Finding a year of under-reported revenue in the wells nobody was checking
Design That Earned Finance’s Trust Back.
Problem
Finance was asked to automate a month-long quarterly reconciliation; nobody had said the real problem was that they’d stopped trusting the numbers.
Role
Sole product designer. I ran the reconciliation by hand, well by well, for a week before designing anything.
Scope
A $400M portfolio, ~100,000 wells, the finance team and the data team.
Result
A year of under-reported interest found across the small-well tail; lookback prep cut from a month to a week; a new anomaly-detection initiative with the data team.
Problem
We were asked to automate the quarterly lookbacks.
Every quarter, Raisa's finance team reconciles what each well actually paid against what it should have earned. They were understaffed, and the process took a month, pulling data from scattered sources, running manual calculations, well by well. We were asked to make it faster.
The legacy tool: every field, every table, for every well.
Devil lies in the details
I wanted hands-on experience with the flow, not just the formulas. I asked our backend engineer to export the raw data from the database, took the last lookbacks from finance, and ran the actual reconciliation well by well. It took about a week. I had to redo it multiple times because my numbers kept not matching theirs.
That's when it got interesting.
The reconciliation, run by hand: well by well, column by column.
Finance knew the tool's numbers didn't match theirs, and had quietly stopped trusting it
They never mentioned it in stakeholder interviews. It was something they'd internalized: they'd written the tool off. But when I ran the reconciliation, the tool turned out to be more accurate than their manual calculations.
The losses weren't in the big wells. They were spread across thousands of small ones, accumulating for a year; each discrepancy too small to notice on its own. Summed across the tail: hundreds of thousands of dollars of under-reported interest.
Solution
The well declares its own condition
The old tool showed everything. Finance trusted nothing. The redesign flipped the approach: show only what changed, and make the math behind every number visible.
Each interest figure expands into the formula for well interest, so finance could see exactly what the tool was doing. Divergence between the calculated and the overridden value is flagged on the line. If they could see the formula, they could trust it, or override it with a reason.
The well declares its own condition, and every number shows its math.
We dropped the full lookback table from the tool entirely; finance exports to their spreadsheets for that. Instead, the tool gives them a change log for quick triage. Focus on what changed, only. The tradeoff was deliberate: removing the full table meant finance couldn't do their entire workflow inside the tool, but it meant what the tool did show was trustworthy by design, not by faith.
The change log: what changed, and nothing else.
Outcome
The finding changed more than the tool
I brought the discrepancy to my PM and our CPTO. What started as “automate the quarterly lookback” turned into a new initiative with the data team to auto-detect anomalies across the entire portfolio, so this couldn't happen again without someone noticing.
1 week
of quarterly lookback prep, down from a month.
1 year
of under-reporting, stopped.
The brief was to make the report faster.
The problem was that nobody trusted the numbers.
The work was to make the math visible.