E3

Resources

Mortgage intelligence and AI

3 pieces.

All writing
FIGURE 1What current models solve, and where they stopThe boundary is not capability. It is reproducibility.WORKS WELL, READINGNOT SOLVED, REASONINGReading documentspoor scans, unfamiliar layouts, formats not trained forClassification and segmentationidentifying documents, splitting packagesLocating valuesfinding a field and reporting where it was foundSummarising and answeringuseful answers about a fileWhether a value is righthigh reading confidence, wrong figure for the fileReproducibilitythe model has a version; the conclusion still may not re-deriveKnowing what is missingabsence is a rules question, not a reading questionSurviving a rule changea finding must name the standard applied at the timeReading confidence and correctness are different properties. A platform reporting one number for both is describing something it does not have.Reading is a commodity capability with a real error rate. The mortgage reasoning above it is neither.

How to Evaluate a Mortgage AI Claim

Six questions that separate a checkable AI claim from an unverifiable one, with the failure modes each question is designed to surface.

7 min readFIGURE 1A finding that can be examinedSix things a third party needs in order to reconstruct a decision without trusting you.WHAT TRAVELS WITH A FINDING1The values comparedthe actual figures, not a summary of them2Where each came fromdocument, page, and position on the page3Confidence on eacha clean read distinguished from a marginal one4How a conflict resolvedwhich source won, and on what basis5Which rule appliednamed and readable, not an opaque score6Which version of that rulethe one part usually missingWithout the sixth, every historical review silently re-anchors to today's rule set.Most systems retain the first two. The last is what makes a review defensible after the rule changes.

Explainability Is Not a Feature of AI in Mortgage. It Is the Condition of Using It.

What a reviewer must be able to reconstruct about a finding years later, and why an adverse-action notice and a QC finding are different obligations.

6 min readFIGURE 1Five waves, one boundaryThe boundary has moved from arithmetic to reading. It has not been crossed.WAVEWHAT IT AUTOMATEDWHAT IT DID NOT0Standardisationthe uniform formsnothing: it enabled everything after1Mainframe servicingpayments, escrow, arithmeticorigination stayed on paper2Automated underwritingeligibility against guidelineswhether the values were right3LOS and imagingmovement and storagereading, and checking4Compliance as softwarethat events occurred in orderwhether the substance was correct5Machine readingreadingdecidingthe boundary, unmoved in fifty yearsEach wave automated a computable layer and stopped at judgement. The waves that lasted made something reproducible.Desktop Underwriter 1995; ATR/QM effective 10 January 2014; TRID effective 3 October 2015.

Fifty Years of Mortgage Technology, and What Each Wave Left for the Reviewer

Five waves of mortgage technology since the 1970s, what each automated, and which review tasks each one left behind.

6 min read