How AI Saved a Small Factory $200K in Supply Chain Waste

The Factory That Was Drowning in Its Own Inventory

Meridian Precision Components makes stamped metal brackets, housings, and fasteners for three Tier 1 automotive suppliers in the Midwest. Forty-seven employees, $18 million annual revenue, and a 60,000-square-foot plant in Akron that has been running since 1987. They are the kind of company that does not make headlines. They make the small parts that hold larger parts together, and they have been doing it well enough to survive three recessions.

But by early 2024, Meridian’s supply chain was quietly bleeding money. Their plant manager, who had been running operations for fourteen years, knew something was wrong but could not pinpoint exactly where the waste was accumulating. Finished goods sat on shelves for months before shipping. Rush orders from one customer triggered overtime that threw off production schedules for everyone else. Raw steel coils arrived three weeks early and sat in the yard, tying up cash. Other times, the same material showed up two weeks late, forcing expedited freight at four times the normal rate.

The numbers told a painful story. $87,000 in dead stock — finished parts that sat unsold for over 120 days. $64,000 in expedited shipping charges for materials that should have been ordered earlier. $52,000 in scrapped materials from production runs that overshot actual demand. Total: roughly $203,000 in waste that was entirely preventable, if anyone had been able to see the patterns in time.

The problem was not incompetence. It was complexity. Meridian tracks 1,400 active SKUs across 23 customer accounts, each with different lead times, order cadences, and seasonal fluctuations. Their planning ran on a combination of Excel spreadsheets, an aging ERP system from 2011, and the plant manager’s intuition built over a decade of watching orders come in. That intuition was good. But it was also slow, and it could only hold so many variables at once.

In March 2024, Meridian’s ownership group approved a $45,000 investment in an AI-powered demand planning and inventory optimization platform. It was not a flashy decision. No press release. No digital transformation roadmap. Just a practical bet that a machine might notice things humans were too busy to see.

Demand Forecasting: Where the AI Found Money Nobody Knew Was Missing

The first module Meridian deployed was demand forecasting. The platform ingested three years of order history, customer PO data, raw material pricing, and even publicly available automotive production schedules from their customers’ parent companies. Setup took about four weeks, including data cleaning that exposed some embarrassing gaps — several hundred SKUs had inconsistent unit-of-measure codes across systems, and one customer’s orders had been logged under two different account numbers for seven months.

The initial results were underwhelming. For the first eight weeks, the AI’s forecasts were roughly as accurate as the plant manager’s manual estimates — within about 15 percent of actual demand on a weekly basis. But by week twelve, something shifted. The model started picking up on cadence patterns that no human had documented. One Tier 1 customer, it turned out, reliably increased bracket orders by 22 percent in the six weeks before their own model-year changeover. Another customer’s ordering pattern correlated strongly with steel futures prices — when steel prices rose, they front-loaded orders to lock in Meridian’s existing quotes.

By six months in, Meridian’s forecast accuracy had improved from roughly 68 percent (their historical average using manual methods) to 91 percent at the weekly SKU level. That improvement sounds technical and abstract until you translate it into what it meant on the shop floor: they stopped making parts nobody had ordered. Production planning shifted from reactive — scrambling when a big PO arrived — to proactive, with schedules built around predicted demand windows. The result was a 34 percent reduction in dead stock inventory within the first six months.

This tracks with broader industry data. Companies deploying AI-driven demand forecasting typically report 20 to 50 percent improvements in forecast accuracy and 20 to 30 percent reductions in inventory carrying costs. Meridian’s results fell squarely in that range, which is the honest thing to report. The AI was not performing miracles. It was performing basic pattern recognition at a scale and speed that humans cannot match.

Inventory, Logistics, and the Domino Effect of Getting One Thing Right

Better demand forecasting created a cascade of downstream improvements that Meridian had not fully anticipated. When you know what customers will order before they order it, every other decision in the supply chain gets easier.

