What it means
Traditional quality control uses a pass or fail test: a part either sits within its tolerance range or it does not. Taguchi argued that this misses something, because a part that just barely passes is almost as troublesome as one that just fails.
His view is that loss grows steadily as a product moves away from its ideal target. That idea is captured in the loss function, a simple formula that links the size of the deviation to a monetary cost.
The cost grows with the square of the deviation, so being twice as far from the target costs four times as much. This turns a vague quality aim into a number that finance can use.
The second part of the method is designing a product to be insensitive to noise, meaning factors that cannot be easily controlled, such as temperature, wear or operator differences. Engineers run structured experiments to find the settings at which the output stays on target despite that noise.
It is cheaper to build in stability than to inspect and scrap afterwards. For a finance professional, the method is useful because it connects quality to cost of poor quality.
It shows that reducing variation can save money even when scrap rates look acceptable. It also helps justify spending on process improvement, which otherwise looks like a cost with no clear return.
The main nuance is that the loss function is an estimate. The constant in the formula comes from a judgement about what a unit of deviation really costs customers and the company, so it needs to be set carefully.
In practice
Real-world examples.
Example
An electronics manufacturer finds that circuit boards inside tolerance still fail more often in customers' hands as the measurements approach the limit. Using a loss function, it assigns a cost to each small deviation and justifies a new calibration machine costing $150,000. Warranty claims fall in the following year.
Example
A food company fills jars with a target weight of 500 grams. Overfilling wastes product and underfilling risks complaints, so it measures the cost of each gram of deviation. Tightening the filling process saves more in product than it costs in equipment.
Example
A car parts supplier runs experiments to find the machine settings that keep a component's size steady across summer and winter temperatures. The better settings remove the need for a costly climate-controlled workshop. The saving is recorded as a quality improvement.
Formula
Calculation
Loss per unit = k x (actual value - target value)^2, where k = cost at the tolerance limit / (tolerance)^2
A factory makes a shaft with a target diameter of 10.00 mm and a tolerance of plus or minus 0.50 mm. A shaft at the limit would need rework costing $20, so k = 20 / (0.50)^2 = 20 / 0.25 = 80. A shaft measuring 10.30 mm deviates by 0.30 mm, so its loss = 80 x (0.30)^2 = 80 x 0.09 = $7.20. If the plant produces 50,000 such shafts a year with that average deviation, the hidden cost is 50,000 x 7.20 = $360,000.Case study
Seen in the real world.
Brightwater Components is an illustrative, fictional maker of pump seals. Its seals passed inspection 98% of the time, yet customers kept returning units that were technically inside tolerance but wore out early.
The quality manager applied a loss function with a cost of $30 at the tolerance limit and found that the average seal was carrying a hidden loss of about $4 per unit across 300,000 units, or $1,200,000 a year. A process redesign using experimental settings cut the average deviation by half.
Because loss grows with the square of deviation, halving the deviation cut the average loss to a quarter, saving roughly $900,000 a year. The illustrative lesson was that passing inspection is a poor measure of quality when the cost curve is smooth.
Watch out
Common mistakes.
- Believing that any product inside the tolerance limits is equally good, when the Taguchi view is that loss rises steadily as the product drifts from target.
- Treating the loss constant as a precise fact, when it is an estimate that depends on realistic costs of rework, warranty and lost customers.
- Relying on end-of-line inspection alone, when the method aims to make the process stable in the first place.
Questions
People also ask.
Who was Genichi Taguchi?
He was a Japanese engineer and statistician whose methods for improving quality through experimental design became widely used in manufacturing.
Why does the loss function use a squared term?
Squaring makes small deviations cost very little and large deviations cost a great deal, which matches how customers experience poor quality.
Does the method only apply to factories?
It started in manufacturing, but the idea of designing a process to tolerate variation also applies to services and software.
From the founder's library
Take it further with the book.
Build your financial confidence beyond this definition. Shihan's full-length guide, Accounting Fundamentals, takes the same plain-English approach and turns it into a complete, practical playbook for non-finance managers, business owners and students - with chapter-end quiz answers and presentation slides included.
25% off with code MMHQ25, applied at checkout. Priced in USD - checkout may show the equivalent in your local currency.
View the book and save 25%