How to Read Indonesian Tax Invoices and Invoices Automatically with AI

Every month, finance teams receive a stack of tax invoices (faktur pajak) and invoices from vendors. Some arrive as PDFs, some as scans, and some as phone photos sent over chat. The contents are then retyped into a spreadsheet or ERP, line by line, before they can be matched and paid.

The work looks simple, but it takes time and invites mistakes. One swapped digit in a VAT amount or invoice number can lead to differences during reconciliation or tax reporting. An AI document reader moves that data automatically and, more importantly, checks whether the numbers make sense before they reach your system.

In short:

• The problem: tax invoice and invoice data is still widely retyped by hand.
• How AI works: it reads the document, extracts key fields, checks the math, and sends the data to your system.
• The key: checking lines, DPP, VAT, and totals, not just reading text.
• The limits: invoice validity and tax treatment are still confirmed through official channels and your tax team.
• How to start: test on samples of your own documents before rolling out fully.

Tax Invoices and Invoices: What Is the Difference?

The two often arrive together, but they serve different purposes. An invoice is a commercial bill from seller to buyer. A tax invoice (faktur pajak) is proof that Value Added Tax (PPN) was collected, issued by a VAT-registered business (Pengusaha Kena Pajak, or PKP) when it delivers taxable goods or services.

Tax invoices are created through the tax applications of the Directorate General of Taxes (DJP), including Coretax DJP, part of the renewal of Indonesia's core tax administration system. For the receiving team, this means the format is fairly uniform. Invoices are the opposite: every vendor has its own layout, terms, and table structure.

Data That Is Usually Extracted

The fields you need vary by company, but they usually include:

• Document number and date
• Seller and buyer identity, including NPWP (tax ID)
• Goods or services, quantities, and unit prices
• Tax base (Dasar Pengenaan Pajak, or DPP)
• VAT amount, and luxury goods sales tax (PPnBM) if any
• Total amount, payment terms, and due date
• Purchase order number or other references, if the vendor includes them

For invoices, the PO number and due date usually matter most because they drive matching and payment scheduling. For tax invoices, the focus is on the parties' identities, DPP, and VAT.

Why Manual Entry Is Still a Problem

The main issue is not only time. Manual entry is hard to check at scale:

• Typos are hard to spot. A wrong number still looks reasonable in a spreadsheet.
• The math is rarely rechecked. Staff seldom re-add every line to confirm the total.
• Formats keep changing. A new vendor means a new layout to learn.
• Queues pile up at month end. Workloads spike near book closing and reporting deadlines.

Basic OCR helps turn images into text, but it does not solve this. The output is still raw text that has to be sorted, and OCR does not know whether DPP plus VAT equals the stated total.

How AI Reads Tax Invoices and Invoices

An AI document reader works in several stages:

1. Receive the document. The file can be a PDF, a scan, a phone photo, or a Word file.
2. Read the content and layout. The AI recognizes the text and its structure, including tables of goods and services.
3. Extract the fields. Data is extracted by document type, or by a template you create for a specific format.
4. Check the results. Validation rules confirm that the numbers agree, for example lines, DPP, VAT, and total.
5. Flag what looks wrong. Documents whose math does not add up or whose data is incomplete are flagged for human review.
6. Send the data. Checked results go to Excel, Google Sheets, or your company systems through an API.

Stages 4 and 5 are what separate trustworthy automation from plain OCR. The goal is not to remove human review, but to focus the team's attention on the documents that really need it.

The Checks That Make Data Trustworthy

Reading numbers correctly is only half the job. The other half is making sure those numbers agree with each other and with the data you already hold.

Some of these checks are done by the document reader. Others, such as matching against purchase orders, usually run in your ERP or finance system after the data arrives.

Lookdocu, for example, checks lines, DPP, VAT, and totals on tax invoices and invoices. If the subtotal plus VAT does not equal the stated total, the document is flagged before its data is sent to your system.

What Remains the Team's Responsibility

Automated reading speeds up data entry, but it does not replace tax and internal control functions:

• Invoice validity. Confirming that a tax invoice is valid is done through official DJP channels, not by a document reader.
• Tax treatment. Whether input VAT can be credited and how to record it remain decisions for the tax team under the applicable rules.
• Payment approval. Payments should still go through your existing approval flow.
• Data access. Tax invoices and invoices contain NPWP and transaction data, so control who can view and download the results.

Integration with Company Systems

For small volumes, exporting to Excel or Google Sheets is often enough. The team can review the results in a spreadsheet and then import them into the accounting system.

For larger volumes, the flow can be fully automated. Your system sends files to Lookdocu through a REST API and receives checked data through HMAC-signed webhooks. The signature lets your system confirm that the data really came from Lookdocu and was not altered on the way before it is processed.

How to Start

1. Pick one flow. For example, vendor invoices going to accounts payable, or input tax invoices for the VAT recap.
2. Define the fields you need. List the fields and their destination format, such as spreadsheet or ERP columns.
3. Collect sample documents. Take samples from several vendors at different quality levels: original PDFs, scans, and photos.
4. Measure accuracy per field. Compare the extracted results with the correct data, field by field.
5. Set review rules. Decide which documents pass straight through and which must be reviewed.
6. Connect to your systems. Start with spreadsheet exports, then move to API integration once the process is stable.

For companies, Lookdocu offers an accuracy test with 50 to 100 sample documents. The Lookdocu team measures accuracy per field and tunes templates and checks, so the decision is based on numbers from your own documents. Teams that want to try first can use the free plan with 30 AI credits a month.

Frequently Asked Questions

Can AI read a tax invoice from a phone photo?

Yes, as long as the text is clearly visible. Blurry, cropped, or glare-affected photos lower the quality of the results. That is why documents with doubtful results should be flagged for review.

How is this different from basic OCR?

OCR turns images into text. An AI document reader also understands the document's structure, extracts the right fields, and checks whether the numbers agree.

Does the tool confirm that a tax invoice is valid?

No. A document reader checks the content and the math. The validity of a tax invoice is confirmed through official DJP channels.

What about invoices from vendors with unusual formats?

Create a template for that format, then test it with several samples before using it routinely.

What accuracy can we expect?

Accuracy depends on the quality and variety of your documents. The most honest way to find out is to measure it directly on samples of your own documents, field by field.

Conclusion

Reading tax invoices and invoices automatically is not just about turning images into text. The value lies in the checks: making sure lines, DPP, VAT, and totals agree, then flagging the documents a person needs to see.

Start with one flow, measure accuracy on your own documents, and let your tax team keep the decisions that belong to them.

Want to try it on your own tax invoices and invoices? See Lookdocu or start free at lookdocu.id.