Rework is hidden capacity. Here's how to measure it.

Every shop I have walked through has rework. Almost none of them can tell you how much. It shows up as a part that goes back to the weld booth, a rack that runs through the finishing line a second time, an operator who spends Tuesday afternoon fixing Monday's run. Nobody logs it, so nobody prices it, so it never makes the list of things to fix.

My position is simple. Rework is capacity you already paid for. Cut it and two things come back: the labor you spent doing work twice, and, when that rework was passing through your bottleneck, the output you were not shipping. This post explains why both are real, when each one applies, and how to measure rework in your own shop with a one-page log and the math behind our ROI calculator.

Rework is a waste with a name

The Lean Enterprise Institute lists correction as one of Taiichi Ohno's seven wastes and defines it as inspection, rework, and scrap.^[1] Art Smalley, in LEI's series on the seven wastes, calls it the waste of defects: scrap, rework, or any nonconformance, none of it something the customer wanted.^[2]

Smalley also makes the point this whole post rests on. He describes plants that cannot keep up with their schedules, where part of the constraint problem is quality.^[2] Something coming from a supplier, an internal process, or final assembly is not right, and that is what stops the plant from hitting its numbers.

That matches what I saw in 15 years of stainless steel manufacturing and 13 years of fixing late orders by walking the floor. The pattern was never that people did not care about quality. It was that nobody could see the rework in one place. The data was scattered, and nobody was looking at the same numbers.

Two frameworks that explain why rework costs more than it looks

Cost of quality: four buckets

Quality professionals have sorted quality costs into four categories since Armand Feigenbaum first laid the concept out in detail: prevention, appraisal, internal failure, and external failure.^[3] Prevention is what you spend to keep defects from happening, such as training and mistake-proofing. Appraisal is inspection and testing. Internal failure is what a defect costs when you catch it in-house, including scrap, rework, and unscheduled downtime. External failure is what it costs after the customer has the product, including returns and warranty charges.^[3]

Here is how those buckets look on a shop floor:

Rework sits in internal failure. That bucket is where most small shops have the most to recover and the least data. The same paper that lays out the categories notes that most companies report the obvious costs, such as scrap, returns, and rework, and miss the hidden ones.^[3] It also cites Juran's estimate that unplanned quality-related costs could run as high as 20% of sales.^[3] I cannot verify that number for your shop, and neither can you until you measure it. That is the point.

The Juran Institute's working definition of cost of poor quality is the one I use: the difference between the costs that would disappear if everything were done right the first time and the actual costs that come from errors and inefficiencies.^[4] That gap is what the rework log measures.

Theory of Constraints: where the hour was lost matters

Eliyahu Goldratt's Theory of Constraints was introduced in his book The Goal. Its core idea is that every system has a constraint that limits its output, and that focusing improvement on that constraint is normally the fastest way to improve profit.^[5] The book states the rule that matters for rework in two lines, as taught in MIT's operations course built around it: an hour lost at a bottleneck is an hour lost for the entire system, and an hour saved at a non-bottleneck is a mirage.^[6]

Apply that to rework. A part that goes through your constraint twice has used two constraint hours to produce one unit of output. If the constraint is loaded and you have orders waiting, you shipped less that day. The rework did not just cost labor. It cost throughput, which The Goal defines as the rate at which the system generates money through sales.^[6]

The Theory of Constraints Institute puts the same idea in operating terms. The output of the constraint governs the output of the organization as a whole, and the constraint must never be starved for inputs or fed poor-quality materials.^[7] On its list of ways to get more out of a constraint without buying anything: install quality checks immediately before it.^[7] Catching a defect before the bottleneck instead of after it is the difference between a labor cost and a lost sale.

If you are not sure which station is your constraint, read Find your bottleneck from WIP data, not gut feel first. Everything below assumes you know where it is.

What this looks like on a real floor

The MEP National Network publishes case write-ups, and several of them are this exact story.

TMI Systems, a casework maker in North Dakota, had a final assembly line that had become the bottleneck. Upstream processes were leaving excess glue and black marks on panels, and final assembly operators were spending time cleaning them off before they could do their own work.^[8] The company's first instinct was to expand the assembly area. Impact Dakota, the state MEP center, instead identified the constraint and put quality at the source into the upstream steps. The company reports a 10% improvement from layout and lean tools and says quality at the source is now used in the factory and the office to improve flow and reduce rework.^[8] Read that again: the bottleneck was not short of capacity. It was spending its capacity on other people's defects.

