Production Sequencing: A Practical Guide

Production sequencing decides the exact run order of jobs already assigned to a resource. Scheduling decides what runs this week and on which line; sequencing decides what runs next, using dispatching rules, setup similarity, and constraint priorities. On most floors, the sequence is where the schedule is won or lost.

Two plants can run the identical weekly schedule, the same jobs, same lines, same hours, and one ships everything while the other burns a day in avoidable setups and misses two due dates. The difference is sequence. This guide covers the classic dispatching rules and what each one optimizes, how sequence-dependent setups change the arithmetic, how sequencing differs at the constraint, and how to write a sequencing policy your supervisors can actually follow. It sits one level below our production scheduling primer: scheduling allocates, sequencing orders.

What is the difference between scheduling and sequencing?

Scheduling assigns work to resources and time buckets; sequencing orders the queue within each resource. The schedule says line 2 runs jobs A through F this week for roughly 60 hours. Sequencing decides that D runs first, then A, then F, and that decision changes total setup time, which due dates are met, and how long jobs wait, without changing a single assignment. Sequencing is also the decision made most often: a schedule might be rebuilt weekly, but every time a job completes, someone or something decides what runs next. That is why floors with excellent schedules and no sequencing discipline still underperform; the highest-frequency decision is the least managed. The formal machinery for making it well is the dispatching rule, covered broadly in dispatching rules.

What do the classic dispatching rules optimize?

Each rule optimizes one thing and sacrifices another; there is no universally best rule, only a best rule for what a resource is trying to protect. The four worth knowing cold:

The deeper dives on earliest due date and critical ratio cover the mechanics with worked examples. In practice most plants run a hybrid: a primary rule per resource, with a small set of explicit overrides for hot orders, and that is fine. The failure mode is not choosing the wrong rule; it is having no stated rule, so the sequence becomes whoever shouts loudest at the morning meeting.

How do sequence-dependent setups change the game?

When changeover time depends on the job pair, the sequence itself creates or destroys capacity, and rules that ignore setups leave hours on the table. Running products in an order that respects the transition structure, light colors before dark, allergen-free before allergen-containing, same tooling family back to back, can cut total changeover hours dramatically without touching a single setup procedure. The knowledge usually exists in a setter's head; the work is writing it down as a transition matrix, the from-to grid of changeover times and cleaning levels between products.

The tension is real: setup-optimal order and due-date order usually disagree, so the policy has to say how much lateness risk the plant will trade for setup savings. In food and consumer goods the matrix carries regulatory weight on top, allergen transitions demand validated cleaning, which is why changeover sequencing and our companion piece on production scheduling for CPG treat run order as a food-safety control, not just an efficiency lever. And every hour the matrix says a transition costs is an hour SMED can attack directly.

What does a transition matrix look like?

A transition matrix is a from-to grid: rows are the product just finished, columns are the product about to run, and each cell holds the changeover time and cleaning level for that pair. It is the sequencing document that turns a setter's experience into something a scheduler, a new supervisor, or a piece of software can use. Building one takes an afternoon per line with the people who do the changeovers, and it pays back immediately: the cheap diagonals and expensive corners of the grid tell you which product families to group and which transitions to schedule only at sanitation breaks.

Should you batch similar jobs or level the mix?

Batching by setup family and leveling the mix pull in opposite directions, and the right balance depends on what downstream needs. Pure setup batching runs each family in one long block, minimizing changeovers but making every product's availability lumpy: customers of the family that ran last week wait longest. Mix leveling, the heijunka idea, repeats a short cycle of every product so each is always recently made, at the price of more changeovers. The transition matrix prices that trade explicitly: if the extra changeovers cost two hours a week and the leveled mix cuts finished-goods inventory and shortage risk across the range, leveling wins; if a transition costs a two-hour validated clean, batching around it wins. Most plants land on a hybrid, a repeating family cycle with leveled products inside each family block, which is exactly the product-wheel structure common in process and packaged-goods plants.

How does sequencing differ at the constraint?

At the constraint, sequence for throughput and due-date protection; everywhere else, sequence to serve the constraint. The bottleneck's hours are the plant's hours, so its sequence deserves real optimization: due-date order first, setup grouping where it does not endanger a promise, planned by a person or system looking at the whole queue. Non-constraint resources need nothing that sophisticated, their queues are short if release is controlled, so a simple rule such as first-available or buffer priority is enough, and over-optimizing them adds complexity without throughput. This split is the sequencing half of bottleneck scheduling, and the full machinery, drum schedules, buffers, and rope-controlled release, is worked through in our drum-buffer-rope guide. A useful habit even outside TOC shops: when two sequencing goals conflict anywhere in the plant, resolve the conflict in favor of the resource that cannot recover lost time.

How do you build a working sequencing policy?

A sequencing policy is a one-page decision rule per resource, built like this:

1. Classify each resource. Constraint or non-constraint, setup-sensitive or not. The combination determines how much sequencing sophistication it deserves.
2. Write the transition matrix where setups are sequence-dependent. From-to changeover times and cleaning levels, captured from the people who do the work.
3. Pick one primary rule per resource. EDD or critical ratio where due dates dominate, SPT where queues and WIP dominate, setup-family grouping where the matrix dominates.
4. Define the overrides. Who may break the sequence, for what reasons, recorded where. Hot orders happen; invisible hot orders are what destroy trust.
5. Freeze a short window. The current job plus the next one or two stay fixed, so setups can be staged and crews are not whipsawed.
6. Measure the outputs. Setup hours per week, average lateness, and flow time per resource; review monthly and adjust the rule, not just the exceptions.

If sequence positions live in a spreadsheet nobody on the floor can see, start simpler: structure the queue in our free production schedule builder and publish it where operators actually look.

What does the research say?

The theory here is old, solid, and worth trusting:

• For a single machine, shortest processing time provably minimizes average flow time, and earliest due date provably minimizes maximum lateness, classical results summarized in the single-machine scheduling literature. These are the honest defaults when no setup structure dominates.
• The general problem is hard: multi-resource job-shop scheduling is NP-hard, which is why practical sequencing uses dispatching rules and heuristics rather than exact optimization, and why a good-enough sequence published on time beats a perfect one computed too late.
• The ASCM/APICS body of knowledge defines the standard rules (FCFS, SPT, EDD, critical ratio, slack-based variants) and treats dispatching as the execution-level layer of priority planning.

Where Harmony AI fits

Sequencing decisions are only as good as the queue data behind them: what is actually waiting at each resource, true remaining work per job, real transition times rather than the estimates from three years ago. Harmony AI is an AI-native MES, a real-time operational layer that connects machines, scheduling software, and floor paperwork into one live record with no rip-and-replace, so the queue, job status, and measured changeover times are current facts at the moment the next-job decision gets made. Harmony AI's agents apply the sequencing policy continuously, flag when an override breaks a due-date commitment, and keep the published sequence in front of operators instead of trapped in a planner's spreadsheet, part of the same connected-floor foundation described in Harmony AI's features. Deployment is white glove and in person: Harmony AI's team stands at the line with your schedulers, captures the transition knowledge that lives in setters' heads, and wires it into the system alongside what you already run. The CLS case study shows that groundwork in a real plant.