Conservar

Inspiration

Most banks spend thousands of dollars in an effort to attract new customers into their business. But even after a successful attainment, most banks don't really have a means of 'preventing' a customer from leaving.

What it does

Conservar provides banks with the opportunity of detecting customers who might potentially be on the verge of leaving them hence allowing them to do something to entice them into staying. The detection is based off of two categories: Transactional History Conservar processes each customer's data and provides an aggregate of their transactional history. Accounts with an extreme drop in activity or below a certain set threshold, are flagged as a potential defector. Characteristics Exploration Through machine learning, Conservar is able to relate a customer's current credit score to their location in order to provide banks with predections of their the customer is most likely to stay or not. Along with this, this tool can also be used by banks to target a certain customer audience.

How I built it

We used Google's Tensorflow along with sklearn to build the brains of the system in Python. A splash of Javascript, CSS and HTML made the dashboard possible. Thanks to Capital One's API we were able to get the data format required to feed into our classifier.

Challenges I ran into

Machine learning pretty much. This was our first time doing it and we really wanted to challenge ourselves and my oh my were we challenged..

Accomplishments that I'm proud of

Getting the machine learning portion working.

What I learned

Tensorflow, SKLearn and some data analysis methods

What's next for Conservar

Improve the machine learning algorithm by not only adding more attributes, but also collecting more data in general. The end goal is to have the platform even find different potential earning means for businesses and not just banks.

Built With

• css
• html
• javascript
• python
• sklearn
• tensorflow