Client is an American multinational financial services firm known for its credit card, charge card & traveller’s cheque business. It’s credit card business accounted for 30% of US credit card transactions (in dollar terms) in the US.
Corporate Credit business team is responsible for the growth of corporate portfolio globally. Sales team across geographies identifies companies which can be the target customer for corporate card business. The first step of data collection involves using B2B databases like Dun & Bradstreet along with manually scanning open & paid third party websites. The manual process was handled by various team members and data thus collected in spreadsheets is consolidated, cleaned and fed into the database for action.
This process resulted in the loss of time and hence lost opportunity from the sales perspective. It was therefore desired to automate data collection process so that corporate acquisition cycle could be sped up. It was further desired to identify prospects that were more likely to respond to the offerings and products they would buy.
Switzerland was chosen to be the target geography by client’s corporate sales team for an initial run. The prospective solution involved four components:
- Data collection layer comprising web crawlers that would gather corporate data through open & paid websites specific to geographies. These programs had to be customized for each website to some extent.
- ETL layer to process scrapped data for validity & consistency before storing in relational databases like SQL or MySQL.
- Algorithms to identify most likely to be converted prospect along with suitable product offering for it. Creation of such algorithms was possible due to existing customer data being provided by the client.
- Web-based visual interface for exposing results to the sales team for action.
Our team had a mix of technology and data science folks. We worked with the client to identify appropriate sources of data collection, open & paid both along with data fields to be captured. Once decided our tech team created automated web crawlers to collect data, clean it and push in into MySQL storage. There were 50+ data points collected for every corporate.
Technology team also created a web-based platform for the corporate sales team to view prospect data and segment it using different fields including propensity score for conversion.
For propensity score, data science team using likelihood modelling identified best prospects along with potential products to be sold to them using historical client acquisition data for the Switzerland market provided by the client. These models were built offline using R since the dataset was small. Records were scored offline and then updated in the database.
We were able to make the complete platform go live for the Swiss market in three months of time. Microsoft Azure platform was used for storage and hosting web platform. Later, we implemented this model for Middle East market.
- Initial response to the platform was encouraging. We were able to reduce manual workloads of data collection by 50% in the first month itself and by 100% in three months.
- Client’s corporate sales team was enabled to slice & dice the data and segment by chosen parameters.
- Propensity algorithms yielded 75% accuracy on validation datasets. We are still awaiting the results in the live scenario as B2B sales cycles are quite lengthy.