Raw material ordering became surgical. Instead of placing monthly bulk orders for steel coils based on rough estimates, the AI recommended staggered weekly orders aligned with predicted production schedules. This cut Meridian’s average raw material inventory from 23 days of supply to 14 days — freeing up approximately $140,000 in working capital that had been sitting in their storage yard as idle steel. Their expedited freight charges dropped from $64,000 annually to under $11,000, because materials arrived when they were actually needed rather than on an arbitrary monthly schedule.

Production sequencing stopped fighting itself. The AI’s scheduling module grouped production runs to minimize die changeovers on their stamping presses. Die changes take 45 minutes to two hours depending on the part, and Meridian was averaging 11 changeovers per day across three presses. The optimized schedule reduced that to seven changeovers per day by batching similar part geometries together. That recovered roughly 16 hours of productive press time per week — the equivalent of adding a fourth press without buying one.

Outbound logistics got smarter without trying. Meridian ships via a mix of LTL carriers and their own two box trucks for local deliveries. The AI did not directly optimize routing — that was not its primary function — but better demand visibility meant they could consolidate shipments. Instead of sending half-full trucks three times a week to their largest customer 40 miles away, they shifted to two full-truck deliveries. Fuel and driver costs for local delivery dropped roughly 28 percent. Industry-wide, companies using AI-optimized logistics routing report 15 to 30 percent reductions in transportation costs, according to case studies from Singapore-based logistics firms and European freight operators that have deployed similar platforms at larger scale.

Quality Control and Supplier Risk: The Parts Nobody Talks About

The supply chain savings at Meridian were dramatic enough to justify the investment on their own. But the AI surfaced two additional problems that turned out to matter just as much.

Quality defect patterns became visible. Meridian’s stamping presses produce parts at tolerances measured in thousandths of an inch. They run statistical process control on critical dimensions, but their SPC system was reactive — it flagged out-of-tolerance parts after they were already stamped. The AI platform, drawing on production data and machine telemetry from press sensors, started identifying pre-failure signatures. Specifically, it detected that a gradual increase in press vibration amplitude, combined with ambient temperature above 88 degrees Fahrenheit in the plant, preceded dimensional drift in their tightest-tolerance bracket by about 90 minutes. That gave operators a window to adjust or perform preventive die maintenance before scrapping a run.

The impact was measurable. Scrap rates on their five highest-volume parts dropped from 3.2 percent to 1.8 percent over the first year. At Meridian’s production volumes, that translated to roughly $38,000 in saved material and avoided rework. These numbers align with broader manufacturing data — companies deploying AI-driven quality inspection systems report 30 to 50 percent reductions in scrap and rework within the first 12 to 18 months, with automotive manufacturers like BMW and Toyota documenting similar gains in their own AI quality programs.

Supplier risk stopped being a surprise. Meridian buys steel from four domestic mills and two importers. Before the AI, their supplier evaluation was annual — a spreadsheet scoring each supplier on price, quality, and on-time delivery. The AI platform added real-time monitoring by tracking publicly available data: steel mill capacity utilization reports, shipping port congestion indices, trade policy announcements, and even local weather patterns that could affect rail transport. In August 2024, the system flagged one of Meridian’s steel importers as high-risk three weeks before the supplier actually missed a delivery — port congestion in Houston had increased 40 percent, and the importer’s typical transit corridor was backed up. Meridian shifted that month’s order to a domestic mill and avoided a production stoppage that would have cost an estimated $25,000 in idle labor and missed delivery penalties.

This kind of predictive supplier management is growing rapidly. Nearly three-quarters of manufacturers now report losing money to supplier disruptions, with the average annual cost reaching $228 million for large U.S. companies. AI systems that monitor external signals — weather, port data, geopolitical risk, financial filings — can flag disruptions weeks before they hit, giving procurement teams time to react rather than scramble.

▲ Against a $45,000 implementation cost and roughly $18,000 in annual licensing, Meridian’s AI investment paid for itself in under five months. Their second-year projected savings are higher, because the models continue learning and the quality-prediction algorithms get more accurate with each production run.

Frequently Asked Questions