Security Transaction Note

The train the automated extraction Konfuzio was set up with the following labels:

Wertpapierabrechnung Compliance Check AI

Use Case: Why train a Document AI?

To check for illegal trading activities financial institutes, hereafter employers, are required to check trades of their employees. The process to check the trades via security transaction notices is quite tedious as the information is sent as a printed document or as a PDF to the employer.

Security Transaction Note - Process of creation

On a high level, the chart illustrates reasons why a security note is created.

graph LR Trader --> Account subgraph Bank or Broker Account --> Buy Buy --> Holdings Holdings --> Sell Holdings --> Dividends Holdings --> Splits[Splits and other Activities] Buy --> Note Sell --> Note Dividends --> Note Splits --> Note end

Submission process of Security Transaction Note

The first step in digitalization is to collect those notices, preferably merge several input channels to one, e. g. one email inbox. Paper-based documents can be scanned individually or as stacked scans. Stacked scans have the disadvantage to require a split up afterward.

graph LR subgraph Bank or Broker Note --> Print end Note --> E-Mail[E-Mail plus Attachment] E-Mail --> Inbox Print --> Scanner Scanner --> E-Mail

Security Transaction Note structured information output

Konfuzio can connect via Mail forwarding, SMTP, or API to the selected input channel to create a DigitalNote.

Definition of a DigitalNote

By DigitalNote we refer to one achievable PDF containing one Note and one machine-readable file, which could be CSV, XML or JSON. One machine-readable file can refer to multiple DigitalNotes. The PDF will be named uniquely, e. g. by combining the day of the note, the BLZ or BIC, and the bank account number, so it is easy to find.

graph TB Inbox[Inbox or Scanner] --> Konfuzio Konfuzio --> |one PDF one Note| DigitalNote Konfuzio --> |one PDF multiple Notes| Split Split --> |Rename per File| DigitalNote DigitalNote -->|optional human verification| RPA RPA --> |CSV / JSON import| System[Compliance System] RPA --> |PDF/A| Archive[File Archive] System --> User User --> |per case basis| Archive

How does an example DigitalNote created by Konfuzio can look like?

An example file might look like 20210605_70150000_123456789.pdf. The machine-readable format will include all attributes of the note that are required for further processing and compliance checks. The relevant attributes are not limited to the following list:

  • Broker / Bank like 1822direkt, Baader Bank, comdirect, Commerzbank, Consorsbank, Degiro (DE) , Deutsche Bank, DKB, ebase, Erste Bank (AT), flatex, Gratisbroker, ING (DE), onvista, Oskar, Peaks, Portfolio Perfor, Postbank, Scalable Capital, Smartbroker, Trade Republic, Union Investment

  • Bank account number

  • ISIN or WKN

  • Date of Note / Date of noted action

  • Type of Note: Buy and Sell notes as active actions and dividend and stock split notes as passive notes.

  • Limit to distinguish Stop Loss vs. Stop Win or Trailing Stop Los

  • Quality Check: This will be true if the price * share = amount - further quality checks can be added to verify the quality of the automatic conversion process

Accuracy after retraining

Our research shows that you should label 10 Status: Training documents as they provide the best first model.

img.png