Lean for high-mix, low-volume (HMLV) is the practice of applying flow, cells, and leveling to plants that make hundreds of part numbers in small lots rather than a few products in huge runs. The principles are unchanged; the tools that dominate are setup reduction, product-family cells, and leveling by interval instead of by rate.
Most lean literature is written around the mass-production case: one product, a stable takt, long runs, a moving line. If you make 400 part numbers and no two weeks look alike, that picture does not fit, and copying it directly leads to frustration, you cannot balance a line to a takt that changes every order. HMLV lean starts from a different question: not "how do we run this one product faster," but "how do we make switching between many products cheap enough that small lots flow." This is the HMLV version of the general lean playbook.
What is lean for high-mix, low-volume?
Lean for high-mix, low-volume means designing production so a wide variety of parts can flow in small quantities without the plant grinding to a halt at every changeover. The customer value and waste definitions are identical to any lean effort. What changes is where the waste lives: in HMLV, the dominant wastes are the setups between products, the work-in-process piling up in front of shared machines, and the time operators spend hunting for the right program, fixture, or material for the next job.
The strategic move is to stop treating variety as one impossible problem and start treating it as a set of repeatable patterns. You almost never have 400 truly unique routings; you have a handful of routing families with 400 variations. Find the families, build flow around them, and make the switch between variants cheap. That is the whole game.
Why doesn't classic mass-production lean fit HMLV?
Because the assumptions underneath a moving assembly line break when volume is low and mix is high. A mass line balances every station to a fixed takt time, runs the same product for hours, and amortizes a long setup over a huge lot. HMLV has none of those luxuries.
Takt time is not fixed, because demand for any single part number is lumpy and small, you might build a part twice this month and not again for six weeks. Runs are short, so a two-hour changeover that a mass plant hides inside a two-day run becomes catastrophic: set up two hours to run twenty minutes. And you cannot dedicate a line to a product that ships in tens, so machines are shared, which creates queues. The tools that matter therefore shift. Line balancing to a single takt matters less; quick changeover flexible cells, and controlling work-in-process matter far more.
How do you group products into families?
You group by how parts are made, not by what they are called. The tool is a product-quantity-routing analysis: list your part numbers, their volumes, and the sequence of operations each one travels, then sort until clusters appear. Parts that share a routing, cut, mill, deburr, belong in one family even if they serve completely different customers. Most plants that do this for the first time find that a large share of their volume runs through two or three families.
Once the families are visible, you can build a cell for each: the machines that family needs, arranged in process sequence, sized for the family's combined volume. A cell turns a tangle of trips across a functional layout, WIP hauled from the mill department to the drill department and back, into a short, contained flow. Value-stream mapping one family end to end, using value stream mapping tells you where the queues and the setups are before you move a single machine.
HMLV is not a niche. The typical U.S. metalworking shop already lives here: most machine shops are small, low-volume operations building parts on a job or order basis, not mass producers.
| HMLV reality (U.S. machine shops, NAICS 332710) | Figure | Primary source |
|---|---|---|
| Estimated employment in machine shops, a classic high-mix, low-volume sector | ~226,000 | U.S. Bureau of Labor Statistics (OEWS) |
| Machine-shop work characterized by the Census Bureau as low volume, on a job or order basis | By definition | U.S. Census Bureau (NAICS 332710) |
How does leveling work when demand is lumpy?
You level by interval, not by rate. Mass-production heijunka smooths a steady demand into an even rate, so many units per hour, every hour. HMLV cannot do that, because any one part's demand is too small and too spiky. Instead, HMLV plants level by committing to an interval: every part, every so often. The idea, often called every-part-every-interval or EPEI, is to fix a repeating cycle, say, every family runs at least once a week, and hold to it.
The interval is powered entirely by setup time. If a changeover takes two hours, you can only afford a few per week, so the interval is long and lots are big. Cut that changeover to twenty minutes and you can run each part more often in smaller lots, shortening the interval and the lead time together. This is why setup reduction is the master lever in HMLV: it does not just save the setup time, it shrinks lot sizes, inventory, and lead time all at once. A shorter EPEI is the direct payoff of every minute you take out of changeover.
Why is quick changeover the linchpin?
Because in HMLV the setup is a bigger share of everything. When runs are short, changeover time is not overhead you amortize away, it is a large, direct chunk of your available capacity, and it is the ceiling on how small your lots can be. Halving setup time in a mass plant is nice; halving it in an HMLV plant changes what the plant is capable of, because it unlocks smaller lots, shorter intervals, and less inventory in one move.
The method is SMED: separate the changeover tasks into internal (must happen with the machine stopped) and external (can be prepared while it still runs), push everything possible to external, then simplify what is left. In a machine shop that means the next job's program loaded and proven, fixtures staged, tools pre-set in holders, and material at the machine before the current job finishes, so the changeover is a swap, not a scavenger hunt. Capturing the proven setup as standard work keeps it from decaying into tribal knowledge that only one setter remembers.
How do you standardize work across so much variety?
You standardize the pattern, not every part. It feels impossible to write standard work for 400 part numbers, and it is, so you do not. You write standard work for the family's process: the setup routine for the cell, the inspection method, the way material is presented, the sequence of operations. The part-specific details, dimensions, the program, the fixture, become parameters that plug into a stable pattern. Ninety percent of what an operator does is the same across a family; only the last ten percent changes per part.
This is also where flexibility in people pays off. HMLV plants live or die on cross-training, because a cell that can only run when one specific person is present is not a cell, it is a bottleneck with a pulse. A skills matrix that shows who can run which family, and a plan to widen it, is as much a lean tool in HMLV as any changeover cart.
What is an HMLV lean roadmap?
High-mix plants get the most from a sequence that finds the families first and attacks setup before layout. Work this order.
- Run a product-quantity-routing analysis. List part numbers, volumes, and operation sequences, then sort into families. This is the map everything else depends on.
- Pick the highest-volume family and map it. Value-stream one family end to end to expose the queues, the shared-machine bottleneck, and the worst setups.
- Attack the dominant changeover with SMED. Convert internal setup to external, stage tooling and programs ahead, and standardize the setup. This is the move that unlocks smaller lots.
- Form a cell for the family. Arrange the family's machines in process sequence so parts flow a short loop instead of touring the plant.
- Set an every-part-every-interval target. Commit to running each part in the family on a fixed, repeating interval, and use the shorter setups to shrink that interval over time.
- Cross-train and standardize the pattern. Write standard work for the family's process, widen the skills matrix so any shift can run the cell, and bank the gains before moving to the next family.
This roadmap shares DNA with the job-shop version of lean but the two differ in one way: HMLV parts repeat, so intervals and cells pay off, while a pure job shop builds one-offs that may never return.
Where does real-time data fit in HMLV?
HMLV is where paper-based tracking hurts most, because there is so much to track. With hundreds of part numbers moving through shared machines in small lots, the questions that matter, where is this job, how long did that setup really take, which family is eating our capacity, are nearly impossible to answer from paper travelers and end-of-shift tallies. Setup times, the one number that governs everything in HMLV, are exactly what paper logs estimate worst.
Digitizing capture at the machine, so real setup and run times land in the system as they happen, turns the family analysis and the EPEI targets from guesses into measured facts, without ripping out the scheduling or ERP systems a shop already runs. When CLS moved production logging off paper, supervisors could finally see problems during the shift they happened in. See how the connective layer fits on our features overview and start by finding your families.