The Illusion of Local Efficiencies (and Breaking Bottlenecks)

The robot cell runs hot. Heat-treat queues for days. Assembly waits on both. Customers do not care about robot uptime—they care whether the order ships. The Goal (Goldratt & Cox, 1984) names this the illusion of local efficiencies: optimizing each step in isolation damages the system.

This is Lesson 3 in our series. You already know the three measurements and the flow engine mental model. Now we find the constraint and manage it.

#Find Herbie on the floor

On Alex Rogo's Boy Scout hike, slow scout Herbie set the troop's pace. In the plant, bottlenecks are resources whose capacity is less than or equal to demand placed on them. Find them by WIP piles—inventory accumulates directly upstream of the constraint.

You do not need a capacity study to start. Walk the line. Where does work wait longest? Where do supervisors expedite? Where does material sit in racks while downstream operations starve? That pile is pointing at your Herbie.

Alex's plant had two constraints at different stages—a machining center and heat-treat furnaces. Most real plants have one dominant constraint at a time, but the lesson holds: name it on the whiteboard so every shift argues about the same resource.

Tip. Photograph WIP piles at the same time each day for one week. If the pile grows at the same operation every day, you have found your constraint candidate.

#An hour at the bottleneck is an hour for the whole plant

Because the constraint sets total output, downtime at the bottleneck costs roughly total operating expense divided by constraint hours available—not the machine's hourly rate. A lunch break at the constraint is not thirty-two dollars of labor; it is an hour of lost throughput for the entire plant.

That math changes how you schedule maintenance, breaks, and QC. Non-bottleneck downtime is expensive as operating expense, but it rarely caps ship rate. Constraint downtime always does.

Protect constraint time like flight hours on an aircraft—every idle minute at the bottleneck taxes the whole system.

#Exploit before you buy

Exploit means never waste constraint capacity:

• Run through breaks when safe and staffed
• Inspect before the bottleneck so bad parts never consume constraint time
• Prioritize orders that protect due dates at the constraint
• Eliminate setup and changeover waste at the constraint before anywhere else

Exploit is step two of the five focusing steps. Most plants jump to "buy a new machine" (elevate) while the constraint still sits idle two hours a day for lunch, changeover, and waiting on material.

#Subordinate the rest

Non-bottlenecks must pace to the constraint—not run flat out. Priority tags helped Alex's team prioritize bottleneck parts, but release timing mattered more: withhold raw material until the bottleneck needs it. Flooding the floor to "keep people busy" creates WIP that chokes the constraint.

Subordinate is the step most plants skip because it feels wrong. Idle time at a non-bottleneck looks like waste. But idle beats a WIP flood every time—because WIP hides the constraint, extends lead time, and traps cash.

Tip. Ask your planners one question: Are we releasing to keep upstream machines busy, or to feed the constraint at its consumption rate? The honest answer tells you whether subordinate is happening.

#Elevate without capital first

Before new equipment, Alex elevated by:

• Pulling retired machines from scrap to offload constraint work
• Sending select parts to outside heat-treat
• Changing engineering rules to bypass heat-treat where brittleness was policy, not physics

Elevate is step four—after identify, exploit, and subordinate are honest. Vendor capacity at an outside processor is elevate, not admission of failure. Compare at system margin and lead time, not unit cost at one operation.

#Constraint elevation checklist

Before you approve capital, outsourcing, or engineering changes, confirm exploit and subordinate are honest:

• Constraint downtime logged with reasons for two weeks
• QC moved upstream of constraint where applicable
• Release rate matches consumption rate for pilot family
• Overtime at non-constraints stopped while constraint waits
• WIP trend at constraint flat or falling

If any box is unchecked, elevate will rebuild the swamp faster than it drains.

#Constraint elevation checklist (continued)

Document your elevate decision in one sentence: We are adding capacity at [resource] because [exploit action] recovered [X] hours and [subordinate action] held WIP flat for [Y] weeks, but demand still exceeds constraint output by [Z].

That sentence protects the investment from "why did we buy this?" six months later—and proves you ran the loop, not just the capital request.

#When the constraint moves

Break one bottleneck and another appears—or demand becomes the limit. Step five of the focusing process warns against inertia: policies built for yesterday's constraint become tomorrow's choke point.

Red-tag priority systems, finished-goods buffers, and batch-size rules that made sense for the old constraint often survive long after the constraint moves. Audit those policies every time throughput jumps.

#Red/green tag lesson

Priority tags helped Alex until they starved assembly sequences. Sequence discipline—first come, first done at the constraint—sometimes beats constant expediting. Expediting feels responsive; it often rearranges WIP without increasing throughput.

#Setup time mirage

Saving setups at non-bottlenecks rarely saves system money; that hour was not convertible to throughput. Challenge setup reduction projects upstream of the constraint. Setup savings at the constraint are gold; setup savings elsewhere are usually mirage time.

#Engineering policy as bottleneck creator

Process rules that require extra operations for historical efficiency targets can create artificial constraints. Question whether a routing rule exists for physics or for a cost-accounting target that no longer serves the plant.

#Communicate with sales

When due-date reliability jumps after subordination, sales can sell reliability—a throughput-world sales weapon. Operations and commercial teams often fight because they use different scorecards. Shared three-measurement vocabulary ends some of those fights.

#The Boy Scout hike as management training

Alex's Boy Scout hike is more than a metaphor—it is a training exercise you can replicate with your leads. Send a group on a timed walk with a mixed-pace team. Watch the line spread. Identify the slowest member. Try reordering (putting the slowest member first). Watch how the constraint sets the pace for everyone behind.

The hike teaches three lessons that transfer directly to the plant:

1. The constraint sets system output — not the fastest member, not the average pace
2. Local speed does not help — running faster when blocked does not increase group arrival time
3. Load matters — removing weight from the constraint (offloading work) can elevate capacity without adding people

Use the hike story in onboarding. New supervisors who understand Herbie understand why idle time at non-bottlenecks is acceptable.

#Sales, planning, and the constraint

Operations is not the only function that creates local efficiency traps:

• Sales promises short lead times based on the fastest operation's capacity, not the constraint's
• Planning releases work to hit monthly production targets, not constraint consumption rate
• Engineering adds routing steps that protect quality locally but consume constraint time systemically

Each function optimizes its own scorecard. TOC gives them one shared scorecard: throughput, inventory, operating expense. When sales, planning, and engineering join the constraint conversation, subordinate becomes possible.

Tip. Invite sales to one gemba walk per quarter. Show them the WIP pile that determines their promise date. Promise-date reliability improves when commercial teams see the constraint.

#Daily constraint check

Run this rhythm until it becomes habit:

• Morning: Is the constraint running? What is WIP at the constraint?
• Afternoon: Did any release violate the rope?
• Evening: One exploit action planned for tomorrow

#Maintenance on the constraint

Schedule preventive maintenance on bottlenecks with the same rigor as production runs. Defer PM on the constraint only when you account for equal throughput risk. PM on non-bottlenecks during constraint idle time is often the cheapest window.

#What to do this week

Name one constraint on a whiteboard. Protect one constraint hour daily for two weeks—no idle lunch, no waiting on material, QC before not after. Measure throughput, inventory, and expense—not utilization alone.

Next we explain why balancing every workstation to demand guarantees failure: Why Balance the Plant Backfires.

← Start the series · ← Flow Engine · Next: Balanced Plant →