Judgment calls stay with finance: the Genesco build

What we found

The judgment calls stay with the finance team, and the build was shaped around that line. The client is Genesco Sports Enterprises. Every month, the finance team rebuilt job-profitability numbers by hand from Sage 50 exports. We built a two-pass web app that does the mechanical part of the close and hands the rest back. The reconciliation took the CFO about 4 days by hand and now runs in about 60 seconds. That earlier Field Note has the before and the after. This piece is about where the line sits inside the build, how the software holds it, and how we decided where to put it.

The figures here are Genesco's and they are public with their permission. They describe one company and one close. Your close will have its own shape.

Month-end by hand

Most of the old close was people moving numbers between files. Exports came out of Sage 50. Someone imported them, reconciled the totals across several spreadsheets, then re-keyed prior-month profit figures so the new month started from the right place. It was slow, and it was easy to get wrong.

The decisions sat at the end of all that. A job that needed a call about how it should be treated waited behind the importing and the re-keying, so the person making the call reached it after days of typing. That ordering is a thing we look for in an audit. When the judgment comes last in line, the calendar is being set by the mechanical work in front of it.

The two-pass build

The app runs the close in two passes with the finance team in the middle. In the first pass it ingests the Sage exports, computes the job totals, and pre-fills every figure that follows a written rule. It then hands back a clean workbook for the items that need a decision. In the second pass it takes the reviewed workbook and produces the final reconciled master.

The app remembers each month's numbers, so the next month starts pre-filled and nobody carries last month's agreed figures across by hand. Each run leaves a record of what went in and what came out, so the close can be checked afterwards.

The arithmetic runs on fixed rules written in code. The same files produce the same answer every time the app runs. A reconciliation has to behave that way, and written rules behave that way.

The handback workbook

The handback workbook is where the line lives. It is a familiar spreadsheet, and it holds the short list of items that need a decision with the arithmetic already done. A reviewer who disagrees with a figure changes it in the workbook, and the second pass takes that change as given. The decision is made in a file the finance team controls, and the app reads it as an instruction.

That arrangement is deliberate. Anthropic, writing about their own commerce agents, keeps the rules that matter in code outside the model and treats an approval typed in a chat window as setting nothing. Their reason for the split is that when the text goes wrong, "the failure is a misstatement to correct and no action needs reversing." Genesco's close has the same shape. The figures the app owns come from rules in code. The figures a person owns arrive as a changed cell in a workbook, which is a mark the software cannot make for itself.

Drawing the line used one test. If every step and every criterion for a good answer can be written down, the software can own the figure. If a person regularly weighs something nobody ever wrote down, the person keeps it. We drew it with the CFO, going line by line through a real month.

The test gets used again after the build ships. Twice, a rule we thought was fully written turned out to carry an exception that lived in someone's head. Those items moved back into the workbook, where a person decides and the record shows who decided. A rule belongs in code when it is complete.

The change at month-end

The close now opens with the decisions. The finance team reaches the short list first, with the sums and the carry-forwards already done, and the rest of the month-end work no longer stands in front of it.

Two other things changed. A category of error stopped, because transcription mistakes between files cannot happen when nobody types figures from one file into another. And the close can be run again. When a late number arrives, the team re-runs the pass and reviews a fresh workbook, which was hard to justify back when a re-run cost days.

What left the finance team was the searching, sorting and re-keying. What stayed was the part of the close they are paid for. Mark Tatum, Genesco's CFO, described it this way: "Every month our team burned hours rebuilding job-profitability numbers from Sage by hand. Plainpath built us a tool that does the heavy lifting and leaves us just the judgment calls. It has saved us real time and headaches, and they were quick to adjust it when we asked."

Limits

These numbers belong to one company and one process. Genesco's close had clear inputs, clear outputs, a large amount of manual effort, and a rule set that could be written down with the people who run it. That combination is what made it a good first process, and we wrote about picking one in which process to automate first.

A close where judgment enters at every step compresses far less. The useful question for a reader is how much of their own close is rule-following they could describe to a new hire in writing. That share is the share software can take.

The build also needs an owner on the client's side. At Genesco that is the CFO, who used it, told us what to adjust, and got the adjustments. A build with nobody asking for changes drifts away from the work, which is the pattern in why most AI pilots fail. We wrote about designing the review step itself in the human review step in an automated finance workflow.

The plain take

The design decision in this build was where to stop. The app owns the figures that follow written rules and leaves a record of each run. The finance team owns the figures that depend on context, and the handback workbook is how they say so. Drawing that line took one pass through a real month with the person who does the close. If you want the same line drawn on your own process, that work happens in Discovery, and it starts from the free audit.

Frequently asked questions

Which parts of a month-end close can software take over?

The parts where every step and every criterion for a good answer can be written down. At Genesco that meant importing the Sage 50 exports, computing job totals, filling in the figures that follow a written rule, and carrying last month's agreed numbers forward. Anything that depends on context, on a conversation, or on a call about how a job should be treated stayed with the finance team.

How does the Genesco app hand work back to a person?

It runs in two passes. The first pass produces a clean workbook holding the short list of items that need a decision, with the arithmetic already done. A reviewer changes the figures in that workbook. The second pass takes those changes as given and produces the final reconciled master.

What happens when a rule turns out to be incomplete?

The item moves back into the workbook and a person decides it. Twice at Genesco, a rule we thought was fully written carried an exception that lived in someone's head. A rule belongs in code when it is complete, and the handback workbook is where everything else goes.

Does the app use an AI model to do the reconciliation?

The live app does its arithmetic with fixed rules in code. A reconciliation has to give the same answer every time it runs on the same files, and written rules do that.

How do we find the line in our own process?

Start with the free audit, which is an interactive chat that asks about your work and scores it. Discovery is the paid engagement after it, where a person digs into your top processes and writes the plan, including which figures a build would own and which ones stay with your team.

References

1. The anatomy of effective commerce agents — Anthropic

The plain take

The design decision in the Genesco build was where to stop. The app owns the figures that follow written rules, computes them from the Sage 50 exports, and leaves a record of each run that can be checked afterwards. The finance team owns the figures that depend on context, and the handback workbook is how they say so. We drew that line with the CFO in one pass through a real month, and we moved items back across it when a rule turned out to carry an exception nobody had written down. The reconciliation took the CFO about 4 days by hand and now runs in about 60 seconds, and the share of a close that compresses that far is the share you could describe to a new hire in writing.