FiveS Digital Logo
Automation for accident reports processing

About the Client


The client is a commercial web-based service that supplies vehicle history reports to individuals and businesses on used cars and light trucks for American and Canadian consumers.


Story Line

FiveS Digital works on document digitalization of accident reports of various cities through a portal, and situations have dramatically changed after post covid which affected the working procedure, so to provide a solution to the client we implemented various technologies for automation.


Challenges

  • Manual data entry processing required human resource which was a challenge due to conditions arising during Covid, and due to this TAT achievement went down.
  • Processing time taken for each image was 210 secs which was affecting the delivery and TAT.
  • The process had 100% audit allocation which almost costed 50-60% of the processing cost, so with higher volumes, 100% audit was a challenge, and random method affected the quality target which was 99.5%.


Automation Challenges

  • Multiple type of accident reports of various agencies with many pages and fields had to be identified through the system.
  • The entire data extraction process was a challenge due to unstructured pdf's along with implementation of process rules which were different for each agency.
  • At start of the project the processing time was reduced to 120 secs per image but still it was not optimized.


Solution

  • Robotic Process Automation(UiPath) technology was identified and implemented for automating the data entry processing along with AI and ML algorithms.
  • Identified the behavior, tenure, past performance and attendance-based logic, A quality sample size recommendation engine built on the defined logic through UiPath, Daily update of quality scores generates quality audit % user wise.
  • Basis the type of data an AI/ML Classifier was built through Python which supported in classification and data cleansing and alignment.
  • For attaining the best quality, data annotation and OCR was applied for machine learning and extracting data from pdf and converting to text.
  • Multi threading programming construct was used which reduced the overall execution time of per image processing.


Result

  • TAT achievement increased from 85% to 100%.
  • Audit % reduced from 100% to 16-30%
  • Handled 1500 records daily with consistent Quality achievement which was above the client’s benchmark 99.50%
  • Processing cost reduction by 50%.
  • Per image handling time reduced from 210 secs to 40 seconds per image.


Process Automation Flow


  • Data uploaded by client on portal
  • Data allocation user wise UiPath - BOT
  • Pdf downloading from user portal and movement in respective folders UiPath - BOT
  • AI engine trigger from PuTTY terminal emulator UiPath - BOT
  • After the trigger, AI engine will convert pdf to JPG AL/ML Algorithms – Python coding
  • Classification, cleansing and alignment through Al classifier and python coding AL/ML Algorithms – Python coding
  • Data annotation for machine learning AL/ML Algorithms – Python coding
  • Data extraction field wise done by Detectron2 AI models AL/ML Algorithms – Python coding
  • OCR implementation on extracted data for covert images to text AL/ML Algorithms -Python coding
  • Field wise text generated in csv on shared path AL/ML Algorithms -Python coding
  • CSV detection done by BOT batch wise UiPath - BOT
  • Automatic data processing done on client portal UiPath - BOT