Sharing resource: Code for computer-aided customer needs detection

Hi peers,

I just wanted to share a piece of code I developed some of you might find useful:

It is a script for the open-source robotic process automation software “Kantu UI-Vision” (Selenium-based) that allows you to automatically scrap negative users reviews from an app in Google Play and then save it to a CSV file.

USE CASE: Say you want to develop a logistic worker-owned platform coop, with this script you can first extract a huge amount of negative reviews of incumbent or substitute products/services, manually label a sub-sample and detect which are the key flaws on current services in order to develop something with higher added value and focus your development energy on the features that matter the most at the end of the day.

1 Like

By the way, the script DOES NOT extract personal names nor customer private data, just the review’s text, score and date.


A+ business intelligence, definitely going to be using this


Might need some walking through to figure out how to use this. Looking to setup a food delivery coop as I can see a gap in quality of service offered by current service providers

1 Like

I’m assuming you are a coder or you have a coder in your team but essentially you have to:

image clone the repo

image go to and click “add RPA to browser”

image install the extension according to your browser

image you then have to install a set of modules called xmodules which are used for computer vision at Download XModules for enhanced features, Robotic Process Automation (RPA)

Once all installed, you go to the extension

image, change the option storage mode to filesystem and click on “Import JSON” selecting the file you download from the git clone command.

Once loaded, you go to the variable “Product” and change the name for the competitor you want to examine. Afterwards, click play macro and wait for the magic to happen. (You can then download CSV)


Once you got the CSV, there you proceed to manually labelling a sample of reviews. There multiples ways of doing so. Here is an example:


You will that from a larger sample of reviews, a select a representative sub-sample and label them. Essentially, I read the text, categorise what was the problem and it’s main theme.

When you label a subsample you can then project an estimate to the totality of users (of course this is just an estimate!!). Here is the way I like to illustrate it.


Further analysis can include UX flows where you illustrate in which particular part of the process your competitor or own’s ux flow “breaks”:


I hope it helps.

Thank you for the breakdown. I will revert on this