Case Study: How a Windshield Distributor Recovered $150K in Trapped Inventory

The Problem: $300K We Could Not See

We carried thousands of windshield SKUs across several vendors. We knew, in a general way, that too much cash was tied up on the shelf — orders we wanted to place kept getting delayed because the capital was not there. But our inventory system could tell us what we had; it could not tell us what to do about it. Asking "which specific SKUs are dead, and how much would we free up by clearing them?" produced a spreadsheet exercise that took days and was stale the moment it was finished.

Why the Tools We Had Failed Us

Two specific failures stood out.

Single-threshold classification. The standard rule for flagging dead stock is "no sale in X days." That rule is wrong in both directions. It flagged windshields we had just received as "dead" because they had not sold yet, and it missed genuinely obsolete glass that had sold once a year ago and never again. We could not trust the list it produced, so we did not act on it.

The multi-vendor variant blind spot. In auto glass, the same physical windshield is available from several vendors — NMC, XYG, PLK, CMX — each under its own SKU. Our system treated those as separate products. When a customer bought the part, the sale was credited to whichever vendor variant filled it, and the other three looked like they had never sold. The software wanted us to liquidate windshields that were, in fact, selling fine at the parent-part level. ABC analysis alone could not fix this — it needed variant-group awareness.

What We Built

We stopped relying on a single rule and built a classification that uses six signals together: forecasted next-period demand, sales-history depth, the most recent purchase-order date, ABC/XYZ class, variant-group sales attribution, and a manual operator override. Each signal acts as a fallback when a higher-priority one is missing.

Run against our own catalog, it finally separated three piles that had been lumped together: genuinely new inventory (recently received, no sales yet — leave it alone), special-order and catalog SKUs (never stocked to sell off the shelf — not a problem), and real dead stock (received long ago, no demand, no forecast — clear it). The variant-group attribution meant a windshield selling under any vendor code counted as selling, so we stopped getting false dead-stock flags on healthy parts.

The Results: $150K Recovered

The analysis surfaced close to $300,000 in dead and overstocked inventory we had not been able to pinpoint before. We acted on it in three ways:

• Liquidated true dead stock — windshields with no demand and no forecast, cleared at discount to recover cash and warehouse space.
• Stopped reordering slow movers — SKUs we had been quietly restocking out of habit, where the data said demand did not justify it.
• Redeployed the freed capital — into new SKUs our customers were actually asking for, and into funding orders we had previously been forced to delay.

Net, we recovered roughly $150,000 of working capital that had been sitting frozen on the shelf. The point was never just the discount on the liquidated glass — it was getting capital moving again. Money tied up in dead stock has a velocity of zero. Every dollar we freed went back to work.

Why We Opened It Up

We built this to fix our own warehouse. But the problem is not specific to auto glass, and it is not specific to us. Every small and mid-sized distributor we know is sitting on trapped capital they cannot precisely see — because the enterprise tools that classify inventory this well cost six figures a year and take six months to implement. So we turned what we built into Tru-Stock AI: the same forecast-first classification, priced and packaged for distributors our size, with self-serve onboarding in under 24 hours.

We are still our own first and most demanding customer. Every feature gets used against our real catalog before it ships to anyone else.