The hidden cost of poor quality in manufacturing

A batch fails inspection. Production records the scrap, the quality team opens an investigation, and the monthly report carries a number everyone recognises. Meanwhile, a planner changes the schedule, a technician repeats a setup, customer service handles a return, and a replacement order goes out by express freight. Those activities may all belong to the same quality problem. They rarely arrive in the same report.

That is the question behind Embiggen X’s latest quality-control shorts: what does poor quality cost beyond the material in the scrap bin? In Episode 3 of AI in Swiss Industry, Stéphane Fallet points to consequences across the business, including returns, reprocessing and reputation. The useful next step is to trace those consequences to evidence, rather than attach a dramatic multiplier to scrap.

For a plant manager, quality leader or finance partner, the objective is practical: identify one recurring failure, assemble its cost across departments, and decide which change would prevent that bill from coming back.

01 Separate cost of quality from cost of poor quality

The terms need a shared definition before anyone adds up a total. ASQ’s cost-of-quality framework distinguishes prevention, appraisal, internal failure and external failure. The last two categories form the cost of poor quality: consequences of failures found before or after delivery.

A training session or a planned inspection is therefore different from the work required to recover a failed batch. Both belong in the wider quality picture, but they answer different management questions. Reducing inspection expenditure while increasing returns could make one budget look better while damaging the operation.

Agree that vocabulary with finance and quality. Keep ordinary prevention and appraisal visible alongside failure costs, so a proposal to improve a process can show both its additional investment and the failure costs it aims to avoid.

02 Follow the incident beyond the scrap report

Start with an actual incident that people can recognise: a rejected lot, a customer return or a recurring dimensional defect. Follow its identifier through production, quality, logistics and finance. Ask each team what changed because of it and what record supports that answer.

• On the line: material written off, additional processing, sorting and repeat checks attributable to the incident.
• In planning: a replacement order, a changed sequence or another setup with an identifiable effect.
• At the customer: return handling, agreed credits, replacement delivery and documented service work.
• In investigation: recorded engineering and quality hours spent containing and resolving the failure.

These are prompts for investigation, not charges to apply automatically. If a schedule changed but no additional cost or displaced production can be established, record the disruption as an operational measure. Do not invent a financial loss simply because the incident was inconvenient.

Reputation deserves the same discipline. A complaint or a deteriorating supplier score is a real signal. It is not a defensible lost-revenue figure by itself. Keep that exposure separate unless there is evidence of a lost order, an agreed commercial penalty or another attributable outcome.

03 Build a ledger around one event identifier

A useful first ledger can be a controlled spreadsheet. Each row needs an event ID, affected lot or order, date, cost category, evidence reference, amount or hours, owner and confidence status. Include both when the work happened and when the cost was posted. A return received this month may relate to a batch made much earlier.

Use explicit links where possible. A lot number shared by the inspection record and the production order is stronger than a similar description. If a return cannot be linked confidently, hold it for review. Unmatched records are a data issue to resolve, not a reason to assign them to the nearest convenient incident.

Separate documented charges, estimates and unresolved exposure. Finance should be able to open a row and see why it exists. An estimate can still support a decision, provided its method, assumptions and uncertainty are visible. A precise-looking total assembled from guesses is harder to challenge and easier to misuse.

04 Count each consequence once

The hardest part of a cost ledger is often deciding what is already included. A scrap write-off may include absorbed labour and overhead. Adding the original production hours again can duplicate part of the loss. A replacement invoice, credit note and returned item may also represent different accounting steps in one commercial remedy.

Ask finance to define the treatment for each cost type before the pilot begins. Document whether material values are gross or net of recovery, which labour rates apply, and how credits, recoveries and warranty provisions are reconciled. Use the same basis in the baseline and the test period.

Keep paid-out costs, staff capacity and potential revenue effects in separate totals. They can all matter, but they are not interchangeable. An hour released from rework is useful capacity. It becomes cash savings only when spending changes, or an evidenced contribution to output when that capacity is actually used.

05 Work through a bounded example

Consider a fictional incident ledger for one product family. It contains €8,000 of scrap already valued on an agreed accounting basis, €2,000 of additional rework labour excluded from that write-off, €1,200 of incremental replacement freight, and €800 of separately verified customer remedy costs. On those assumptions, the documented incident total is €12,000.

The purpose of the example is reconciliation: the visible scrap figure is €8,000, while the supported ledger contains another €4,000. These are illustrative amounts, not an Embiggen client result, a typical factory ratio or a promised saving. A different incident might contain no additional cost at all.

