Hello to my industry friends out there.

DFA has always worked well with partners to sell our products, but the diversity of what we have created goes beyond some of these niche relationships. Consequently, I think my most regular contacts may be unaware of what we're selling. This is a short list (newest at top) and over the next few weeks I'm going to provide some detailed case studies.

AI Monitor: An LLM comparing sampled frontline communications with regulatory, ethical, and business rules to verify compliance. Launched to provide standardized metrics and audit reports for AI chatbots, but now being deployed for monitoring human and mixed staffing environments as well. Honestly, it's really cool. :) See www.deepfutureanalytics.ai.
APR Benchmarking and Credit Trends: For all loan types we don't just provide industry pricing data. We estimate the pricing function with far more detailed segmentation than you'll find elsewhere. And, we match this with trending of adverse selection that shows how much you have to adjust bureau scores to reflect the real risk of the applicant pools. Client testimonials of strategic saves put ROI around 20,000% per year. Not bad, eh?
Yield Forecasting and Portfolio Valuations: We do a steady business in creating account-level, economic scenario-based cash flow models that combine vintage stress testing concepts with origination and behavior scores. Great for campaign performance tracking, valuing loan pools for transactions, and identifying outlier loans. Incorporating adverse selection analysis, this has helped avoid some weak pool purchases and justified some strong pool pricing.
Pricing Optimization: It should be no surprise that if we can do 2 & 3, we can run an algorithm to optimize pricing. Studying our shared pool of 300+ clients, we have found that standard economics textbooks don't quite understand how price-demand works. It's that messy behavioral economics stuff. We have developed excellent pricing sensitivity functions that mesh with yield forecasting and constrained optimization to help you target your pricing strategy.
Machine Learning Scores without Overfitting: Ya, really. We started by experimenting with combining ML scoring with cash flow estimation so that ML scores become yield models. To my own surprise we discovered that overfitting disappeared, because of the separation of performance drivers. And, the same mechanism solves the pandemic data scoring problem. We use these methods for our models and run training-consulting projects to share what we've learned.

Before the year is out, we're going to have additional major product announcements, but you don't need to wait. :) If you just want to chat about analytics, lending, the economy, AI, ML, I'm a scientist, not a hard-sell guy. I'm happy to take a call and see if we can help.

Joseph Breeden
Posted on LinkedIn