on 07-02-24 08:16 AM - edited on 12-04-24 01:13 PM by Michael_S
The automation sought to improve the speed and quality of data capture from bank statements, as well as reduce human error.
Discovery
Our customer reported the problem and provided an overview of the goals.
Design
I sought to understand the process step by step. This included listing all the compliance rules that the process required, documenting the "AS IS" state of the process and the TO BE desired state. I then requested approval from the client, and acquired the necessary system access.
Build
The automation was built using :
Blue Prism Enterprise to extract data from the system, to capture PDF files from the system, to fill in template files and for the messaging / alerts steps of the process.
We used Decipher IDP to capture and interpret data from bank statements.
Quality Assurance
The biggest factor here was machine learning. Training Decipher IDP using templates until it achieved a 100% confidence rate took time, but was worth it for the sustainability of the process.
Huge. We successfully improved both the speed and quality of data capture from bank statements!
------------------------------
Guilherme Morandini
Consultant RPA
Avanade Brazil
São José dos Campos
+5512981637307
------------------------------
@Guilherme Morandini soooooo gooooood!
What did you train Decipher IDP with? Did you use past examples of statements?
Hi Guilherme,
I am currently working on a similar usecase. I would like to know more about your process.
Can you please explain more about the how many different bank statement were trained? Did you used single DFD or multiple DFD ?
Hi Athiban Mahamathi Mathialagan ! How are you ?
Three different types of DFD were trained. One type of DFD for each type of bank statement template.
Hi Michael ! How are you ?
I trained Decipher IDP using multiple examples of the three types of bank statements.
I used da DFD and a Document Type for each type of bank statement.
Hi, Guilherme,
impressive. How long did it take to reach 100% in Decipher and is it really 100% or close to 100%?
Regards
Zdenek
@GuilhermeMorandini given we are a couple of months down the track since your original post, how many bank statements are you currently processing through Decipher IDP ?
Hi @zdenek.kabatek ! How are you ?
To reach 100% accuracy, I believe it is more linked to the number of document examples that the developer will have to train the model in Decipher. In my case it took 2 to 4 weeks to get close to 100% accuracy because I had to capture more sample documents to train the model.
Regards,
Hi @timappleyard ! How are you ?
The process has a volume of 5 daily bank statements