I was a stat analyst in the insurance industry, but I ended up testing and debugging COBOL programming when I was with the state and later on I ended up creating entire MS Access applications prior to Y2K. Since I was both the user and the programmer I brought projects in quicker and within a short period of time.
However, I did run into one project where I was initially told to create a policy rating engine for workers compensation policies. I know the algorithms to rate a policy better than the user did, but the scope of the project was poorly defined. What they actually wanted was a transaction system to replace the old mainframe system. The problem was that none of old data was migrated off of the mainframe so that when the policies were audited at expiration it would require the initial entries when the policy was written to be offset--there was no data available to offset so the original policy rating data had to be reentered. To top all of those problems off was that all of the data from the transaction system then needed to be transferred into a data warehouse (so I had to be able to distinguish which policies were originally rated on the mainframe against those rated on the replacement system). We only tested about 20 policies and the systems were not run in parallel to work out any discrepancies. It ended up being a humongous clusterf**k that required me to spend an extraordinary amount of time manually rating and correcting the transaction system and data warehouse, plus the user also had to do about three times the amount of work so she was upset at me. Considering that the only computer class that I had was Intro to BASIC in 1984 I felt particularly abused by the whole mess.