This site uses different types of cookies, including analytics and functional cookies (its own and from other sites). To change your cookie settings or find out more, click here. If you continue browsing our website, you accept these cookies.
on 08-31-201810:18 AM - edited on 06-19-201911:24 AM by ichand
2018 Excellence Awards Entry: Cheque-to-Cash Factoring with Predictive Analytics
Name: Altan Atabarut
Title: Analytics Consultant
Company: Ulusal Factoring, TR
Approved Company Logo:Overview of Use Case:
Factoring is a financial transaction and a type of debtor finance in which a business sells its accounts receivable (in our case cheques or notes) to a third party (called a factor) at a discount. A business will sometimes factor its receivable assets to meet its present and immediate cash needs so these factor firms become the heart of any economy that supply immediate cash flow.
For Ulusal, almost 2000 checks and notes are received daily for cheque to cash businesses. There is a need for near real-time risk assessment for credit decisioning (Accept a cheque or Reject). The factor will decide to buy or not to buy the customer’s receivable by looking at the default risk levels of each and every counterparty, the cheque issuer, the customer and any other party like endorsers in between. Will at least one of the counterparties be able to pay back the debt on time at the due date?
Normally this is a cumbersome manual task handled by underwriters that may take 10 minutes to hours. And there are almost 1700 data points for a single transaction that is literally impossible for a person to judge.
We have built a scoring model for the accept/reject decisions. This model runs every month and learns customers past payment behaviour and updates risk predictions in a batch automated fashion. The scoring model is run for every transaction to come up with a probability of default.
For those cheques agreed to be accepted, dynamic pricing is essential. Factor has to decide on the discount rate for the receivable based on the risk levels of both the issuer, endorsers as well as the customer itself. We have also built an estimation model for the discount rates given the counterparty risk levels.
Describe the business challenge or problem you needed to solve:
There are at least two counterparties to a single transaction: the customer and the cheque issuer. This can increase up to 12 with addition of endorsers. For each counterparty, there are multiple internal and external sources of data for risk assessments that needs to be cleansed and blended. The sources of data are;
Customer relations data - RFM (recency, frequency, monetary value)
Previous risk-taking information related to issuers and endorsers that can be gathered from credit bureaus for the past 3 years.
Historical cheque usage and late payment information from central bank database for the past 1 year.
Also some fraud, collection and intelligence data exists in multiple onpremise and cloud databases.
Credit Bureau and historic cheque usage data makes around 30-40 million rows of data at a minimum. This entire data set needs to be blended into a single modelling table that will be fed into predictive models.
Describe your working solution:
We used Alteryx Designer for ETL and Predictive modelling. Today, an end-to-end ETL process creates a modeling table. Before Alteryx, this project took a month’s time. Today, it takes approximately three to four hours. All data is filtered, joined and cleansed through a series of workflows using in-database tools. In-db tools are connecting to Oracle databases running millions of rows of data pretty quickly. The modelling table is then fed into Random Forest algorithms to predict 30 days past due for each and every cheque or note transaction received for the past 3 years. This produces a modelling file which is then used for transaction scoring purposes.
Describe the benefits you have achieved:
Before Alteryx, the manual process of risk assessment and accept/reject decisioning would take anywhere from 10 minutes to hours. It can now be done with automated scoring without human intervention. Similarly, pricing levels for each and every transaction are now based on the level of counterparty risks and can be decided objectively every time. This will enable a lot more transaction to be handled with the help of digital transformation and mobile financial services integrations and more through pricing decisions consistent with every decision taken.