Metal Finishing Technologies, a Connecticut plater working with CONNSTEP, a NIST MEP center, had a zinc plating line running at a 15 percent rework rate and estimated the line was costing nearly $80,000 a year in revenue through rework, scrap, and waste.^[9] The team set a goal to cut rework in half and identified 17 improvements, 13 of which could be done in under a month.^[9]

Amphenol Aerospace in Arizona took the appraisal route. Its assembly operations manager put a screening gate at the marking station, wrote work instructions for it, and trained operators to reject defects to the same standard used upstream. He tracked every defect caught for 60 days by defect code and originating facility and documented $60,000 in savings.^[10]

Electronic Design and Manufacturing in Virginia reports that standardized work from a GENEDGE line-balance event reduced defects and rework rates while output rose.^[11]

None of these are my shops, and the numbers are the companies' own reports to their MEP centers. But the pattern is the same one I saw everywhere: good people, scattered data, and a bottleneck quietly absorbing rework.

The math: how the ROI calculator prices rework

The ROI calculator on our home page prices rework with two lines. Here they are in full.

Rework labor saved = monthly direct labor × rework (% of labor hours) × reduction %

Added throughput = monthly shipments × rework % × reduction %

Four inputs:

• Monthly direct labor. Your loaded direct labor cost for a month, in dollars. Payroll will have it.
• Rework % of labor hours. The share of direct labor hours that went to doing something a second time. This is the number almost nobody has. The log in the callout below gets it.
• Monthly shipments. Dollar value shipped in a typical month.
• Reduction %. How much of the rework you expect to remove. Be conservative. Removing the top cause on your Pareto chart is usually worth more than you think and less than the whole thing.

Worked example

The calculator's default inputs make the arithmetic easy to follow.

Notice the shape of it. Rework in that shop is about $24,000 a month in labor. Cutting one fifth of it returns $4,800 in labor, which is real but modest. The throughput line is more than three times larger. That is the capacity argument in one table, and it is also why you need to be honest about when the second line counts.

When the labor line applies

Almost always. If someone spent an hour redoing work, that hour cost you money whether or not the station was busy. The only caveat is that the saving is real cash only if you redeploy the hour or stop paying overtime for it. If the operator simply stands at the same station with less to do, the labor line is capacity, not cash, until you fill it.

When the throughput line applies

Only when three things are true.

1. The rework consumed bottleneck time. Either the defect was made at the constraint, or it was made upstream and sent through the constraint a second time, or the constraint spent time cleaning up upstream defects, as TMI's assembly line did.^[8] Rework in a downstream cell with slack capacity, like a deburr bench with idle time, returns labor and nothing else. An hour saved there is Goldratt's mirage.^[6]
2. You have demand to fill the recovered hours. If the constraint is already waiting on orders, freed time becomes idle time. Check your backlog and your past-due list before you count it.
3. You treat the result as revenue, not profit. The calculator multiplies shipments, so the throughput line is gross shipped dollars. What you keep is that number less the material and variable cost of making the extra output. For most job shops that is still the biggest number on the page, but it is not the same as cash in the bank.

If you are not sure whether your rework touches the bottleneck, set the throughput line aside and price labor only. A smaller number you trust beats a larger one you cannot defend in a budget meeting.

How to measure rework % in two weeks

You do not need a system to get this number. You need a sheet at every cell and a rule that nobody gets in trouble for writing on it. Two weeks of honest data is enough to start.

What to do with the number

The log gives you a Pareto chart. The frameworks tell you what to do with it.

Move the catch point upstream. Smalley's point is that a defect gets more expensive the farther it travels.^[2] The TOC Institute's advice to put a quality check right before the constraint is the same idea aimed at the one station where it matters most.^[7] A first-piece check at the cell that feeds your bottleneck is appraisal spending that protects throughput. It is usually the cheapest capacity you will ever buy.

Spend on prevention where the hours are. If your top cause is a dimension that drifts after a tool change, the fix is a setup check, not more inspection. If it is an operator-to-operator difference, it is training. Amphenol's fix was work instructions and a standard everyone inspected to.^[10] Prevention and appraisal are the planned costs; failure costs are the unplanned ones.^[3] You are trading a small planned cost for a larger unplanned one.

Re-measure. Keep the log running for another two weeks after the fix. If rework hours at the constraint dropped and shipments rose, you have your proof, and you can put the actual reduction percent into the calculator instead of a guess.

Make it visible every day. The two turnarounds I can point to had no new system behind them. A steel products maker had over $700,000 tied up in late orders, and six months later it was under $50,000 because everyone looked at the same numbers every day. A textiles operation was three months behind and two months later was shipping on time and making almost twice as much product. A rework number nobody sees is a rework number nobody owns.

When the paper log outgrows itself, this is the kind of record an MES is built to keep: nonconformances, rework, and scrap tied to the job and the cell, so the Pareto chart updates itself instead of waiting for someone to retype the sheets.

The short version

Rework is internal failure cost, and it is the bucket with the least data in most shops.^[3] An hour of it at your bottleneck is an hour of output the whole system lost.^[6] Log it for two weeks by cell, part, cause, hours, and catch point. Divide rework hours by direct labor hours to get your rework percent. Price the labor line always. Price the throughput line only when the rework touched the constraint and you have orders to fill. Then fix the top cause, measure again, and put the real number in the calculator.