You certainly want to make sure that the data you pull together will support your CECL analysis. But that data can also be used to analyze marketing and credit policy initiatives to improve business results. While Oscar Wilde once said 'Consistency is the last refuge of the unimaginative' it is very important to capture your data consistently to support all regular and one-off analyses and to ensure that all users understand and assess the data in the same way.
Customer Lifecycle
First and foremost you need to track customer performance over the account lifecycle, capturing data for each month. At a minimum you should collect balance, delinquency status, and whether the customer has declared bankruptcy or charged off. A nice to have would be payment data. From there it's important to identify the first time (lifecycle month) that negative actions have occurred. This includes when the account first hit 30, 60 and 90 days delinquent and when the account charged off. It's important that you know when these events first happened and not just that they happened. This is especially important for developing lifetime delinquency and loss curves.
If income earned (net interest and fees) you should capture this data as granularly as possible. If average balances for the month are available it is a good idea to include this data, too, for proper weighting of balances. For credit cards and open ended accounts you should also capture monthly usage broken down by cash advances, balance transfers and regular purchases. Ideally this would include both the number of transactions as well as the net amount.
Profile Data
There is a wide range of data that describes both the customer and the account. This is generally point-in-time data that may change (e.g., income) but is used in making decisions about the account. Data that is related such as product characteristics or customer demographics typically are stored separately and then joined together as needed.
The key, of course, is to have a field that is common between tables (assuming a traditional SQL database). Generally this is something that will always be unique to the account or the customer. The common field should not be anything that could be used to identify the individual (e.g., SSN). Instead a depersonalized number should be created that can only be associated with that customer.
Key Events
Another data set you may capture are all of the key events -- actions taken during an account's lifetime -- so that you can measure the impact of the action. This includes marketing actions such as cross-sell programs as well as credit programs like credit limit changes. Chances are credit actions apply mainly to open ended accounts like credit cards.
It's important to track test and control groups, ensuring that accounts are randomly assigned to each group. As more events are executed random assignment becomes more challenging.
If You Don't Measure Performance
It All Looks Random
Data Quality
Data quality addresses both the accuracy of all data as well as full coverage. Accuracy pretty much speaks for itself. You want to know that data is correct for each account and that it has been used consistently. Coverage relates to how many have (or do not have) a given piece of data. If data elements are missing for a large portion of the accounts the value of that data element is diminished.
