*Accept No Substitutes*
**By Jennifer Miller**
***We’re thrilled to see the mortgage industry move toward a more data-driven process. The work of the Mortgage Industry Standards Maintenance Organization (MISMO) has made it possible for lenders to work from a common language based on XML. Huge efficiency gains and easy compliance with GSE mandates are possible now, but there’s an iceberg-sized deal killer we’ve got to remove.
****Some lenders and technology providers have chosen to cut corners, using various error-prone techniques that try to extract the data from appraisal forms. (Some of them call this process “OCR” but “text extraction” is a more accurate description.) Here’s how it works: The company receives a signed report as a PDF from the appraiser. The software reads all of the text in the report and tries to guess how the text in the report should match up the MISMO XML fields. Most of the time, it will guess correctly, but even if its guesses are 99% correct, that leaves 2 or more errors per report – depending on the length of the report. That’s before we even consider lengthy addenda. How many computer-generated errors can you afford?
****Here’s the crux of the matter: appraisal forms are meant for human eyes, not computer reverse engineering. This text extraction approach defeats the entire purpose of data standardization by miring everyone involved in errors, lost productivity, and the massive financial losses resulting from forced loan buybacks and/or extra fees assessed by GSEs to correct your data. Remember, text is not data until it is usable and acceptable to GSEs.
****These vendors are touting the “text extraction” approach because they don’t have any other options. They are getting a first-generation PDF directly from the appraiser, but that PDF is just text and images. Many of the tick boxes on the form are not machine readable, and very little data is labeled consistently in the extracted text. We have pulled perfect text out of first-generation forms and it’s almost impossible to match fields to required data fields in XML. If human eyes can’t do it, how can software do it? It can’t – the result is lots of human review, with eyeballs on the original forms. This leads to wasted time for resubmissions, and data validation – not to mention human error. Each data point re-entered or re-assigned is another potential point of failure.
****How can you avoid this problem? Make sure that the actual computer-readable data from the appraiser is delivered right alongside the human readable form. When you’re evaluating appraisal management and data transmission for your compliance programs, only accept MISMO XML data transmission direct from the appraiser’s desktop. Providing MISMO XML straight from the appraiser allows lenders and AMCs to fully comply with FHFA’s newly mandated Uniform Mortgage Data Program (UMDP) program, and to submit appraisal reports to the UMDP’s “front end”, called the Uniform Collateral Data Portal (UCDP), without fear of rejection due to data extraction or data entry errors.
****Lots of vendors tell you that they are UMDP compliant, and we can hardly blame them. They have to be UMDP compliant if they want to work with GSE-backed loans. But ask them the most important question: Do you have native MISMO XML from the appraiser’s desktop? Or are you extracting the data from the form? It’s an easy question with only one right answer.