OCR Isn’t a Total Solution

Optical character recognition, or OCR sounds like a dream come true for data entry in the mortgage business. Simply scan in tax documents and watch forms get automatically filled out and calculations materialize before your eyes. Ahh, if it were just that easy. OCR is a labor saving technology, but be careful you trust it too much. I do feel that OCR could do some of the light lifting during an income review. For example it can be trusted to pull most required underwriting data off of tax returns, W-2, K-1, and a few other standard tax forms. As promised in the first blog we wanted to expand on these points we reviewed in an effort to share data with other people in the mortgage business that might be tackling the same challenge of improving income review.

The biggest gap OCR has is that a majority of borrowers are employed versus self-employed that means you have find another solution for 65-70% of the loans.

The first example is that OCR can only handle just a few of the thirty-plus income types explained in the agency guidelines. The biggest gap OCR has is that a majority of borrowers are employed versus self-employed that means you have find another solution for 65-70% of the loans. The reason for the gap in coverage is there are a few hundred variations of paychecks and the formats they come in. These variations range from layouts, to how year to date calculations displayed, to the abbreviations used for income and deductions. A simple 401K loan can have several abbreviations some of the examples are TESP, 401K Loan, 457 Loan, Retirement Loan, just to name a few. Take this same problem with any non-uniformed pay statement and the gap widens to retirement income, foster care, child support, VA benefits, all of which have no uniformed pay statements for the computer to read.

even if OCR can read the income form it does not mean that income can or should be used

The second example is even if OCR can read the income form it does not mean that income can or should be used. Let’s says you receive a loan application from John Doe, John is married, but his wife Mary is not on the loan. If you read their tax returns in most cases it has joint income reported. So if the OCR sees income in the Schedule C line (line 12) how will OCR know that the income is John’s or Mary’s? Or if there is social security income listed on 20A how much is earned by John alone?

Our final point was that OCR just cannot provide guidance, training, or safety checks on the income reviewed. On just the two points mentioned above if an untrained user just simply scanned the documents and used the raw numbers provided by the OCR review the QC problems would run most companies out of business. When I train underwriters about income I have to train them to think about many things for one income type. Things like whose income is it, is the income going to continue, what is the income trend in the last three years, do we have the proper forms for the income provided. Bottom line income review is more than just numbers.

So our conclusion, yes OCR can be used to fill in a few fields to speed things up, but don’t leave it to make the final decision on what number goes on your 1003 form.