Is AI taking all of our jobs in underwriting? No.

Last week, we delved into how AI is impacting financial services as a whole. We actually asked ChatGPT how this is happening, and it generated a comprehensive assessment– and fairly well written too! At a glance, it was incredibly scary from a human perspective.?

It seems like AI can and will take away all our (current) jobs tomorrow.

But don’t despair just yet. Here’s our ChatGPT-assisted analysis on one specific aspect of financial services Credit Scoring.

Step 1: We asked ChatGPT: "How will Al upend Credit Scoring and Underwriting?" – the secret sauce at OneBlinc.

“AI-powered credit scoring models leverage a diverse range of data points to assess creditworthiness and make more accurate lending decisions. By analyzing historical data, including credit reports, payment histories, and alternative data sources, these models can evaluate the risk associated with lending to individuals or businesses, thereby streamlining the loan approval process and reducing human biases.”

Sounds simple, right? Not really.?

Step 2: We asked: "What are some concrete steps to apply AI to credit scoring processes at our loans company?"

Applying AI to credit scoring can enhance the speed, accuracy, and fairness of lending decisions. Here's a step-by-step process:

1. Define Your Goals: Clearly articulate what you want AI to achieve. This might include increasing the speed of credit decisioning, lowering default rates, reducing bias, or better complying with regulations.
2. Data Collection: Collect data that is relevant to your credit scoring process. This could include traditional credit data (like credit scores, income, and employment history) and alternative data (like utility payments, rental payments, or social media data). Remember that any data you use must comply with regulations like the Fair Credit Reporting Act (FCRA) and the Equal Credit Opportunity Act (ECOA) in the U.S.
3. Data Cleaning and Preparation: AI models are only as good as the data they're trained on. Remove errors, duplicates, and inconsistencies from your data. This step also involves data transformation and normalization to make the data suitable for machine learning algorithms.
4. Feature Selection: Not all data points are equally important. Use statistical techniques to identify the most predictive features. This might include debt-to-income ratios, payment histories, or the number of open credit lines.
5. Model Building: Use machine learning techniques to build your credit scoring model. You might start with traditional techniques like logistic regression and move towards more advanced techniques like random forest, gradient boosting, or neural networks.
6. Model Training and Validation: Train your model on a subset of your data and then validate it on a separate set. This process helps ensure that your model can generalize to new data.
7. Model Evaluation: Evaluate your model using appropriate metrics. For credit scoring, this might include the area under the Receiver Operating Characteristic curve (AUC-ROC), accuracy, precision, recall, and F1 score.
8. Model Interpretability: It's crucial to be able to explain your model's decisions, both to comply with regulations and to maintain trust with your customers. Techniques like SHAP, LIME, or partial dependence plots can help interpret complex models.
9. Implementation: Integrate your model into your credit decisioning process. This might involve creating a real-time scoring API, a batch scoring process, or a user interface for loan officers.
10. Monitoring and Updating: Continuously monitor your model's performance and update it as needed. Changes in economic conditions, customer behavior, or your business can all necessitate model updates.

Note: out of the 10 steps above, all 10 require heavy human input and adjustment. Humans are FAR from being displaced in the application of AI to credit scoring and underwriting. Furthermore, many of the skillsets above are not native to a traditional lending company, so most lenders will have to hire more people for these.?

Nonetheless, there are other jobs in the underwriting value chain that will be affected by the increases in productivity and streamlining of processes. Reskilling these workers today to become interpreters of data sets, sanitize data, and overall monitoring of outputs could be a great way to leverage old-world wisdom into better applications of AI.?

If you are a lender, how will you start to shift your workforce to leverage AI today??