The Changeover Bottleneck

Test the hypothesis with flow evidence, not with utilisation opinion.

• Queue behaviour. A true constraint has a growing or persistently large queue in front of it and starves the resource downstream. Here: 6.5 days of WIP before die-cutting, and gluing idle 22%. That pattern is consistent with die-cutting as the constraint. It is not proof - the queue could be caused by batch policy or by printing releasing in large lots.
• Capacity versus demand by work centre. Convert demand into required machine hours at each centre and compare with effective capacity (scheduled hours x OEE). If die-cutting is running at close to 100% of effective capacity while printing has headroom, the constraint is confirmed. This one calculation settles most arguments.
• Observe the starved-and-piled-up pattern directly. Walk the floor over two full shifts: is die-cutting ever waiting for work? Is gluing ever waiting for die-cutting output? The resource that is never starved, and always has a queue, is the constraint.

Why it matters. Every improvement dollar spent off the constraint buys nothing in throughput. Overtime on gluing, which is already idle 22% of the time, would add cost and produce no additional output - it simply shortens an idle period. Similarly, overtime on the die-cutters only helps if the constraint responds to extra hours; if the lost time is changeover and breakdown, extra hours at premium cost treat the symptom. Deciding the constraint first is what stops the plant from spending $40,000 proving that local efficiency metrics were the wrong target.

Throughput logic: the plant earns money by shipping orders, and the only improvements that matter are those that increase throughput, reduce inventory, or reduce operating expense at the constraint.

Option 1 - SMED ($40,000). Three changeovers per day at 2.5 hours consume 7.5 machine-hours of 32 scheduled. Reducing changeover to 50 minutes saves 3 x 1.67 ≈ 5 machine-hours per day, roughly 16% more available die-cutting capacity for $40,000. Highest return per dollar by a wide margin, and it attacks the true cause: 40% small orders means changeover frequency is structural, not a temporary condition.

Option 2 - die-cutting overtime. Adds capacity at a premium wage, and only after the constraint is exploited. Keep it as a short-term buffer for backlog recovery, not a permanent plan.

Option 3 - third shift or third machine. $850,000 is unavailable and a third shift adds capacity that a 16% gain may already cover. Defer until SMED results are known; the capacity gap should be re-measured before any capital request.

Option 4 - gluing overtime. Reject. Gluing is idle 22%; 22% x 32 hours x 2 shifts is a large amount of unused capacity. Adding cost there violates the subordination principle.

Recommended sequence: (1) Exploit at zero cost - enforce a no-idle rule so die-cutting is never starved (feed it first, pre-stage materials, break reliefs covered), stop unplanned stops, sequence jobs to group similar dies, and reduce the number of changeovers through scheduling before reducing their duration. (2) Run SMED. (3) Use targeted overtime only to clear the backlog the constraint created. (4) Re-measure and only then decide on capital.

Also examine the 3.8% scrap and 6% rework at die-cutting - quality losses are capacity losses at the constraint. Even a point of improvement there adds effective capacity.

The failure mode is predictable: the moment plant performance is judged on machine utilisation or labour efficiency, someone will run the die-cutters through breaks, build inventory in front of the constraint, and produce large batches that hurt lead time.

Measure at the system level:

• On-time-in-full delivery and quoted-versus-actual lead time (currently 10 quoted, 17 actual) - the outcome the customer feels.
• Die-cutting throughput in good units per scheduled hour, and die-cut queue in days (currently 6.5).
• Changeover minutes and changeover count per day, reported together so a drop in count is not mistaken for improvement if it came from bigger batches.
• WIP value and days, total and by centre.
• Constraint downtime and starvation incidents - count events where die-cutting had no work, and events where it stopped unplanned.
• Scrap and rework at die-cutting specifically.

Retire or reinterpret: individual machine utilisation above the constraint, and labour efficiency variance on non-constraint resources. Explain to the production manager why gluing running at 78% utilisation may be correct behaviour.

Cadence and governance. A daily fifteen-minute constraint review (queue, starvation, top three losses) and a weekly review of the dashboard with the plant manager. Publish the SMED benefit in throughput and lead-time terms, not in cost-per-unit terms, so the improvement survives the next budget cycle. Set the decision point in advance: after 90 days, if die-cut queue is not down and OTIF not up, the diagnosis of the constraint was wrong and we return to the capacity analysis rather than escalating spend.