Kanban in a High-Mix, Low-Volume Line
Precision Interface assembles 420 active SKUs of industrial control modules in lot sizes of 25–200. Demand is lumpy: monthly volume is 18,000 units, but any given SKU may see zero orders for weeks and then a 600-unit spike. The company introduced a two-bin kanban for all purchased parts and a card-based kanban between SMT and final assembly. Within three months, line-side inventory doubled to $1.9M, stockouts on A-class components rose to 14 per week, and planners were overriding cards daily. The SMT line has 35-minute changeovers and 92% uptime; final assembly has 10-minute changeovers. Suppliers deliver twice weekly, with 5-day replenishment lead time for most parts. The operations director wants to abandon pull and return to MRP. The CFO wants inventory down 30% without hurting 96% on-time delivery.
1. Was the failure due to pull logic or to how pull was applied? Analyze the evidence, distinguishing where kanban is appropriate and where it is not.
Read the worked answer
Diagnosis. The failure is mainly application, not pull logic. Kanban works best with relatively stable, repetitive consumption and disciplined replenishment. Precision Interface has 420 SKUs, lumpy demand, zero-demand weeks, and 600-unit spikes. A blanket two-bin system for all purchased parts ignores that segmentation.
The symptoms fit a broken signal loop:
- Line-side inventory doubled because cards are not being removed, returned, or trusted; planners override daily, so the kanban is no longer the replenishment signal.
- A-class stockouts rose because card quantities were not sized to a 5-day supplier lead time plus twice-weekly delivery and demand spikes. The system was probably sized on average usage, not variability.
- The SMT-to-assembly loop may be more suitable because final assembly has 10-minute changeovers and SMT has 92% uptime. But a single card loop across a high-mix SMT line still needs supermarket rules and controlled batch sizes.
Abandoning pull entirely would throw away the visibility. Instead, segment: kanban for runners, min-max or MRP for irregular SKUs, and a controlled buffer or CONWIP for spike items.
2. Design a segmented replenishment model for the 420 SKUs. Specify criteria and at least one sizing calculation.
Read the worked answer
Segmentation. Use ABC by annual usage value crossed with XYZ demand variability.
- AX/runners: high value but stable. Use kanban with a supermarket and daily card audit.
- BY/irregular: medium value, variable. Use min-max with a fixed review period and planner exception review.
- CZ/rare or spike: low value or very lumpy. Make-to-order or MRP with a small safety buffer only where customer service requires it.
Sizing example. A stable A-class component has average daily usage of 20 units, supplier lead time of 5 days, twice-weekly delivery so assume 2.5 days review, and a 2-day safety allowance. Container quantity is 50. Cards = 20 × (5 + 2.5 + 2) / 50 = 3.8, round up to 4 cards, or 200 units. For a 600-unit spike, do not simply multiply every card; use a spike buffer, expedite, or make-to-order rule. Reduce the 5-day lead time through supplier collaboration and review the card count monthly. Put card ownership under operations, with an audit before any planner override.
3. The CFO asks for a 30% inventory reduction while holding 96% on-time delivery. What is a realistic plan and what would you refuse to promise?
Read the worked answer
Realistic plan. A 30% cut on $1.9M is $570k. Do not take it evenly across all SKUs. Sequence the work:
- Remove obsolete and slow-moving stock first; this is the lowest-risk reduction.
- Stabilize the card loop on A-class runners: correct card counts, stop unauthorized overrides, and audit daily.
- Reduce supplier lead time and delivery variability before cutting safety stock; twice-weekly delivery with a 5-day lead time is a major driver.
- Reduce SMT changeover time to allow smaller batches, which lowers the need for buffer stock between SMT and assembly.
- Recalculate safety stock by segment, then release inventory in stages with weekly OTD and stockout review.
What I would refuse. I would not promise 30% in one quarter, and I would not cut A-class safety stock while stockouts are already 14 per week. I would not use a blanket percentage target that rewards starving the line. A credible commitment is a staged reduction with an explicit service guardrail: if OTD falls below 96% or A-class stockouts rise, pause the next inventory release.