Forensic data analysis
Most times someone calls me, a number is wrong and nobody can say why.
Not wrong in an obvious way. The report runs, the totals balance, and the figure is still not what anyone in the room believes. Two systems disagree by four percent. A metric moved and no decision explains it. Someone changed how a field was counted three years ago and the note that said so is gone.
That is the work. I take the number apart, trace it back through whatever produced it, and establish what the records can actually support.
What that involves
Reconstructing how the number was made. Which field it came from, who entered it, what got joined to what, which rows were dropped along the way and whether anyone meant to drop them. Most discrepancies are not in the analysis. They are in a definition that changed quietly, or a join that silently discarded the rows that mattered.
Separating the mechanism from the correlation. Two things moving together is where the work starts, not where it ends. I keep going until there is an account of what is actually driving the result, or until it is clear the data cannot get you there. Both are real answers. The second one is the more useful of the two more often than people expect.
Saying what the evidence does not support. Sometimes the honest finding is that the question cannot be answered with the records that exist. I would rather establish that in the first week than produce a confident-looking number in the sixth that nobody should rely on.
Writing it down so it survives disagreement. A finding has to hold up in front of the people it inconveniences. That means the method, the exclusions, and the parts I am unsure about are written down alongside the conclusion, in language the person making the decision can follow.
What to send me
Describe the number you do not trust and what you expected instead. If there are two sources that disagree, say which two. I will tell you whether it looks answerable, roughly what it would take, and whether I am the right person for it.
TODO: confirm scope — if you need an expert witness, a certified fraud examination, or anything filed with a court, that is not what this is. What I do is establish what the records show.
The method does not depend on the industry. Much of my background is in property management and operations, so I know those systems best, but the failure modes are the same everywhere: definitions drift, systems disagree, and nobody wrote down why.