Ask a finance team what quality costs, and they will usually point at the QA department’s budget: salaries, tools, test environments. That number is visible and easy to cut. What it leaves out is the money spent because quality was not good enough: developers fixing bugs, support teams handling complaints, refunds to customers, emergency weekend releases. Those costs are scattered across many budgets and rarely labelled as quality. The cost of quality model exists to bring them into one view.
The idea comes from manufacturing. Armand Feigenbaum described the prevention, appraisal and failure categories in the 1950s, Joseph Juran wrote extensively on the cost of poor quality, and Philip Crosby’s 1979 book Quality Is Free argued that money spent on prevention more than pays for itself. The framework transfers to software with very little adjustment.
The four buckets
Total cost of quality is split into the cost of good quality (what you spend to get it right) and the cost of poor quality (what you spend because you did not).
- Prevention costs stop defects from being created: requirement and design reviews, training, coding standards, threat modelling, test-driven development, building reusable test infrastructure.
- Appraisal costs find defects that already exist: test design and execution, code reviews, static analysis, test environments, audits, performance and security testing.
- Internal failure costs come from defects found before release: fixing, retesting, regression cycles repeated because of late fixes, delays to the release.
- External failure costs come from defects found after release: incident response, hotfixes, support calls, refunds and penalties, SLA credits, lost customers, regulatory consequences and reputational damage.
Prevention and appraisal are the cost of good quality. Internal and external failure are the cost of poor quality. Some writers group these slightly differently, but the four-bucket split is the standard one.
A worked quarter
Consider a payments product team estimating one quarter’s costs, in lakh rupees.
| Category | Items | Before (₹ lakh) | After a year of change (₹ lakh) |
|---|---|---|---|
| Prevention | Reviews, training, framework work | 15 | 25 |
| Appraisal | Testing, environments, audits | 45 | 40 |
| Internal failure | Fixing and retesting pre-release bugs | 33 | 28 |
| External failure | Incidents, refunds, support load | 40 | 22 |
| Total | 133 | 115 |
In the first column, failure costs (33 plus 40, or 73 lakh) are about 55% of the total. The team then invests in prevention: three-amigos refinement sessions, API contract checks, better unit testing. Prevention spending rises by 10 lakh. Over the following year, external failure costs fall by 18 lakh and internal failure by 5, while appraisal drops slightly because fewer late fixes need full regression. Total cost of quality falls from 133 to 115 lakh, and failure falls to roughly 43% of the total.
The point of the example is the shape, not the numbers. Spending more in one bucket reduced spending in others by more. That is the central argument of the model, and it is the argument QA leaders most often need to make to justify investment that does not directly produce features.
Why failure costs are usually underestimated
Prevention and appraisal show up as clear line items. Failure costs hide in other people’s time. A developer spending a third of each sprint on bug fixes is recorded as development cost. A support agent handling calls about a failed UPI transaction is recorded as support cost. An on-call engineer awake at 2 a.m. is not recorded at all.
Some external costs are also genuinely hard to quantify: a customer who silently moves to a competitor, or a negative news story. It is better to list these as unquantified risks than to invent a figure. Where direct costs exist, though, they can usually be estimated from data the organisation already has: time logged against defect tickets, incident durations multiplied by the number of people involved, refunds and credits recorded by finance, support ticket volumes tagged by cause.
Why the timing of a defect matters
The model connects closely to a familiar observation: a defect costs more to fix the later it is found. A wrong requirement caught in a refinement discussion costs a few minutes. The same mistake caught in system testing costs design, code and test rework. Caught in production, it adds incident handling, customer impact and sometimes data repair.
Be careful with the precise multipliers that circulate (“100 times more expensive in production”). They come from old studies in very different contexts, and their methodology has been questioned. The direction of the effect is well accepted; the exact ratio varies enormously by system and should not be quoted as a law. Use your own organisation’s data where possible.
Using the model in practice
- Estimate the current split. Rough figures are fine for a first pass. The goal is to make failure costs visible next to QA spending.
- Find the biggest failure category. Is it production incidents, late-cycle rework or support load? That is where prevention will pay off fastest.
- Propose targeted prevention. Tie each investment to the failure cost it aims to reduce, not to a general promise of better quality.
- Re-measure after a few quarters. Show the shift between buckets. A falling external failure cost is the most persuasive outcome.
Where the model can mislead
The model can be pushed too far. Prevention does not reduce failure costs to zero, and beyond some point extra reviews and process slow delivery more than they save. Categorisation is also fuzzy: is a code review prevention or appraisal? It does not matter much, as long as the same rule is applied consistently over time.
Finally, cutting appraisal is not the same as improving quality. If a team reduces testing to lower its appraisal costs without strengthening prevention, failure costs will rise, often with a delay of a few releases that makes the connection easy to miss. Watching all four buckets together is what keeps that mistake visible.
Written by the RP Strativa Team. The team and cost figures in the example are illustrative.
Leave a Reply