For product businesses, no number on the income statement is bigger—or more error-prone—
than cost of goods sold. And no year-end ritual is more dreaded than the physical inventory
count. With December approaching, this week is about getting both right, and how AI has turned
inventory from a guessing game into a data discipline.
Why accuracy matters more than it seems. Inventory errors don’t stay in the warehouse; they
flow straight through your financials. Overstate ending inventory and you understate COGS,
which overstates profit—and your tax bill. Understate it and your margins look worse than
reality, distorting every pricing and purchasing decision you make. Your gross margin is only as
trustworthy as your inventory records.
The count itself. If you’re doing one big year-end count, plan it now: schedule it for a lowactivity
window, freeze receiving and shipping during the count, count in teams with fresh eyes
checking each other, and investigate significant variances before adjusting the books—a big
discrepancy is information (shrinkage, receiving errors, theft, or record-keeping failures), not just
a journal entry. Better still, move toward cycle counting: continuously counting a rotating slice of
SKUs all year, prioritized by value and velocity. AI-driven inventory systems make this
dramatically more effective by directing counts toward the SKUs whose records are statistically
most likely to be wrong—items with irregular movement patterns, frequent adjustments, or
variance history—so you find errors months sooner with a fraction of the labor.
Costing methods, briefly. FIFO, weighted average, and (where permitted) LIFO will produce
different COGS in an inflationary year, with real tax consequences. Changing methods is a
decision for your accountant, not December improvisation—but it’s worth a conversation if
you’ve never had one.
The AI layer. Beyond counting, machine learning now touches the whole inventory lifecycle:
demand forecasting sets reorder points that adapt to seasonality instead of gathering dust;
anomaly detection flags shrinkage patterns by location, shift, or SKU; and margin analytics
compute true landed cost per product—purchase price plus freight, duties, storage, and
returns—revealing that some “profitable” products aren’t. Businesses running this analysis for
the first time almost always find a product they should discontinue and one they should double
down on.
A Thanksgiving-week note. For retailers, this weekend is the year’s biggest data event.
Whatever happens—stockouts, surpluses, surprise winners—capture it. Next year’s forecasting
model is trained on this year’s records.
Count carefully, cost honestly, and let the data tell you which products actually earn their shelf
space.
Next week: the skills question—building a finance team for the AI era.


