Solutions
Document Classification
Incoming mail and documents must be manually reviewed, sorted, and forwarded. For inclusion companies and scan service providers, this is core business — but also the bottleneck.
The Solution
feld.ai automatically classifies incoming documents according to your categories. Invoice, contract, reminder, notice — whatever arrives.
How It Works
1. Import documents — Feed incoming files via email, upload or scan folder. feld.ai processes PDF, images and Office formats.
2. Classify automatically — AI assigns each document to the correct category — invoice, contract, notice, reminder or your own custom classes.
3. Split and orient — Multi-page scans are automatically split into individual documents and correctly oriented.
4. Trigger downstream processing — Depending on the class, the appropriate extraction or filing process is initiated. Classification drives the entire downstream pipeline.
Extreme Class Imbalance
Works even with 99% invoices, 1% special cases
Nested Documents
Emails with attachments, ZIP archives, multi-page scans
Pipeline Chaining
First classify, then extract differently depending on class
Document Splitting
Split and correctly orient documents
Related Solutions
AI-Powered Mailroom — Automatically classify, extract, and file incoming mail.
Invoice Processing — From scan to posting suggestion.
Industry Solution: Tax Advisors & Law Firms — Automate client mail and tax notices.
Frequently Asked Questions
How many document types can the system distinguish?
There is no fixed upper limit. In practice, our models distinguish between 5 and over 80 classes — depending on the use case. New classes can be trained with just a few example documents.
How much effort is the initial training?
For most classes, 20 to 50 example documents per category are sufficient. For very specific document types, even fewer are often enough. Training typically takes just a few hours.
Does classification support multiple languages?
Yes. Our models work language-independently based on visual and textual features. German, English, French and other European languages are supported without separate training.
How accurate is the classification?
In production environments we typically achieve over 95% accuracy. With extreme class imbalance (e.g. 99% invoices, 1% special cases), feld.ai employs specialised techniques to reliably identify rare classes.
How does integration with existing systems work?
feld.ai provides a REST API for all classification steps. Webhooks notify your system when processing is complete. Integration with existing DMS, ERP or workflow systems is possible via standard interfaces.