The meta-model analyzes and adapts the way our AI works, making it even smarter and more adaptable.
This article assumes knowledge about the Finmatics machine.
For an overview of how Finmatics AI works, see the following article:
How does Finmatics artificial intelligence learn?
Digitize documents, recognize document features, without templates.
As a Finmatics customer, you benefit from top readout rate and quality throughout Europe.
Accounting professionals must be able to rely on the readout values.
A new feature, the meta-model, further reduces the error rate for reading information from documents.
What does the meta-model do? The check model checks...
Simply put: The meta-model checks whether the read out values were correct for this business partner in the past and which model is the best model for the business partner. This check takes place per business partner and client, so that all sources of error (including OCR errors) can be taken into account.
Our analyses have shown that for certain fields the Deep Learning model works better and for other fields the Auto Machine Learning model works better. Therefore, the meta-model "learns" which model gives the best results for the document.
A small example: the invoice number is read and output as RE404 from Finmatic's Auto Machine Learning model. However, you correct the value to RE1234, exactly the value that the Deep Learning model output. Then the meta-model learns that in the future the Deep Learning model should be used for this business partner.
The next time, the value will be correct.
The meta-model learns where values are incorrectly predicted
In rare cases, both the Auto Machine Learning model and the Deep Learning model are wrong. In these cases, the meta-model learns that the result is not yet correct and offers no read out result.
The result is a reduction of the error rate!
How often must the business partner occur for the check model to work?
The meta-model works as soon as the business partner has been processed and confirmed 2x at the client.
How much does the test model reduce the error rate?
An evaluation of all Finmatics processing operations has shown that the error rate can be reduced by 300%. This means that a completely new level of quality can be achieved in processing.
Can I switch the meta-model on and off or set it?
In special cases, it may make sense to parameterize the meta-model per client. Together with your Finmatics Customer Success Manager, the meta-model can be adjusted specifically to your processing operations in terms of sensitivity.
What do I have to do to activate the meta-model?
The meta-model has been active for customers since June 2021.