All postsJuly 4, 2026

Drum-Buffer-Rope: How to Schedule a CPG Line Around Its One True Bottleneck

scheduling oee cpg operations changeovers smed

Your line only goes as fast as one machine

Walk any CPG floor at 10 a.m. and it looks productive: mixers running, upstream lines humming, forklifts moving totes, WIP stacked three pallets deep. Everyone is busy. And yet output is flat, overtime is creeping, and the ship date for that private-label run is slipping again.

Here is the uncomfortable truth behind that scene: your throughput is set by a single resource. Running everything else "flat out" doesn't make more cases — it makes more WIP. If you want to schedule a line that actually ships, you stop scheduling the whole line and start scheduling the one machine that governs it.

That's the core of Drum-Buffer-Rope (DBR), the scheduling method inside the Theory of Constraints (TOC). This post is the operator's version: how to find your constraint, build the schedule around it, and protect it from the churn that quietly costs you cases every shift.

The one idea behind Theory of Constraints

TOC starts from a claim that sounds too simple to be useful: every process has essentially one governing constraint, and total throughput can only improve when that constraint improves. Time spent optimizing anything else does not lift system output (leanproduction.com).

That's the part most schedulers get wrong. Squeezing another two points of efficiency out of a mixer that already outpaces the filler produces exactly zero additional cases — it just piles inventory in front of the real limiter.

A quick vocabulary note. A bottleneck is the specific resource where work backs up because its capacity is less than the demand placed on it (en.wikipedia.org). A constraint is the broader idea: the thing limiting the whole system. In CPG plants, constraints come in three flavors you'll actually recognize:

Equipment — the filler, oven, retort, pasteurizer, or labeler that simply can't go faster.
Policy / "paradigm" — a rule like "never idle the filler" or "run every SKU to a full pallet" that limits output more than any machine does (isixsigma.com).
Market — when your line can out-produce demand, the constraint isn't on the floor at all; it's the order book (leanproduction.com).

That middle category matters. On a lot of lines the true limiter is a policy, not a machine — and you can fix a policy this week.

Find your drum: identifying the constraint

Before you can schedule around the constraint, you have to name it. The symptoms are consistent:

WIP consistently piles up in front of one asset. Inventory accumulates where the flow chokes. That pallet stack is a map (en.wikipedia.org).
It's the one that's "always running." Non-constraints have idle time; the constraint rarely does.
It dictates your overtime. When you need more output, this is the resource you extend hours on.

This is also where your OEE data earns its keep. A point of lost OEE on a non-constraint costs you nothing at the plant level — there's slack downstream to absorb it. A point of lost OEE on the constraint costs you cases you can never recover. So when you read your OEE dashboard, weight the constraint's losses far more heavily than everyone else's. (If you're building that discipline, our OEE-grounded changeover math shows how the same lost minute is worth wildly different amounts depending on where it happens.)

Drum-Buffer-Rope, explained for the floor

Once you know your constraint, DBR gives you three moving parts.

Drum

The drum is the constraint, and the pace it runs at sets the beat for the entire plant. Its schedule determines total throughput — which means, in practice, the constraint's schedule is your real master production schedule (leanproduction.com). Academic studies of TOC scheduling confirm this is exactly what master schedulers do in practice: they build the schedule around the single critical resource and let everything else follow (sciencedirect.com).

Buffer

A buffer is a time buffer — usually measured in hours — of WIP that arrives ahead of the constraint so an upstream hiccup never starves it. There are two you care about:

A constraint buffer protects the drum from running dry.
A customer / shipping buffer protects your promised ship dates.

Sizing is not a gut call: more process variation upstream means you need a larger buffer. If you'd rather not carry big buffers, the alternative is sprint capacity — deliberate overcapacity at non-constraint resources so they can catch up quickly after a stumble (leanproduction.com).

Rope

The rope is a signal. When the constraint consumes material, that consumption triggers an identical release of new material at the front of the line. It's a pull signal, not a push: it keeps the drum fed without flooding the floor with WIP (leanproduction.com).

A quick worked example

Say your filler runs at 100 cases/hour and it's the drum. Downstream, your labeler and packer can each do 140/hour; upstream, your mixer can supply well over 100/hour but occasionally goes down for 30–40 minutes.

Drum: you schedule the filler to 100/hour and build the day's MPS from its clock, not from the 140/hour packer's theoretical rate.
Buffer: you hold, say, a 1-hour time buffer of mixed product in front of the filler, so a 40-minute mixer stumble never starves it.
Rope: every time the filler pulls a batch, the mixer gets the signal to make one more — no more, no less. WIP stays bounded; the filler never waits.

The packer and labeler? They're free to sit idle sometimes. That idle time isn't waste — it's proof they're subordinated to the drum instead of manufacturing inventory nobody ordered.

