Inventory is the largest line item on most e-commerce balance sheets and the least examined. Too much stock ties up cash, drives storage surcharges, and eventually becomes dead weight you liquidate at a loss. Too little stock breaks the Buy Box, tanks organic rank, and hands customers to competitors during the exact weeks you spent months preparing for. The difference between the two failure modes is rarely luck; it is whether the brand runs inventory on rules or on gut feel.
This guide is the web edition of the full 26-page PDF. It walks through the core disciplines that separate well-run inventory operations from reactive ones: ABC classification, safety stock and reorder point math you can actually maintain, demand forecasting that improves each quarter, multi-warehouse allocation, and a dead stock process that stops slow movers from quietly consuming your margin.
Who This Guide Is For
- Operations leads managing inventory across one or more sales channels
- Founders whose cash is increasingly locked in stock they cannot explain
- International brands shipping into the U.S. with long, variable lead times
- Teams graduating from spreadsheet reordering to a rules-based system
What You Will Learn
- ABC analysis: classifying SKUs so attention follows value
- Setting safety stock levels that reflect real variability
- Reorder points and order quantities you can automate
- Demand forecasting fundamentals and how to improve accuracy
- Allocating inventory across a multi-warehouse network
- Preventing, detecting, and liquidating dead stock
Start With ABC Analysis
Not all SKUs deserve equal attention. Rank your catalog by trailing 12-month revenue contribution and split it into three classes:
- A items: the top roughly 20 percent of SKUs driving about 80 percent of revenue. Review weekly, forecast individually, never stock out.
- B items: the next 30 percent, driving about 15 percent of revenue. Review monthly with standard reorder rules.
- C items: the long tail. Review quarterly, order conservatively, and question whether each SKU should exist at all.
Run the analysis twice a year and after any major catalog change. The most common mistake is treating the classification as permanent; products migrate between classes, and yesterday's A item is often today's liquidation candidate.
Safety Stock That Reflects Reality
Safety stock exists to absorb two kinds of variability: demand spikes and supplier delays. A practical formula for A and B items: multiply your desired service factor (1.65 covers roughly 95 percent of variability) by the standard deviation of demand over your lead time. If the statistics feel heavy, start with a simpler proxy: safety stock equals average daily sales times the difference between your worst-case and average lead time, plus a demand buffer of 15 to 25 percent for volatile SKUs.
Two rules keep this honest. First, use actual lead times from your last 10 purchase orders, not the lead time your supplier quotes. For brands importing from Asia, door-to-door variability of two to three weeks is normal, and ocean freight disruptions make it worse. Second, recalculate quarterly; safety stock set during a promotion period will silently overstate demand forever.
Reorder Points You Can Automate
The reorder point is simple: average daily demand times lead time in days, plus safety stock. When available inventory (on hand plus on order, minus committed) drops below that number, you order. The discipline is in the inputs:
- Track lead time per supplier and per shipping mode, not one global number
- Include inbound receiving and prep time, which adds 3 to 10 days at most warehouses
- Net out unsellable and reserved units so phantom stock does not delay reorders
- Review open purchase orders weekly and chase confirmations, because a PO that slips two weeks invalidates the math
Order quantities should balance unit cost breaks against holding cost. A useful guardrail: no single order should exceed 90 days of forecast demand for A items unless a price break or freight consolidation clearly pays for the extra holding cost and risk.
Forecasting That Improves Every Quarter
Forecasting is not about being right; it is about being systematically less wrong. Baseline approach: use trailing 90-day sales adjusted for seasonality, stockout periods (backfill the demand you would have captured), and known events such as promotions or listing changes. Measure forecast accuracy monthly at the SKU level using absolute percentage error, and investigate your ten worst misses; the causes are usually stockouts, unplanned promotions, or competitor exits, all of which can be encoded into the next forecast.
At scale, statistical and machine learning models materially outperform spreadsheets because they pick up signals humans miss: pricing changes, marketplace search trends, cannibalization between SKUs. This is exactly the problem demand forecasting AI is built for, and it matters most for international brands whose long replenishment cycles amplify every forecast error into a two-month problem.
Multi-Warehouse Allocation
Splitting inventory across warehouses cuts delivery times and shipping cost, but it multiplies allocation decisions. Practical rules:
- Place A items in every node; keep C items in a single low-cost location
- Allocate by regional demand share, which you can read from your order history zip codes
- Hold 10 to 15 percent of network stock as an unallocated reserve to rebalance mid-cycle
- Set per-node reorder points; a network that is healthy in aggregate can still be stocked out on the coast that matters
- Watch split-shipment rates, since multi-box orders erase the savings distribution was meant to create
Brands entering the U.S. often over-commit to a single warehouse lease before they understand their demand geography. Using a vetted multi-warehouse network with flexible terms lets you follow the demand data instead of guessing in advance.
Dead Stock: Prevent, Detect, Liquidate
Dead stock is inventory with no realistic path to full-price sale. Prevention starts at purchasing: smaller first orders on new SKUs, and a rule that no reorder happens on any SKU with more than 120 days of cover. Detection is a monthly report flagging SKUs where days of supply exceed 180 or sell-through has fallen below 30 percent of forecast for two consecutive months.
Once flagged, act on a clock, because aged inventory only gets cheaper:
- Weeks 1 to 4: price promotion, bundling with A items, marketplace deals
- Weeks 5 to 8: secondary channels, off-price marketplaces, outlet listings
- Weeks 9 to 12: bulk liquidation or donation with tax documentation
- Beyond: dispose and take the write-off; storage fees on hopeless stock are a pure loss
The emotional mistake is anchoring to landed cost. That money is spent; the only question is which exit recovers the most cash and frees the most storage.
Key Takeaways
- Classify SKUs with ABC analysis so review effort follows revenue contribution
- Base safety stock on measured lead time and demand variability, refreshed quarterly
- Automate reorder points with honest inputs: real lead times, receiving time, net sellable stock
- Measure forecast error monthly and fix the causes of your worst misses
- Allocate across warehouses by regional demand and keep a rebalancing reserve
- Run dead stock through a timed escalation; storage fees never negotiate
Go Further
The mechanics in this guide work, but running them across channels, warehouses, and time zones is a full-time operation. Pi-Commerce manages inventory optimization end to end for international brands in the U.S., combining a negotiated warehouse network with AI-driven forecasting and real-time visibility through the Pi Data Center. If you would rather set the strategy and delegate the daily execution, talk to our team about what a managed inventory operation looks like for your catalog.