Implementation Engagement · The Close, Before and After

Three days of spreadsheet archaeology became a four-hour close

The old close, hour by hour across three days, set against the new close: a pack that drafts itself at 7:00am and four hours of human judgment where 24 hours of manual assembly used to be.

Executive band

The close went from 3 days to 4 hours. The consolidation motion behind it went from 15 hours a week to about 2 hours of review.

Same company. Same six sources. Same accounting ledger. What changed is that the assembly work moved into a pipeline and the humans kept only the judgment.

3 days → 4 hrs

Month-end close duration. Same-day close on day one of the month.

6 sources

Feeds that had to agree every month, and never did, until the pipeline reconciled them nightly.

38 → 0

Manual copy-paste steps per close. A 12-point review checklist replaced them.

15 hrs/wk → ~2 hrs

Weekly consolidation motion, now about 2 hours of review instead of hands-on assembly.

The old close · 3 days · 24 working hours

Three 8-hour days, every month, to learn what happened last month

The ops lead was the ETL layer. Six exports, one master workbook, 38 manual steps, and a founder who got numbers already 3 weeks stale for the decisions he needed them for.

Worst pain Manual work

Day 1 · 8 hours

Exports and cleanup

Pull six exports

Shopify, Amazon settlements, 3PL WMS, freight invoices, processor payouts, the returns log. Six formats, six vocabularies.

Manual

Fix date ranges

Each source cuts the month on its own boundary. Every export gets trimmed by hand to match the calendar month.

Manual

Re-key wholesale invoices

12 retail accounts, invoices re-typed into the workbook line by line.

Manual

Paste into the master workbook: 38 manual steps

38 copy-paste steps, in order, from memory. Miss one and Day 2 gets worse.

Worst pain

Formulas break

At least one tab breaks every month. Row counts shift, references slide, someone hunts the #REF!.

Manual

End of Day 1: data is in one file. Nothing agrees yet.

Day 2 · 8 hours

Reconciliation disputes

Shopify vs 3PL standoff

Shopify says shipped. The 3PL says it never left. Someone has to decide which system is lying, order by order.

Manual

Amazon settlement mismatch

The settlement report does not match the orders. The gap gets chased or, eventually, shrugged at.

Manual

The monthly "net sales" argument

Founder math vs CFO math. "Net sales" meant three different things depending on who built the tab. The argument re-runs every single month.

Worst pain

Freight allocated by guess

Freight invoices do not map to orders, so cost per order is an estimate defended from memory.

Manual

End of Day 2: numbers dispute settled by exhaustion, not evidence.

Day 3 · 8 hours

Rebuild and publish

Rebuild what Day 2 broke

Every dispute resolution meant edits. Edits meant broken tabs. The morning goes to putting the workbook back together.

Manual

Produce the pack

Tables copied into the report. Formatting by hand. Commentary written against numbers nobody fully trusts.

Manual

Founder reads stale numbers

By publication, the numbers are already about 3 weeks stale for the decisions he needed them for.

Structural

"final final v3.xlsx"

At least one number changes after publication, every month. The filename tells the whole story.

Worst pain

End of Day 3: pack shipped. Corrections already incoming.

The six sources · 3 versions of "net sales"

Six feeds that had to agree, and never did

Each source spoke a different language. "Net sales" meant three different things depending on who built the tab.

Shopify exportsCalled orders shipped that the 3PL said had not left the building.

Amazon settlement reportsSettlement totals did not match the orders they were supposed to settle.

3PL WMS exportShipments, inventory, and returns received, all cut on the warehouse's own date ranges, not the calendar month.

Freight invoicesNever mapped to orders, so freight per order was allocated by guess.

Payment-processor payoutsPayout totals never tied cleanly to the sales tabs built beside them.

The returns logA spreadsheet CS maintains on its own schedule, disagreeing with returns received at the 3PL.

The new close · 4 hours · same-day

Day one of the month, 7:00am: the pack drafts itself

The pipeline assembles the pack from the same nightly views the daily dashboard reads. The four remaining hours are all judgment: review, decide, publish.

Control layer Human judgment

7:00am · 0 hrs human

The pack drafts itself

Assembled overnight from the pipeline. No exports, no pasting, no rebuilds.

2 hrs

CFO reviews flags and annotates

Works through the reconciliation flags and adds annotations. Nothing auto-corrects, ever.

1 hr

Founder and CFO walkthrough

One hour on the numbers and the decisions they call for. No argument about definitions: one signed version of net sales.

1 hr

Publish with commentary

Pack goes out same-day with commentary attached. No "final final v3.xlsx".

The 12-point review checklist · replaced 38 manual steps

Zero assembly steps remain. Twelve review checks replaced them.

All six sources loaded in the overnight run, full month covered.

Reconciliation flags reviewed, each with a resolution or an owner.

Metric definitions unchanged from the signed set, or a change noted.

Channel table checked against plan: DTC, Amazon, wholesale.

Returns agree between the returns log and the 3PL's returns received.

Settlement tie-out confirmed: Amazon settlements against orders.

Freight per order reviewed against the trailing average, no guessing.

Inventory position sanity-checked: sellable by location, aging flags.

Cash-relevant flags confirmed on the one-page summary.

Open items listed with owners in the reconciliation appendix.

Decisions page complete: each called decision backed by its numbers.

Commentary attached and published. One version. No corrections queued.

How the four hours became possible: the system-design exhibit shows the machinery underneath, the nightly pipeline across all six sources, the 14 signed metric definitions that ended the net-sales argument, and the reconciliation agent that flags mismatches for human review every night.