Quantifying what poor quality really costs you, to defend a budget

Ask a quality manager what poor quality costs their site. They will give you an order of magnitude, with a hesitation in their voice.

They know the real figure is scattered across four systems that do not talk to each other:

• scrap lives in production data;
• rework hides in labour time records;
• returns sit in logistics;
• complaints sleep in a separate spreadsheet.

Nobody adds them up. And that is precisely why the quality budget is the first to be cut: you cannot defend a cost you cannot quantify.

Why does the real cost stay invisible?

The cost of quality breaks down classically into four families: prevention, appraisal, internal failures, external failures. In theory, a simple table is enough.

In practice, each family is fed by a different system, with its own units and its own owners. That is where the total gets lost.

Table scrolls horizontally on small screens.

The result is familiar: everyone sees their share, nobody sees the total.

The production manager knows their scrap rate, but not what the customer return cost in freight and credit notes. Customer service handles complaints without knowing they point to the same defect as the internal scrap.

These are exactly the hidden costs of quality: not concealed spending, but real spending that is never consolidated. Poor quality is paid in full, in silence, because the invoice is never brought together in one place.

What does AI actually change?

The aggregation work is exactly what a human hates doing and what an AI does well: going to fetch data in different formats from different sources, linking it, and putting a value on it.

An agent can read scrap data from production, rework hours from time records, credit notes from logistics and the flow of complaints, then output a consolidated annual cost.

In practice, aggregation always follows the same sequence:

Above all, AI can link what belongs together: showing that the internal scrap, the rework and the customer return trace back to the same defect, and therefore to the same euro spent three times.

This costing, which took days of manual collection when someone tackled it for a management review, becomes an operation of a few hours, repeatable every month. That is what separates a real calculation of the cost of poor quality from a rough once-a-year estimate. It is also one of the least glamorous but most useful uses of AI in industry: it predicts nothing, it simply gathers figures that already exist and puts them side by side.

How do you produce an honest number rather than a flattering one?

Quantifying poor quality exposes you to the same trap as a business case: the temptation to inflate in order to convince. It is counterproductive.

A cost measured on real data, with owned assumptions, withstands management's first question. An inflated cost collapses at the first check, and discredits everything else.

AI helps hold that discipline. It cleanly separates two kinds of cost:

Table scrolls horizontally on small screens.

The first goes into the number, the second is cited without being quantified. It is the same rule that applies when defending a project before a committee: you do not mix what you prove with what you assume.

• Inflating the total to impress.An exaggerated figure collapses at the first check and takes the credibility of the whole exercise with it.
• Quantifying the unquantifiable.Putting a precise value on brand image or lost customers gives the illusion of rigour; these effects are cited, they are not added up.
• Counting the same euro twice.Without tying items to their common cause, a defect can appear at once as scrap, rework and return, and inflate the total.
• Presenting a total without stating its limits.A figure drawn from incomplete data stays useful, provided you say where it comes from before management finds out.

How does this number unlock a decision?

Consolidating the cost of poor quality in industry is not about producing one more indicator. It is about unlocking an investment decision.

As long as the invoice stays scattered, prevention looks expensive and its return invisible. Once the total is on the table, the reasoning flips: you compare what the defect costs, every year, with what it would cost to treat it at the root.

It is the same mechanic as defending a project before management: a cost measured today against a gain owned tomorrow.

The consolidated figure also surfaces priorities. Out of dozens of defects, a few concentrate most of the cost.

Those are the ones to attack, and that is where the link with recurring non-conformities becomes concrete: the recurrence that weighs most in volume is often the one that weighs most in cost.

Detecting the repetition and quantifying its cost are the two halves of the same work:

• one says what to fix;
• the other says why it is worth it.

Does AI invent data, or does it just aggregate?

You have to stay honest about what AI does here. It creates no data: it links and adds up what already exists, scattered.

If your scrap data is wrong, the total will be wrong. AI does not fix incomplete cost accounting, it exposes it.

That is useful too, by the way: discovering that you cannot measure your own scrap is a first result in itself. Many sites learn, on this occasion, that their tracking of scrap, rework and returns was never set up to be added together.

But the value of the final figure will never exceed that of the data feeding it. It is a limit to state before presenting a total to management, not after.

What to take away

Poor quality is expensive, but it is expensive in silence, because its invoice is scattered.

Consolidating it is not one more accounting exercise: it is what turns a problem management ignores into a budget it funds.

AI does the aggregation work; the assumptions and the decision to invest remain yours.