Somewhere right now, a warehouse is packed with products nobody can buy. Not because they are sold out. Because they are in the wrong place. IHL Group ran the numbers on this and landed at $1.7 trillion a year, which is more than six percent of every dollar spent in retail, gone. Not to competitors. Not to theft. To boxes sitting where no customer shops. The fix has a boring name: inventory allocation. Deciding which shelf each unit belongs on. Boring name, expensive consequences.
What This Actually Means
Take 5,000 units of a single product. It sells through a website, through Amazon, and across twelve physical stores. Somebody has to slice that pie: a chunk for Amazon, a chunk for the website, a portion per store, some held back. When that decision gets made carefully, every channel has what it needs. When it gets made on autopilot, the website runs dry on Tuesday while stores sit on three weeks of dead stock.
The customer in the middle of this does not complain. They wanted the product, could not find it, bought it elsewhere. That silent departure, multiplied across thousands of products, is where the trillions go.
And no, ordering more does not fix it. That is the first instinct, and it is wrong most of the time. Plenty of businesses drowning in stockouts have warehouses full of product. The product is just concentrated where buyers are not. More inventory in the wrong places is still wrong.
The Two Ways to Bleed
Overstock is the visible wound. Pallets accumulate, the monthly storage invoice arrives like clockwork, and eventually someone waves the white flag with a clearance event. The company pays to store the stuff and then pays again in lost margin. Painful, but at least there is a meeting about it.
Stockouts are the invisible one, with no meeting and no alert. A shopper hits a dead end, buys from whoever has it in stock, and the original store never learns what happened. Ask any retailer with twenty years in the business about the customers they lost this way. They will tell you they have no idea, which is exactly the problem.
Now run both problems through multiple sales channels. The website and the Amazon listing drink from the same well with no agreement on who gets what. Both sell the last forty units. One buyer gets a cancellation notice. Amazon treats cancellations as a character flaw, and the seller rating built over months takes the hit, because nobody set the split and everybody pays the price.
A Short History of Getting It Wrong
One store never had this problem. Fifty stores plus a website invented it. The question mutated from “how much should be ordered” into “where does every unit go,” and the second question turned out to be much harder than the first.
Early on, the answers were folk wisdom. Ship more to stores that sold more last quarter. Stock up before the holidays. Fine when the business was simple. Then came e-commerce, marketplaces, and customers expecting delivery tomorrow. The folk wisdom started failing quietly while companies kept using rules built for a simpler era.
Eight Methods, Ranked by Honesty
No single approach works everywhere. Most competent operations blend several. Here is what each one actually does, with the marketing stripped off.
Push: Bet on the Forecast
Look at last year’s sales, predict what each location will need, ship inventory before demand shows up. Winter coats head north in September. When the prediction holds, this runs beautifully. When it does not, real money was spent delivering products to the wrong addresses. Push is a bet, and the uncomfortable part is that most companies are worse at forecasting than their planning meetings suggest. It feels decisive, but decisive and correct are different things.
Pull: Wait for the Sale
Nothing ships until a customer buys. A location burns through its stock, the system notices, replenishment rolls out. Inventory stays thin because it only moves toward proven demand. For most businesses, this is the same starting point, less glamorous than forecasting but harder to get badly wrong. The weak spot is reaction time. A demand spike hits, the system is one shipment behind, and competitors are already fulfilling the orders.
Just-in-Time: Live Dangerously
Pull with the safety net removed. Products arrive more or less as needed, buffers near zero across the chain. Toyota made this famous, and for Toyota it works, because Toyota controls its supply chain the way a conductor controls an orchestra. Most retailers are not Toyota. Most retailers have suppliers who miss deadlines, trucks that break down, and ports that close. One failure cascades through the whole system. The wave of companies quietly rebuilding safety stock over the past few years was not being cowardly; they were being realistic. For anyone without Toyota-level supply chain control, JIT is the most overrated idea in this list.
Centralized: One Brain
A single team makes every allocation call for every location. During shortages, this is genuinely useful, because someone with the full picture can prioritize instead of letting every store grab what it can. The cost is local blindness. Headquarters does not feel what a regional manager feels walking their floor. They see numbers, and numbers miss things. Centralized suits companies where every market behaves roughly the same and struggles everywhere else.
Decentralized: Many Brains
Hand the decisions to regional teams who know their territories. In diverse markets, this beats headquarters guessing from a spreadsheet. The catch arrives with scarcity. Ten regions want the same limited units, nobody has final authority, and the allocation starts reflecting internal politics instead of customer demand. The region with the loudest director wins. That is not a supply chain strategy; that is an office dynamic wearing a costume.
