dcrypt helped client increase its capture rate of potential credit candidates by 12%
The Business Challenge
A large financial services company wanted to improve its analytics and intelligence support. Challenges included:
- Working with large sets of unstructured addresses’ data to map/match different transactions and clients using only names and addresses.
- Using the client’s fuzzy matching framework for names and addresses, which could not provide a single view of customers across different data sources when no unique ID was provided.
Fractal solution dCrypt, helped improve overall matching accuracy by 9 percentage points. It solves problems with the same technology used to translate languages.
- dCrypt is a proprietary self-learning, Natural Language Processing (NLP) algorithm to predict a new product description to the best-suited attribute, in real-time.
- The new framework was created in two parts: first, to incorporate heuristic features for data cleaning, normalizing and standardizing; second, to use dCrypt’s superior fuzzy matching prowess to map standardization addresses and add new features.
This new solution helped to decrease false positives by 7% and 9% respectively, while improving the overall accuracy of matching by 9%.
By deploying dCrypt helped our client track nearly 120 million more candidates for credit rating agencies not captured by the existing framework.
- Fractal discovered that the original address data was very erratic, with no standard conventions followed to write addresses or to correct incorrect pin codes, though the first three digits of the pin code are always correct (state, city, town or district).
- One single method of fuzzy matching cannot capture all the elements of address variations.
- The final Fractal solution’s capability is defined by the calibration done for using different fuzzy matching scores
- Fractal’s new solution framework helped this client increase its capture rate of potential credit candidates by 12%, from 70% to 82%.
- Fractal was able to track nearly 90 million more candidates for credit rating agencies not been previously captured by their existing framework, plus another 30 million for a total of 120 million.
- Fractal incorporated a classifier model in the fuzzy matching framework to improve the output.
- This classier model helped Fractal in segregating addresses on the basis of characteristics not previously captured by the client’s fuzzy matching algorithm.
- This innovation helped increase the capture rate by another 4%, translating into almost 30 million more customers tracked above the 90 million already tracked.