Marcus Biel.

AI document processing

Turn incoming documents into data your team can use

Every PDF order means another round of opening a file, looking up item codes, and copying quantities. I help you cut down on that manual work.

I build software that reads the fields you need, flags anything uncertain, and sends checked data to the system your team already uses.

From incoming document to usable data

An emailed purchase order needs to reach your order system. A delivery note needs to match an existing order. A form might contain details for several departments.

I connect those steps so your team can see where each value came from and what happened to it.

For example, the software can help with these tasks.

  • Entering orders. Capture the order number, customer, line items, quantities, delivery address, and requested dates.
  • Matching delivery notes. Extract references and line items and match them against existing records.
  • Processing requests and forms. Supply contact details, the request and other required information as structured data.
  • Sorting incoming documents. Identify the document type and route it to the right queue.

I assess which steps are worth automating using your actual document volumes and representative examples.

From an emailed PDF to an approved order

The sample order below shows where a reviewer comes in. Twelve mounting plates match the original. The six bearings need a closer look before the order is ready for the receiving system.

Here’s one way to handle an order from arrival to approval.

  1. 1. Receive the order

    A PDF arrives through an agreed mailbox, an upload or an existing system. The application routes it to the order workflow.

  2. 2. Extract the required details

    The software extracts the agreed fields, such as order number, delivery address, item codes and quantities.

    The review screen shows extracted data alongside the relevant passages so staff can check it against the original.

  3. 3. Apply business rules

    Are required fields present? Do the customer and items match your master data? Do quantities and dates satisfy the agreed rules?

    Missing fields, unknown items and conflicting information are flagged for review.

  4. 4. Review and approve the order

    Initially, you can require approval for every order. Staff check the extracted details, make corrections and approve the transfer.

    I use the pilot results and your business requirements to help you decide which cases can proceed without manual approval.

  5. 5. Transfer the data

    The application sends the approved data to your order management system through an API or produces a file you can import.

    Staff can see which orders arrived, which need checking, and which transfers failed.

Check the item code against the sample order

Compare the item code in the draft with the original order. One character needs a closer look. The demo uses fictional data and runs in your browser.

Original document

Open PDF
WERKSTATT NORD

Order

No. BN-1042 · Sample document

Line items on the original order
ItemCodeQuantity
Mounting plateMP-2412
Round bearingRL-186

Please deliver both items together.

Draft order

BN-1042
Mounting plateMP-24 · 12 units Find line 1 on the original
Confirmed
Round bearingRL-1B · 6 units Find line 2 on the original
Awaiting review

The last character could be a letter or a digit. Which code appears on the original?

Try it on one document workflow

Start with one manageable document workflow and a result you can assess.

Before the pilot, I ask which questions you need it to answer.

  • Which document types and layouts will we test?
  • Which fields do we need, and which mistakes would matter most?
  • Who checks or approves the result?
  • What format does the receiving system need?
  • Do we need a live integration, or will sample output do?
  • How will we judge the results?

You receive a working demonstration, an evaluation and the information you need to decide what to build next.

Measure accuracy on your documents

I use representative test documents held separately from the examples used to set up the extraction.

I count correct, missing, and incorrect fields, then measure how much work it takes a person to fix them. The test set includes different layouts, incomplete documents, and the exceptions your team sees most often.

This shows whether the workflow fits everyday use and lets you compare the time saved with the cost of building and running it.

Discuss a document processing pilot

Connect to your existing systems

I tailor the output to the application where work continues, such as your ERP, CRM, order management system or custom software.

I check the available APIs and import formats, data mapping and the acknowledgments the receiving system can provide.

I design for duplicate records and corrections and check how to tell whether a failed transfer is safe to repeat.

What affects the time and cost?

Two PDFs with the same number of pages can take very different amounts of work to process. These are the factors I use to scope a first release.

  • Document variety. How much do structure, language and content differ?
  • Legibility. Are there digital PDFs, scans or photographed documents?
  • Fields and tables. Does the workflow involve extracting individual fields or multi-page tables with many line items?
  • Business rules. Which checks, exceptions and approvals are required?
  • System integration. How does the software send the data and confirm that it arrived?
  • Day-to-day operations. How many documents arrive, how quickly must they be processed and who runs the workflow?

I use this information to propose an initial scope, considering development and running costs together.

Who will build your workflow?

I’m Marcus Biel. I’ve been building software with Java since 2001.

I’ll work through sample documents with your staff and design and build the workflow. I can work alongside your developers or bring in specialists where needed.

That includes connecting the receiving system and making sure someone knows what to do if a document needs attention.

More about working with me
Marcus Biel
Marcus BielSoftware developer and architect

Frequently asked questions about AI document processing

Can you handle scans and different layouts?

I test that with your own examples. Scans need text recognition, and accuracy can vary with image quality, tables, and changes in layout. I measure those differences during the pilot.

What happens when a value is extracted incorrectly?

I tailor validation and human approval to the impact of an error. The review screen lets staff compare the extracted data with the original and correct it. Test results show which checks need to stay in place.

Where are our documents processed?

Before I start building, I check who is allowed to access your documents, where they can be processed and stored, and which services you’re comfortable using.

What information do you need for a first conversation?

Tell me which documents you handle, roughly how many arrive each month, the fields you need and the target system. It also helps to describe today’s workflow and where people spend the most time.

Which documents are taking up your team’s time?

Tell me what comes in, which details someone has to copy, and where they need to end up.

We can use that to pick a small batch of documents for a first trial.

Other project inquiries: Automate a broader workflow · Connect two systems.

Prefer to talk first? Call +49 171 1151013 or message me on LinkedIn.

Discuss your document workflow