The Defensibility Gap in Automated Valuations

When a borrower challenges a low AVM estimate, what can the lender actually document? Most AVMs produce a number without a rationale.

A borrower under contract at $498,000 receives notification that the automated valuation came back at $461,000 -- a 7.4% gap that may prevent the transaction from closing without a price renegotiation or additional down payment. The borrower asks why. What does the lender tell them?

In too many cases, the answer is: the model determined this value based on comparable sales data. That answer is legally and practically useless. It does not identify which comparables were used. It does not explain which property characteristics were weighted. It does not show the condition adjustment methodology. It is a black box assertion dressed in vague language, and it does not survive a motivated borrower's follow-up questions.

What "Defensible" Means in Practice

A defensible valuation is one where a reviewer can reconstruct the analyst's reasoning from the documented evidence. In a traditional appraisal, this means the 1004 form: the subject property description, the three to five comparables listed with their sale prices and attributes, the dollar adjustments for each characteristic difference, and the reconciled value. A desk reviewer, examiner, or borrower's representative can evaluate each step of the analysis, agree or disagree with specific adjustments, and identify whether the reconciled value is reasonable given the evidence presented.

This standard exists not because appraisers enjoy paperwork but because the lending industry discovered through regulatory experience -- particularly after the savings and loan crisis -- that undocumented valuation judgment produces exactly the kind of systematic error that regulatory oversight is designed to prevent. Documentation is the mechanism through which valuation quality is reviewable, not just asserted.

The Documentation Gap in Most AVMs

Most automated valuation tools in the market were built for speed and scale. The design goal was to produce a point estimate with reasonable aggregate accuracy across large transaction volumes -- not to produce a documentable comp analysis on individual properties. The proprietary model weights are not disclosed because they are the vendor's intellectual property. The specific comps used for any individual estimate may not be retrievable in a reviewable format. The output is a number, perhaps with a confidence interval, and a list of "relevant recent sales" that may or may not have been the actual inputs to the model.

This design works well for the consumer use case and for portfolio monitoring at institutional scale, where aggregate accuracy is what matters and the review process is statistical rather than transactional. It does not work for loan origination, where every individual valuation may be contested and the lender needs to be able to defend each one in an adverse scenario.

The Regulatory Documentation Standard

The Interagency Appraisal and Evaluation Guidelines require that evaluations include a description of the methodology used, the data sources consulted, and the property's estimated market value consistent with the market value definition. For automated valuations used as evaluations, this means the output document needs to contain enough information that a subsequent reviewer can assess whether the methodology was appropriate and the data sources were reasonable.

Vendors who describe their AVM output as "supporting desk review" without making that output reviewable by the desk reviewer are describing a workflow that does not satisfy the spirit of the evaluation requirement. The desk reviewer cannot evaluate what they cannot see.

Borrower Reconsideration Requests

Reconsideration of value (ROV) requests are common in purchase money transactions and becoming more common in refinance transactions as borrowers are better informed about their rights. An ROV request asks the lender to consider whether the original valuation appropriately reflected the property's market value, typically by identifying comps the borrower believes are more relevant than those used in the original analysis.

An undocumented automated valuation cannot be meaningfully compared against a borrower's proposed comps. The lender does not know which comps the AVM used. The borrower does not know which comps the AVM used. The ROV process is supposed to be a comp-by-comp comparison of competing analyses; without the original analysis being visible, the comparison cannot happen. The lender is left in the position of either accepting the borrower's comps blindly or asserting the original AVM result without documentation -- neither of which serves the lender's regulatory position.

The Defensibility Standard in Action

A valuation that is defensible under adverse conditions -- borrower challenge, regulatory examination, or investor audit -- is one where the supporting work is visible, the methodology is disclosed, and the reasoning chain can be followed from data to conclusion. This is not a higher standard than what appraisers are already held to; it is the same standard, applied to automated output.

Plotgleam's report structure was designed against this standard from the start. The comp selection table is in the output. The condition score is in the output. The block-level signals are in the output. When a borrower submits an ROV, the desk reviewer has an actual analysis to compare against the borrower's proposed comps, rather than an assertion to defend. That is what defensibility means in a loan file context.

The Examiner's Perspective

A regulatory examiner reviewing a lender's collateral files looks for evidence that the valuation methodology was appropriate, the data sources were reasonable, and the outcome was consistent with the methodology. An AVM output that passes this review is one where the examiner can trace the estimate back to specific comparable data, understand how the data was used, and assess whether the methodology was consistent with safe and sound banking practices.

An AVM output that fails this review is one where the examiner sees a number with no visible supporting work. Even if the number is accurate, the absence of documentation is a finding. Examiners do not grade on curve; they assess whether the documentation in the file meets the standard. Lenders who rely on undocumented automated estimates take on documentation risk that is separate from and independent of the accuracy of the underlying model.

Building defensibility into every AVM-supported transaction is not primarily a regulatory compliance exercise -- though that is a genuine benefit. It is a credit quality discipline: requiring that every valuation be supportable forces the team to engage with the evidence rather than accepting a black-box number. The loan officers and desk reviewers who develop comfort with documented automated estimates are also developing the market judgment that helps them identify when a number looks wrong and deserves a second look. That judgment is not scalable from a black-box output alone.