Coding by Machine Learning

What does machine learning in Dooap mean? How to utilize Dooap's zero-configuration automation to predict Non-PO coding patterns?

Dooap Machine Learning

 

Machine learning is a new way of automating invoice handling based on the handling history.

1. The model self-learns your process by comparing the relevant invoice data* against the applied coding, to find the common factors that drive your coding selections.
   
2. When a new invoice arrives, invoice data is evaluated against the learning data. The three best coding predictions are returned to the user.
3. Adjustments, when needed, are fed back into the system, constantly improving the model for better accuracy. Re-training happens overnight.

 

*Learning model is based on a combination of virtually any digitized invoice data, such as company, vendor name and id, currency, amount, contract number, day of month, invoice type, payment method, or even your own customized header data and other xml fields.

Depending on your process, the system automatically recognizes the data relevant to your coding selections - usually at least Vendor - using a complex machine learning algorithm. For the incoming invoices, the most relevant data is compared against the past coding selections, to produce accurate coding predictions with zero configuration required up front.

Your data is safe, as it never leaves Dooap's Azure server.

 

 

Coding Pattern Detection

ML predictions are based on full 'coding patterns'. A coding pattern means any exact combination of dimensions across an exact number of coding lines. Order of the coding lines does not matter, but the selected dimensions and number of lines does. For multi-line coding, the amount is allocated by hand.

Tax Predictions (optional)

Tax group and Item tax group predictions may be enabled. This will automatically apply taxes for each coded or predicted line.

Predictions are based on invoice base data and selected dimensions for each line separately.

Selected tax groups are treated separately from coding patterns, and therefore do not affect the coding predictions.

 

Using Dooap Machine Learning

Here you will find more detailed instructions on how to utilize Dooap machine learning.

Coding Pattern Selection During Handling

 

• Once enabled, all users with access to invoice coding have access to the coding predictions.
• During invoice handling, Dooap returns 3 best predictions to the user's coding view.
• Users can quickly see details of the prediction by hovering the cursor over the buttons.
• Pattern is applied on the invoice by clicking on the icon.
  
• Active prediction is highlighted with a black icon.
  
• The icon also shows relative accuracy of the prediction.
  
  Note, the percentage is calculated across all possible permutations and should not be interpreted strictly as a measure of accuracy or quality. A prediction with 20% is likely to give close to correct results, if the alternatives are 1-2%.
  

Coding Pattern Selection During Import

• During invoice import, Dooap can apply the best coding prediction for invoices, where minimum percentage (default 50%) is exceeded.
• In ideal situations, together with Dooap's ability to auto-send invoices to workflow, this allows touchless NPO pre-processing.
• Predictions cannot override vendor default coding during invoice import.
  
  

Additional Notes

• ML learning models are company specific. Each company has their own set of learning data and coding logic. Invoices assigned to one company are only showing predictions from that same company.
• Configurations are available for:
  
  • Limiting learning data and predicion visibility per company.
  • Selecting up to 5 customized XML data fields (such as order reference, delivery address), in addtion to Dooap's standard data inputs.
• ML Coding is not supported during dimension-based approval.

 

Activation

Please contact Dooap support, to activate the ML functions.

Once activated, visibility of the coding predictions can be administrated per company.