The Five Focusing Steps as a scheduling routine

TOC packages the work into five steps. Read them as a repeatable operating loop, not a one-time project (leanproduction.com):

Identify the constraint. Use the WIP-and-overtime symptoms above.
Exploit it — get more out of the constraint with the resources you already have. On a CPG line that means: stop starving or blocking the filler, move minor changeovers and cleaning tasks off the drum wherever possible, and don't run trial or QA samples through it during production.
Subordinate everything else to the constraint. Upstream lines pace to the drum, not to their own local efficiency numbers. This is the step that feels wrong to a shift lead measured on their own asset's utilization — and it's the step that makes DBR work.
Elevate the constraint. If it's still limiting after you've fully exploited it, now add capacity — a second shift on the drum, a parallel machine, faster tooling.
Repeat — but watch out for inertia. Once you elevate, the constraint may move somewhere else, and the old rules you built around the old drum can quietly become the new constraint (leanproduction.com).

One practical scheduling payoff: products that never touch the constraint — TOC literature calls them "free products" — can be scheduled flexibly, because they don't compete for the drum's time (sciencedirect.com). That's real freedom in your sequencing you probably aren't using.

Protecting the drum: changeovers, maintenance, and quality holds

Here's where DBR connects to changeover discipline. A sequence-dependent changeover is expensive everywhere — but it's catastrophic at the constraint, because every minute the drum spends changing over is a minute of plant-wide throughput gone. So the rule isn't "minimize changeovers everywhere equally." It's minimize and sequence-optimize changeovers on the drum specifically, and be relatively relaxed about setups on non-constraints that have slack anyway.

That's a direct application of your SMED work: the fastest changeover payback in the whole plant is the one on the constraint. If you've read our take on the hidden cost of changeovers, apply that math to the drum first.

The broader bottleneck-protection playbook is well established (en.wikipedia.org):

Keep the constraint well-maintained — an unplanned drum breakdown is the most expensive downtime you have.
Hold a constant buffer stock upstream of it so it's never starved.
Strip non-value activity off it — inspections, staging, minor tasks that could happen elsewhere.
Only add capacity after you've genuinely exhausted the exploit and subordinate steps.

Where DBR meets EPEI and Heijunka

DBR doesn't replace the leveling and cadence methods you may already run — it sits above them and answers a different question.

DBR decides what paces the plant — the drum sets the beat.
EPEI decides how often each SKU comes around on that beat.
Heijunka decides how you level the mix so demand variability doesn't whip the drum around.

Think of it as a stack, not a competition. DBR keeps the constraint fed and protected; EPEI turns that protected capacity into a repeatable wheel; Heijunka smooths the mix so the wheel stays stable. And all three depend on demand signals that survive contact with the floor — see forecasting that survives the floor.

Common failure modes

Chasing local efficiencies off the constraint. Running a non-constraint at 100% only builds WIP. Utilization is not the goal; throughput is.
Buffers sized by gut. Too small and the drum starves; too large and you're back to WIP piles. Size to actual upstream variation.
Letting policy inertia become the new constraint. After you elevate a machine, revisit the rules built around the old drum.
Ignoring the market constraint. If the line already out-produces demand, more scheduling cleverness on the floor won't help — the constraint moved to the order book, and that's where the work is (leanproduction.com). This is also where relentless overproduction turns into the stockout-and-glut firefighting cycle.

Put it into practice this week

You don't need a transformation program to start. You need four moves:

Locate the drum. Find where WIP piles up and which asset dictates overtime.
Set a time buffer. Pick an hours-based constraint buffer sized to your worst realistic upstream interruption.
Define the rope. Write one release rule: material enters the line only when the drum consumes it.
Protect the drum's schedule. Freeze it against churn, and move minor changeovers and non-value tasks off it.

The hard part isn't understanding DBR — it's holding the discipline every shift: monitoring the buffer, honoring the release signal, and refusing to let a busy non-constraint pull the schedule off the drum. That constant buffer-watching and release-signaling is exactly the kind of continuous, rules-based work an agentic scheduler is built to operationalize — keeping the drum fed and the rope taut without a person babysitting the board.

Key takeaways

Throughput is governed by one constraint; optimizing anything else just makes WIP.
The drum (constraint) sets the beat and effectively is your master production schedule.
A time buffer protects the drum from starvation; the rope pulls new material in at the rate the drum consumes it.
Run the Five Focusing Steps as a loop: Identify, Exploit, Subordinate, Elevate, Repeat.
Attack changeovers and maintenance on the drum first — that's where minutes convert to cases.
DBR stacks with EPEI and Heijunka; watch for the market constraint when you out-produce demand.