2021. It has never been replaced.
What makes this result significant is not just the time saved, it is the elimination of human error at scale. Manual data entry across 30,000 documents per month produces errors. Those errors create downstream problems in finance, reconciliation, and reporting. The tool removed that failure mode entirely.
What does this mean for your business?
If your business has any repetitive document processing, data extraction, or consolidation task that someone is doing manually every month, it is almost certainly automatable.
The key question is not "can this be automated?" Almost everything can. The key question is: what is the actual cost of the manual process, and what would a one-time automation investment return over 12 months?
In Huawei's case: IDR 750 million per month saved. The tool cost a fraction of that to build.
For Indonesian SMEs, the calculus is different in scale but identical in logic. If your team spends 3 days per month on a task, that is 36 days per year, roughly one full-time employee month on something a script could handle in minutes.
FAQ
What is document automation?
Document automation is the use of software to extract, clean, and process data from documents, PDFs, Excel files, scanned forms, without manual human intervention. It replaces repetitive data entry and consolidation tasks with automated pipelines.
How much does document automation cost in Indonesia?
Simple automation projects in Indonesia typically range from IDR 5 million to IDR 20 million as a one-time cost, depending on complexity. Projects involving multiple document types, custom parsing logic, or integration with existing systems cost more.
Can Python automate PDF processing without OCR?
Yes. Python libraries such as pdfplumber, PyMuPDF, and pdfminer can extract structured text from text-based PDFs directly, without OCR. This approach is faster, more accurate, and requires no machine learning model — as long as the PDF contains selectable text rather than scanned images.
How long does it take to build a document automation tool?
A focused document automation project — from requirement gathering to working tool — typically takes 2 to 6 weeks depending on the complexity of the documents and the number of edge cases to handle.
Is document automation worth it for Indonesian SMEs?
Yes, provided the manual process is recurring and takes significant time. Any business spending more than 2 to 3 days per month on repetitive document processing is a strong candidate for automation. The return on investment is typically realized within 2 to 3 months.