Fixed reference example, illustrative defaults. Each channel uses the same normalised defect cohort. Glyphs round detection to a whole percentage; read-outs retain the calculation.
Capital recovered / illustrative default11.4 months
Month 0Month 24
GFT instruments / Inspection economics
A missed defect
has a second cost.
Compare escaped garments and their cost. Model detection, returns and camera expenditure, then see whether avoided losses repay the system. All commercial defaults are illustrative.
01 / An instrument for the decision
What does manual inspection miss that vision could catch, in garments and money?
Change the inputs below. The tally, the cash curve and the rest of the page follow. In this scenario vision catches 135 additional defective garments a month, for a net saving of 525,000 XAF a month and a simple payback of 11.4 months.
Illustrative inputs / XAF The manual range comes from the programme annex. The vision point and every commercial value are illustrative. These figures describe an editable scenario, not GFT operating results.
The instrument rack / every control writes to the whole page
Ten inputs, one cohort of threads.
The illustrative case recovers capital within the selected horizon. Confirm incremental costs and measured detection before relying on it.
Cumulative net cash at month 12: 300,000 XAF
Solid mint: current B. Dashed gold: saved A. Common axes; no cash past a horizon.
Reconcile the tally
Catches, escapes and money add up to one saving.
The cash line above ends here: every figure on it is a row of this ledger.
Reconcile catches, escapes and money.
- Defective garments / month
- 500
- Manual caught / month
- 325
- Manual escaped / month
- 175
- Vision caught / month
- 460
- Vision escaped / month
- 40
- Incremental cost per escape
- 5,000 XAF
- Manual escape loss / month
- 875,000 XAF
- Vision escape loss / month
- 200,000 XAF
- Gross saving / month
- 675,000 XAF
To recover capital inside this horizon, each additional catch must avoid at least 2,963 XAF of incremental escape cost.
Ledger row / the model line behind it
Gross saving / month
Line 5, Net saving, then payback: net = catches × C − running cost
Hover or focus a row to read the line that produces it.
Keep the alternative beside the decision.
Save scenario A, change an input, then compare with current scenario B. A is held in this tab until reset; shared links and CSV files include both cases.
| Measure | A | B |
|---|---|---|
| Net saving / month | ||
| Simple payback | ||
| Cash at own horizon | ||
| Vision escapes / month |
The model stays visible
Five lines carry every figure on this page.
Pick a line. The card lifts, its thread lights, and the inputs that feed it are listed below with their live values.
A planning aid. Not investment, legal or tax advice.
Method and source
Expected escaped garments = monthly volume × defect prevalence × (1 − detection). Additional catches × incremental cost per escape − incremental monthly running cost = net monthly saving. Simple payback = installed capital ÷ positive net monthly saving.
Method source: R29, GFT programme annex: MDPI and industry inspection literature. Broad claims, not a verified factory benchmark. Verify inputs and boundaries before an operational decision.
D = volume × defect rate / 100garments / month
E = D × (100 − detection) / 100manual and vision escapes / month
E manual − E visioncatches vision adds / month
C = return + rework + reputationper escaped garment
net = catches × C − running costa month, then months to recover capital
Line 1 selected / inputs that feed it
- Monthly inspected output10,000
- Defective garments5%
Line 2 selected / inputs that feed it
- Manual detection65%
- Vision detection92%
Line 3 selected / inputs that feed it
- Manual detection65%
- Vision detection92%
Line 4 selected / inputs that feed it
- Returns and handling3,000 XAF
- Late rework premium2,000 XAF
- Reputation allowance0 XAF
Line 5 selected / inputs that feed it
- Incremental running cost150,000 XAF
- Installed camera capital6,000,000 XAF
Hover a ledger row above to light its line. Hover a line to light its inputs in the instrument. Values follow the sliders.
02 / The calculation stays visible
Compare escapes before comparing cameras.
Five steps take the sliders to the payback. The stage redraws on every step, using your current inputs.
Same defective cohort / current scenario
escapes a month. Whole thread: caught. Broken thread: escaped.
D = 10,000 × 5% / 100
defective garments / month
Each thread is one percent of inspected output. Broken threads are the defective share.
Manual 65% detection
escapes / month
Vision 92% detection
escapes / month
E = D × (100 − detection) / 100
- Caught by both
- Caught only by vision
- Missed by both
additional catches / month
175 − 40
- Returns and handling
- 3,000 XAF
- Late rework premium
- 2,000 XAF
- Reputation allowance
- 0 XAF
per escaped garment
simple payback on 6,000,000 XAF
Cash at horizon: 6,600,000 XAF
Step 1 of 5
Count garments, not marks.
The unit is a defective garment, not a defect mark. A garment with several defects counts once. Use the same inspected volume and defect prevalence for both systems.
D = volume × defect rate / 100
Step 2 of 5
Detection means recall, not accuracy.
Detection means the share of truly defective garments identified, sometimes called recall. It is not the percentage of all garments classified correctly. In a low-defect population, overall accuracy can look high even when a system misses most defective garments.
E manual = D × (100 − manual detection) / 100
E vision = D × (100 − vision detection) / 100
Step 3 of 5
The gap between the escapes is what vision adds.
Each detected population plus its escapes must equal the same defective population. Expected values may be fractional because this is a rate model; actual garment counts are whole numbers.