Demand-Based: Chase the Signal
Watch what is selling right now and feed stock toward it. Product takes off in Chicago, Chicago gets more units. Simple, adaptive, grounded in what is actually happening instead of what was predicted. This is probably the most practical method available to modern retailers. The fraud version is everywhere: companies claiming demand-based allocation while working from stale numbers.
Rule-Based: Automate the Obvious
Top ten stores get served first. Twenty percent stays reserved for the website. Fill everything else in rank order. Rules like these run fast and handle catalogs where individual judgment is impossible. The rot sets in silently when a rule written eighteen months ago keeps executing while the business around it changes.
AI-Based: Trust the Model
Pour sales velocity, margins, seasonality, weather, and forty other signals into a model that reallocates continuously. IHL Group ties this to measurably faster growth for companies that do it well, and the results are real. So is the prerequisite: clean, deep, honest data. Models do not fix bad inputs; they accelerate them. A company with messy records gets wrong answers at machine speed, delivered with total confidence. Sort the data before buying the software, not after.
Side by Side
| Method | How it works | Where it shines | Where it breaks |
| Push | Ship on forecast | Stable demand, good predictions | Bad forecasts waste money |
| Pull | Ship after sales | Lean operations, sane default | Slow on sudden spikes |
| Just-in-time | Ship at the last moment | Toyota-level supply chains | Almost everyone else |
| Centralized | HQ decides all | Shortages, uniform markets | Diverse regions |
| Decentralized | Regions decide | Varied local markets | Scarce supply, politics |
| Demand-based | Follow live sales | Modern retail, real data | Stale data dressed up |
| Rule-based | Fixed policies | Huge catalogs | Rules nobody reviews |
| AI-based | Models rebalance | Clean data, scale | Messy data, false confidence |
How Companies Sabotage Themselves
The classic move is allocating from expired numbers. Last quarter’s data drives this quarter’s decisions while the market has already moved on. The current numbers exist in the system, but nobody opens them before the meeting.
Then there is the equality trap. The website gets the same as the flagship store because that feels fair. It is not fair; it is lazy. Different channels have different customers, different velocities, different return rates. Equal splits ignore all of that.
The shared-pool collision is next. Every channel pulls from one pile with no priority rules. Two channels sell the same final units. Somebody’s order gets cancelled. On Amazon, that cancellation follows the seller for months.
And the slow rot of unreviewed rules. Markets shift, channels grow unevenly, products cycle. The allocation policy from two years ago keeps running because changing it requires a meeting nobody wants to call. Meanwhile, it misdirects stock every single week.
What Actually Works
Visibility first, with real-time numbers across every channel and every warehouse. Allocation is arithmetic, and arithmetic with bad inputs produces bad outputs, so fix the inputs.
Forecast at the level where decisions get made: specific product, specific location. Blended averages hide the exact variations that determine where units should go. Anyone allocating from company-wide averages is guessing with extra steps.
Size the safety net per product, not per policy. Steady sellers from reliable suppliers need almost nothing in reserve. Seasonal items with long lead times need plenty. One buffer rule for everything manufactures overstock and stockouts simultaneously, which is quite an achievement.
Revisit on a calendar, not on a crisis. Monthly or quarterly, sit down with the allocation rules and ask what changed. Seasons turned, a channel grew, a product died. Static rules in a moving business are just slow-motion errors.
Know when the spreadsheet years are over. At some point manual allocation becomes theater. An ecommerce fulfillment partner takes the whole function- storage, allocation, and dispatch together. Understanding how ecommerce shipping fits in helps make that call. For more on methods and providers, this guide to inventory allocation covers the options in depth.
Eight Questions, Straight Answers
Deciding where every unit of stock should sit so each sales channel has enough without cash getting trapped in excess.
Management is everything: buying, storing, tracking, selling. Allocation is one slice: the distribution of available units across the places that sell them.
Push, pull, just-in-time, centralized, decentralized, demand-based, rule-based, and AI-driven. Almost nobody uses just one.
When done right, yes. Supply sits where buyers are instead of draining from one shared pile, and buffers plus rebalancing do the quiet work.
Absolutely, at small scale. Reserve a fixed share for the website, split the rest by past sales, and review quarterly, since an afternoon of discipline prevents most of the damage.
Fill rate, the percentage of orders shipped complete. Back it up with stockouts per location, turnover speed, and carrying costs. All of them should move together.
Mostly, yes. Software enforces rules without fatigue and models rebalance on their own, though it also enforces bad rules without fatigue, so the data and the reviews matter more than the tool.
Setting the rules and never looking at them again, because everything drifts and unmaintained policies become the problem they were designed to solve.
Final Thoughts
Good allocation is invisible, and that is the entire point. Orders go out whole, shelves look right, and nobody has a story about it. Behind that calm sits a set of explicit rules, numbers worth trusting, and someone revisiting both. Build that before buying software.