What a farmer's cash book taught me about records

In the autumn of 1995 I sat at a kitchen table in Reno County with a wheat farmer and his cash book. He had recorded a payment on a quarter-section of land as an operating expense, because it left the same account on the same day as his fertiliser bill. Nothing in the book was dishonest, and every figure in it was correct. The record simply answered a different question from the one our benchmark report was asking. I have spent thirty years meeting that farmer's cash book in other forms.

I grew up in central Kansas and studied agricultural economics at Ardellis State College, finishing in May 1994. My first job, at the Prairiefold Farm Records Association, was to reconcile member farms' records and prepare annual benchmarks comparing costs per acre and per head. I learnt to ask how an entry had been made before deciding what it meant, and to tell a member plainly when his figures could not support the comparison he wanted.

In 1999 I moved to Corvenhall Bank in Omaha as a regulatory reporting analyst. The bank lent to feedlots, grain operations and Nebraska businesses, and my work was the quarterly call report. I found that its schedules were exact about categories and silent about how a loan officer had chosen one. A purpose code entered to close a file on a Friday afternoon could decide where a loan appeared in a federal return. While I was there I completed a part-time master's degree in applied statistics at Pellwyn University.

From 2005 I worked on the other side of the relationship, at Arbelune Core Systems, which supplied core processing to community banks. I led data conversions, moving a bank's loans, deposits and customers from its old core onto ours over a single weekend. I saw how many banks carried fields whose meaning had drifted over the years, and how often a vendor release altered a definition without anyone at the client noticing. That experience shapes how I read our own contracts today.

I joined Kellanbrook Bank in March 2011 to run management information and regulatory-reporting data. In 2015 I took on data governance and built the bank's first fair-lending analytics team, doing testing work for the chief compliance officer and ownership work for the chief operating officer. I became chief data officer in February 2020. The role is narrower than the title suggests at larger banks. Technology operations, information security and credit decisions sit elsewhere. My teams own definitions, data quality, the reporting data behind our call report and mortgage disclosure submissions, the data feeding our models, and the analysis compliance uses to test lending for fairness.

The cases I describe all involved giving something up. Amending two call reports pulled credit administration away from operating-line renewals in the busiest weeks of their year. Reporting a pricing gap before our method was settled meant presenting a finding I could not yet explain, and it slowed lenders who compete with dealer finance and other farm lenders. Choosing data terms in a core contract cost the bank about US$370,000 a year in discount it could have taken. I would make each choice again, though not always in the same way, and each case says how.

I have also learnt the limits of my discipline. Data cannot tell a loan officer whether a family operation will survive a dry year, and a proxy estimate of a borrower's ethnicity is a statistical tool, not a fact about a person. I use machine-learning methods where a human can check the result, such as flagging loans whose collateral and purpose records disagree. I have declined to put generated summaries of borrowers' tax returns into credit memos until someone can show me how their errors would be caught.

I present to our audit committee twice a year on the controls behind the call report, and to the risk committee each year, alongside the chief compliance officer, on fair-lending testing. I have never served as a director. I would like to bring a board the view of someone who has worked inside the record at four points: the farm, the reporting desk, the vendor and the bank. I know how complete a filed figure looks on the page. I also know how much judgement sits underneath it, and I think directors should see some of that judgement before they attest to it.

Open to Board appointments

Languages / English