How to calculate the cost of poor quality (with a worked example)

Cost of poor quality (COPQ) is the money a process burns because it doesn't work first time. It is the single most useful number when you argue for change, because it converts "this is frustrating" into "this costs £X a year", and finance only funds the second one.

• Internal failure, rework, scrap, re-keying, chasing, re-approvals. Caught before the customer sees it.
• External failure, complaints, credits, refunds, warranty, churn, penalties. The customer saw it.
• Appraisal, checking, inspection, audit, double-signing. The cost of looking for problems.
• Prevention, training, better forms, automation, standard work. The only category you generally want to spend more on.

A healthy improvement plan moves money out of the first three and into the fourth. If your plan only adds more checking, you have increased appraisal cost and fixed nothing.

For any recurring failure, the annual cost is:

annual cost =
    occurrences per year
  × time lost per occurrence (hours)
  × fully-loaded hourly cost of the people involved
  + direct cash losses (credits, penalties, scrap, expedite fees)

"Fully loaded" means salary plus employer's national insurance, pension, holiday cover and overhead, typically 1.25 to 1.4 times base salary. Using base salary alone understates the number by roughly a third, and a good finance director will spot it.

An onboarding team re-requests missing customer documents on 40% of new accounts. They open 250 accounts a month.

• Occurrences: 250 × 0.40 × 12 = 1,200 a year
• Time lost: 35 minutes per occurrence across the analyst and the relationship manager = 0.58 h
• Fully-loaded cost: £42 an hour blended
• Direct losses: 15 deals a year slip past the quarter end, costing £900 in expedite fees

1,200 × 0.58 × £42 = £29,232, plus £900 = ~£30,100 a year from one avoidable failure mode.

Real data is always incomplete. That is not a reason to skip the number, it is a reason to state what you assumed, right next to it. Every COPQ figure should carry:

• Where each input came from (system report, sample, or someone's estimate)
• The date range it covers
• Which figures are estimates, and how wide you think the range is
• What would need to be measured to firm it up

A number with stated assumptions survives challenge. A confident number with hidden assumptions gets demolished in the first meeting.