What we found
The Genesco Sports Enterprises close runs in two passes, with the finance team's review sitting in the middle. A month-end reconciliation that took the CFO about 4 days by hand now runs in about 60 seconds. The work that needs a finance brain is still done by the finance team. It moved to the front of the close, it arrives with the arithmetic already done, and it covers a short list of items.
This note is about where we drew that line and what holds it in place. The figures here belong to Genesco and are public with their permission. They cover one company and one process, and your close will have its own shape.
The two-pass build
The app makes two passes over the month, and the finance team works in the gap between them. Every month, the finance team rebuilt job-profitability numbers by hand from Sage 50 exports. The first pass reads those exports, adds up each job, and fills in every amount a written rule covers, then hands back a clean workbook. A person opens that workbook and settles the items that need a decision. The second pass takes the reviewed workbook and produces the final reconciled master.
We chose two passes because a single pass leaves only two options, and both are bad. Either the software decides items that nobody ever wrote a rule for, or a person does the arithmetic again by hand to check it. The gap between the passes is where a reviewer decides how a job should be treated, with the sums already settled.
The shape a user sees is an upload, a review and a download. Someone uploads the Sage exports, works through the workbook, and downloads the master. Each run leaves a record of what went in and what came out, so the close can be audited afterwards.
The pre-filled figures
An amount belongs to the first pass when every step and every criterion for it can be written down. At Genesco that means the sums across the exports, the totals per job, and the prior-month profit amounts that used to be re-keyed by hand. Each of those steps was arithmetic and typing.
The app also remembers the amounts agreed each month, so the next month starts pre-filled. Last month's agreed numbers are already sitting in the new month. That one piece of memory removed a whole step from the close, because nobody carries amounts across files any more, and it is what makes the second month lighter than the first.
Pre-filling changes what a reviewer looks at. A reviewer now reads a job that has already been added up and decides whether the costs on it belong there and whether the treatment is right.
The judgment calls
The judgment calls stay with the finance team. An amount that depends on context, on a conversation with an account lead, or on a call about how a job should be treated is one a person owns. We drew the line where the client drew it, and the test is plain. If you can write down every step and every criterion for a good outcome, software can own it. If a person regularly weighs something that was never written down, the person keeps it.
The app presents those cases in a workbook, which is where a finance team already works. A reviewer who disagrees with a pre-filled amount changes it in the workbook, and the second pass takes that change as given. The reviewer has the last word on every line, by design.
Three things changed for the people doing the close. The decisions come earlier, because nobody spends days shuffling files before reaching them. The searching, sorting and re-keying that filled the week is gone, and that work never added anything to the judgment a finance team is paid for. A whole category of error ended, because a transcription mistake between two files cannot happen when nobody types amounts between files.
Controls written in code
The rules that matter sit in code, outside any model. Anthropic's write-up of its reference commerce agents makes the case for the split. Their agents keep provenance, staging, caps and the tool list in the executor, so the same checks run on every path. Their line on why that split matters is that when the model breaks a prompt rule, "the failure is a misstatement to correct and no action needs reversing", because every write behind the text already passed a check in code. You can read their account in the anatomy of effective commerce agents.
We use the same idea when we write what "human review" means into a finance build. A review built into the run is a gate the close has to pass through to finish. In the Genesco flow the final master is produced from the reviewed workbook. The app does not mark its own work as approved, and the record of each run shows what went in and what came out.
Ask any vendor who says a person stays in the loop what their software does when the person skips the step.
The owner and the checker
A build with a named owner on the client's side is a production system from its first week. Someone has to run it on the real work, someone has to check what it produces, and someone has to say by when it should have paid for itself. A build with none of those three is a pilot, whatever it gets called in the kickoff.
At Genesco the owner is the CFO. He ran the close on it, told us what to change, and got the changes. The build shipped, runs live, and was refined on his feedback. 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."
We wrote about what happens when nobody owns a build in why most AI pilots fail.
Limits
These figures describe 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, which is what made it a good first process. A close where judgment enters at every step compresses far less, and we say so before anyone quotes these numbers at their own board.
The review design also depends on someone being willing to say where the line sits. That conversation took the CFO's time. When nobody will commit to which amounts follow a rule, the reviewer ends up checking every line again.
The plain take
Drawing the line between the two passes took a working session with the CFO about which amounts follow a written rule and which ones need a decision. A finance team can make that list on its own close this month, with a pen and the last set of files. If you want a person to check that map and write a build plan against it, that is what we do in Discovery.
Frequently asked questions
What does the human review step do in an automated month-end close?
It settles the amounts that no written rule covers. In the Genesco Sports Enterprises build, the app adds up each job and fills in what the rules cover, then hands back a clean workbook. The finance team decides the items that depend on context, and the second pass builds the final reconciled master from that reviewed workbook.
Why build two passes instead of one?
A single pass means either the software decides items nobody wrote a rule for, or a person redoes the arithmetic by hand to check it. Two passes put the review in the gap, so the reviewer sees a short list with the arithmetic already done.
How do you decide what the software fills in and what a person decides?
The test is whether every step and every criterion for a good outcome can be written down. If it can, software can own it. If a person regularly weighs something that was never written down, the person keeps it.
How do you know a human review step is real?
Ask what the software does when the person skips it. A review built into the run, where the final output can only be produced from a reviewed file, is a gate the process has to pass through.
Will our close get faster too?
We cannot promise that. The Genesco Sports Enterprises figures belong to one company and one process with clear inputs, clear outputs and a rule set that could be written down. A close where judgment enters at every step compresses far less.
References
- The anatomy of effective commerce agents — Anthropic
The plain take
Drawing the line between what software fills in and what a person decides took a working session with Genesco's CFO about which amounts follow a written rule. A finance team can make that list on its own close, with a pen and the last set of files, marking each line as rule-bound or as a decision someone makes. If you want a person to check that map and write a build plan against it, that is what we do in Discovery.
If one of your processes looks like this, the free audit at /compass is where to start.