Suppose the team also records 30 engineering hours and a complaint from a major customer. Show those separately until the labour treatment and commercial consequence are established. Do not add an arbitrary reputation percentage. Equally, do not assume the full €12,000 is recoverable: some expenditure has already occurred, and a proposed intervention may address only one cause.

06 Use AI to prepare a traceable review

The first useful AI task may be assembling evidence rather than making a quality decision. A proposed workflow could extract lot references from approved documents, suggest links between returns and orders, group recurring defect descriptions, and prepare an incident summary. Each suggestion should retain its source record so someone can check it.

Test that proposal on a representative set of incidents that quality and finance have already reconciled. Compare suggested links with the reviewed reference set. Count missed costs, wrong matches and duplicates, as well as the time required to correct them. A fast summary that misassigns a customer return creates more work and a weaker decision.

Access permissions still apply across systems. The workflow should expose only the records its users are authorised to see. Treat retrieved document text as evidence to inspect, and require approval for changes to the ledger, financial treatment, customer communications or product disposition. Reading a record does not grant authority to act on it.

07 Turn visibility into prevention

A fuller bill is a starting point. The result should guide a corrective action owned by the people responsible for the process. If the largest verified cost comes from repeat sorting, investigate why the issue recurs. If it comes from late discovery, examine which earlier signal or process check could have changed the outcome.

AI may help identify patterns for investigation, but a pattern is not proof of a cause. A defect associated with one supplier could also reflect a different product mix, a changed specification or a measurement issue. Validate explanations with process knowledge and suitable testing before changing settings or supplier decisions.

The earlier AI quality-control guide examines inspection more broadly. This cost-ledger approach helps choose where to concentrate that work. It can also reveal a simpler intervention: clearer identifiers, a corrected work instruction or a more reliable handoff may solve the immediate problem without a new model.

08 Measure the pilot against a fair baseline

Choose one product family or failure mode with enough recurring incidents to compare. Agree the observation period, eligible records and finance treatment before testing. Compare rates and costs on a consistent production basis, accounting for changes in volume, product mix and reporting delays. Better detection can initially increase recorded cost because previously missing incidents become visible.

• Evidence quality: correct incident links, missing records, duplicate charges and reviewer corrections.
• Operating effort: review time per incident, unresolved exceptions and time to close an investigation.
• Process outcome: recurrence, scrap and rework on a consistent basis, alongside customer escapes.
• Economics: verified avoided expenditure, realised capacity benefits and the full cost of running the workflow.

Include integration, data preparation, licences or hosting, review, maintenance and training in the business case. Avoided cost is not the same as cash saved, and increased capacity is not automatically sold output. Finance should decide which benefits can be recognised and which remain an operational opportunity.

Run the proposed assistance alongside the existing process first. Keep quality and customer protections in place. A lower cost figure is not a successful result if it comes from weaker inspection, delayed recognition of returns or a shift of work into an unmeasured department.

09 Bring quality, operations and finance to the same review

Give each part of the pilot an owner. Quality validates incident classification and disposition. Operations confirms what happened on the line and whether proposed changes are workable. Finance approves valuation and reconciles the total. The data or engineering team maintains the links, access controls and supporting workflow.

Agree how disagreements are resolved. If AI suggests a return belongs to a lot and the quality owner rejects the match, retain the correction and reason. If finance excludes a cost from the recognised total, preserve it in the appropriate operational or estimated category. That record improves both the reporting and the next test.

Review one question together: which recurring failure now has enough evidence to justify action? This keeps the exercise attached to a decision rather than becoming another dashboard that reports a larger problem every month.

10 Start with the bill you can verify

Choose one recent quality incident. Bring its production order, inspection record, rework activity, customer history and financial entries into the same review. Agree what belongs to it, remove duplication and identify what remains unknown. That is a practical first step even if the records are fragmented.

When the evidence is connected, decide whether the next investment should improve prevention, detection, incident handling or data capture. Then test the change against the same ledger. AI earns its place when it helps the team make that decision more reliably and demonstrate the outcome.

Want to find where quality losses accumulate in your operation? Talk to Embiggen X about one workflow and the evidence needed to assess it.

Source conversation and reading

This article develops the question raised in Embiggen X’s October 30 short and the Episode 3 recording. The cost-category distinction follows ASQ’s framework linked above. The ledger, worked example and pilot design are proposed guidance, not reported webinar results. No universal cost percentage or AI saving is asserted.