The journey from data to decision

By Boris Huard, UK&I Managing Director of Decision Analytics, Experian

Successful businesses within the financial services market are increasingly based on the automation of complex data-driven decisions.

It’s useful for the business to understand the process in order to ensure the right trouble shooting should the steps go awry, and also to ensure that the business benefits from a continuously learning and optimising mind-set to improve on the procedures as customers and the market evolve.

Here’s how the typical process unfolds.

 Stay Updated To Save Money & Time. Join Our Free Newsletter 
. Indepth Analysis & Opinion       . Interviews      . Exclusive Reports  
. Free Digital Magazines      News & updates      . Event Invitations 
                     
& Much More Delivered To Your Inbox For Free.
Submit
We Will Not Spam, Rent, or Sell Your Information.
All emails include an unsubscribe link. You may opt-out at any time. See our privacy policy.

 

Step 1: Data extraction: Data is extracted from operational systems and put into an off-line environment. This is typically done in batches – meaning all the data is gathered at one time – and contains a historical view of information, such as credit card balances each month for the last 12 months. When ad-hoc analytics are done, this may be a one-off, but many companies have established large IT ecosystems that pull data into operational data stores, business intelligence, data warehouses and now data-lakes.

Step 2: Analytics creation (discovery, data-preparation, model building and testing): Once data is available to data scientists, a variety of tools can be used to discover the contents and to prepare the data, such as translating credit card balance information into a trend analysis that shows increasing, flat or decreasing balances. This requires preparing (or wrangling) data, feature preparation (such as transposing values into a trend) and data quality / cleansing (such as getting rid of bad records or learning to ignore badly entered data). Once the data is available in the required format, various statistical, manual or machine learned approaches can be applied to create a model which takes the input (data and derived features) and then predicts the outcome.

Step 3: Model import After a model is done, it needs to be brought into the decision environment. In today’s credit operations, a model is often re-keyed as a security step to ensure that the correct model is used. Various governance and sign-off steps are common, including validating model performance and model simulations. PMML – a mark-up language that defines the steps in a model – is a typical way to import models.

Step 4: Strategy design and governance Once the relevant model is imported, the actual business strategy needs to be designed. A prediction model may say that there is 3.72 percent risk that a customer will default, but the lending decision will also depend on the type of the loan whether it is secured or not, what the customer’s potential lifetime value is and many other factors. These elements could be encoded directly in the model but, doing so, would quickly become very cumbersome and time consuming to manage and change. Instead, decision trees and simulation tools are often used to define and study the impact of different decision factors and strategies.

Step 5: Decision execution To execute a decision – i.e., making the actual decision at a point in a business process, the operational processes and related systems need to invoke the decision software. This is either done as a call out, i.e., one system calls another to receive a decision, or by embedding the decision software directly in the processing system.

Step 6: Data enrichment Most times the data available in the process is not enough and additional data sources are needed to inform the decisioning process. This can include: checking a balance, confirming if a person is already a customer, getting bureau data and a credit score. To do this, the decisioning system needs to integrate with a variety of data-sources and these data-sources need to match the ones used to build the model.

Step 7: Monitoring, Learning and Improvement At the end of the cycle, businesses want to monitor and understand what is going on in their platform. Today, most of this monitoring is performed by collecting the necessary data in warehouses, data-lakes and other systems.

Analytics is a highly dynamic area, with lots of solutions changing and appearing quickly. Most companies are unlikely to feel comfortable with being tied in to a single analytics solution.

The governance aspect of decisions that has been prevalent in credit risk is going to become more attractive and important in other decisions in light of regulatory changes and the increased need to treat customers fairly.

 Stay Updated To Save Money & Time. Join Our Free Newsletter 
. Indepth Analysis & Opinion       . Interviews      . Exclusive Reports  
. Free Digital Magazines      News & updates      . Event Invitations 
                     
& Much More Delivered To Your Inbox For Free.
Submit
We Will Not Spam, Rent, or Sell Your Information.
All emails include an unsubscribe link. You may opt-out at any time. See our privacy policy.

 
Close
Stay Updated To Save Money & Time. Join Our Free Newsletter. 
. Indepth Analysis & Opinion       Interviews          . Exclusive Reports 
. Free Digital Magazines        . News & updates        . Event Invitations
& Much More Delivered To Your Inbox For Free. 
Submit
We Will Not Spam, Rent, or Sell Your Information.
All emails include an unsubscribe link. You may opt-out at any time. See our privacy policy.
 
Close