Additional catches = E manual − E vision
Step 4 of 5
Price one escape, once.
Estimate the average incremental consequence of an escaped defective garment across all escapes, including those that never generate a return. The return allowance should therefore reflect return probability rather than the full cost of every possible return.
Keep late rework separate from handling already counted elsewhere. Leave reputation at zero unless there is defensible evidence of an expected loss.
C = return cost + late rework premium + reputation allowance
Step 5 of 5
Net saving first, then payback.
The difference between manual and vision escape losses must equal gross savings. Adding capital back to horizon cash must equal the horizon multiplied by net monthly saving. These identities keep the garment boundary and the cash boundary aligned.
Simple payback assumes a constant monthly saving from the first month and expresses recovery as a fraction of a month. It is not a discounted return or a financing metric. When net saving is negative, there is no finite payback. A payback later than the selected horizon remains visible, but the interpretation states that the horizon does not recover capital.
Net saving = additional catches × C − monthly opex
Cash at month t = t × net saving − capital
Simple payback = capital / positive net saving
03 / Every default has a status
The reference case is an assumption register.
R29 is a grade-C programme-annex finding attributed to MDPI and industry inspection literature. Its manual statement gives a range from 60 to 70 percent, while the vision statement says above 90 percent. The annex also reports a miss-rate range that is not the mathematical complement of its manual accuracy range. These statements cannot establish a reliable operating recall. The underlying article has not been opened for this tool. The default is a scenario interpretation that needs testing, not a certified benchmark.
R29 manual midpoint with an illustrative vision point. Not a sourced factory case.
| Input and basis | Default | Live now | Payback at +20% |
|---|---|---|---|
| 10,000 garments/month | 10,000 garments/month | 9.1 months | |
| 5 % | 5 % | 9.1 months | |
| 65 % | 65 % | 30 months | |
| 92 % | 92 % | 8.3 months | |
| 3,000 XAF/escape | 3,000 XAF/escape | 9.9 months | |
| 2,000 XAF/escape | 2,000 XAF/escape | 10.4 months | |
| 0 XAF/escape | 0 XAF/escape | 11.4 months | |
| 6,000,000 XAF | 6,000,000 XAF | 13.7 months | |
| 150,000 XAF/month | 150,000 XAF/month | 12.1 months | |
| 24 months | 24 months | 11.4 months |
The cautious preset holds the same commercial assumptions and uses manual detection at the upper end of the annex range, with vision detection at its lower threshold. It tests a smaller detection advantage. The low-defect preset changes only prevalence to an illustrative one percent. These are sensitivity tests, not claims that either operating state has been observed. Costs stay in CFA franc (XAF) throughout; the model applies no exchange rate.
05 / Turn the scenario into a trial
Measure the defects the model cannot see.
A scenario is only as good as the detection rate behind it. The scorecard shows which cells the model fills and which only a labelled trial can fill.
Labelled set / one thread is one percent of the defective cohort
Defective and flagged. True positives. The model counts these as detected defective garments: the defective population times the detection rate.
Counts are expected garments a month from the model, for the system selected. Cells marked for the trial are outside the model.
- 01
Build one labelled set.
Build a labelled trial set that represents the proposed line’s fabrics, colours, sizes and defect severity. Separate training material from evaluation material. Ask qualified reviewers to establish the reference labels and resolve disagreement before scoring either system. Run manual and vision inspection against the same set. Preserve the confusion matrix, including false positives, false negatives and genuinely conforming garments.
- 02
Repeat it under the line’s own conditions.
Then repeat the evaluation under operating light, speed and handling conditions. Camera coverage, occlusion, dirty lenses, product changes and calibration drift can alter performance. The model does not simulate those mechanisms.
- 03
Price what the model leaves out.
Add incremental recurring burdens to running cost and installation burdens to capital where appropriate. Do not assume that catching a defect creates the same cash benefit as avoiding it.
The model also omits
0 of 9 marked to price in the trial. Net saving of 525,000 XAF a month is before any of them.
The instrument is a planning aid, not investment, legal or tax advice, a supplier quotation or a guarantee of inspection performance.
06 / Use the result in its boundary
A pilot must establish the detection rate.
Measure manual and vision inspection on the same labelled garments, under the same operating conditions. A scenario is useful when it changes a test plan, not when it replaces one.
Keep the early-repair cost inside the boundary.
The avoided cost of an escape must be incremental to finding that garment early. Count returns, late rework and reputation only once. This model holds staffing constant and does not predict orders, sales or supplier performance.
Break-even detection for payback inside 24 months: 81%. Below it, capital is not recovered in the horizon.
labelled defective garments to estimate recall within the margin
Normal approximation, 95 percent confidence. Planning arithmetic, not a trial design.07 / Carry the tally forward
Inspection belongs in the production model.
- InstrumentOEE lineTest how rejects take productive time.Open the instrument
- MethodMeasurement approachDefine one operating boundary for every metric.Read the approach
- Paper wp03Seeing Every DefectAccuracy, economics and deployment of vision inspection.Open the paper page
Escapes a month and payback follow the current scenario, illustrative inputs. The paper is a programme paper: check its status on its own page. A defect found earlier may reduce escape loss without raising throughput, so keep those benefits separate in any larger business case.