Anonymized, but real.

A home décor and furnishings manufacturer in the $200–500M range, running its own delivery fleet alongside a 15-site distribution network covering the United States and Canada.

The business was contracting, and a network sized for a larger company had become a fixed cost it could no longer carry. Closing sites was the obvious lever and also the dangerous one: customers across both countries had been promised two-day delivery, and the network existed to keep that promise.

• Network analysis modelling service coverage, freight cost and inventory position for every candidate configuration, rather than closing the sites that looked most expensive in isolation.
• Worked out how far the company’s own delivery fleet could be extended, which turned out to be the hinge: fleet reach, not warehouse count, was what actually determined the two-day window.
• Ran it in three phases — analysis, then a pilot on one region to prove the model against reality, then replication across the remaining network.

• More than half of the 15 distribution centers closed.
• Two-day delivery coverage across the United States and Canada preserved throughout.
• Inventory and operating costs both reduced.

18 months, across analysis, pilot and full replication

A consumer products business in the $200–500M range, selling through national retail accounts.

One of its largest retail customers filed for bankruptcy. The demand signal that had anchored planning vanished, and fill rate, inventory and production capacity all went wrong at once — the kind of shock that exposes a planning process built on assumptions nobody had revisited.

• Mapped the existing planning process end to end, so the conversation was about what actually happened each month rather than what the procedure said should happen.
• Ran Kaizen events with the people doing the work to strip out the steps that existed only because they always had.
• Built a four-step S&OP cycle in their place, with clear owners and a decision at the end of each pass rather than a status meeting.

• Fill rate improved and inventory levels brought back under control.

An industrial products manufacturer in the $25–75M range, supplying big-box retail accounts.

COVID supplier disruption pushed order fill rates down into the 60s. At that level the problem stops being operational and becomes commercial: big-box accounts do not tolerate a supplier that cannot ship, and losing one of them would have taken a large share of the business with it.

• Selected and deployed forecasting software, rather than trying to carry the forecast in spreadsheets through a period when demand was moving weekly.
• Trained the planning team on it — the step most often skipped, and the reason most forecasting tools end up unused six months after go-live.
• Linked the forecast into the S&OP process so it drove purchasing and production decisions instead of sitting beside them as a reference.

• Fill rates recovered from the 60s.
• The major retail accounts were retained.

A precision manufacturer in the $25–75M range, combining multi-level mechanical assembly with electronics assembly.

A small company with all the structural complexity of a large one. Bills of material routinely ran beyond eight levels and carried 100 to 200 components each, with no practical control over the processes behind them. Fill rates sat in the 60s.

• Attached an AI-driven finite scheduling layer to the existing ERP rather than replacing a system that was doing the rest of its job perfectly well.
• The scheduling mathematics is the ordinary part. What the AI adds is tolerance for bad data: where a conventional scheduler hits a broken record and stops, this one resolves the problem and keeps going — which matters when the bills of material are this deep and the data was never clean to begin with.

• Engagement in progress. Fill rates started in the 60s; the target is the 90s.

Ongoing