The primary goal was to process a dataset of 50 images showing website experiment results and rank each test by its performance. Each image contained a test name alongside traffic and conversion numbers for both an original design and a new variation. The main challenge was extracting these specific numbers accurately from a flat picture. Specific objectives included using software to read the text, calculating the success rates for both groups, and measuring the exact uplift %. The final requirement was to export this structured data into a ranked web page, a spreadsheet, and an organized data file to support immediate business decisions.

The solution relies on a Python script utilizing EasyOCR for Optical Character Recognition to read raw text directly from the images. EasyOCR was selected because it works reliably without complex system installations. After reading the text, the script uses regular expressions to locate the exact numbers associated with visitor traffic and successful actions. The code specifically searches for row labels to ensure it groups the correct numbers together. Next, the Pandas library processes this extracted data to calculate conversion rates and the relative uplift % for each test. Finally, the script sorts the results from highest to lowest and generates an interactive HTML web page alongside CSV and JSON data files. Matplotlib and Seaborn libraries then create visual charts to highlight overall performance distributions and the top winning tests.

The automated pipeline successfully transformed 50 static images into a fully structured and ranked database without any manual data entry. The data reveals a highly successful testing program where 41 tests generated positive results. The top performing test achieved a 22.74% improvement, while the worst test only caused a minor 6.54% drop. The analysis also shows that the highest traffic pages resisted significant changes, hovering near a 0% improvement mark. Based on these findings, the business should immediately deploy the top 10 winning features to the live website to capture immediate revenue. The data team should also investigate user behavior on the highest traffic pages to understand why new ideas struggle to succeed there. This automated process can now evaluate all future test results instantly.