01
Defect detection
BuyBuild index
Buy index
Buy or license
Strongest weighted driver: Data history fit.
- Data history fit
- Vendor maturity
- Local skills
- Strategic control
- Build cost advantage
Programme direction: Buy or license a vision system.
Tip the large scale Illustrative
Leans buy
GFT instruments / Digital procurement
Buy the model.
Own the data.
The illustrative cold-start case favours buying vision, forecasting and optimisation, building data capture, and keeping design in a partner pilot. Reweight each decision.
Operate the instrumentProcedural visualisation. Schematic, not a facility.
01 / An instrument for the decision
Should a cold-start factory build or buy each ML capability?
Pick a capability, rate the factory’s fit, reweight the priorities. The scale answers as you type. The default directions follow GFT’s Cameroon textile programme; every rating is an illustrative judgement, not a vendor finding.
02 / The procurement portfolio
Let each capability find its balance.
36.8:63.2
Mean build : buy index across the five capabilities, out of 100, from illustrative inputs. 1 Buy · 1 Build · 1 Pilot · 2 Defer · 0 Review.
02
Demand forecasting
DeferBuild index
Buy index
Defer: capture history
Strongest weighted driver: Data history fit.
- Data history fit
- Vendor maturity
- Local skills
- Strategic control
- Build cost advantage
Programme direction: Buy first, tune later.
03
Inventory optimisation
DeferBuild index
Buy index
Defer: capture history
Strongest weighted driver: Data history fit.
- Data history fit
- Vendor maturity
- Local skills
- Strategic control
- Build cost advantage
Programme direction: Buy or integrate.
04
Generative design
PilotBuild index
Buy index
Pilot with partners
Strongest weighted driver: Data history fit.
- Data history fit
- Vendor maturity
- Local skills
- Strategic control
- Build cost advantage
Programme direction: Pilot with partners.
05
Data capture
BuildBuild index
Buy index
Build lightly, own it
Strongest weighted driver: Data history fit.
- Data history fit
- Vendor maturity
- Local skills
- Strategic control
- Build cost advantage
Programme direction: Build lightly, own it.
Own the record. Test the model. Keep an exit.
Capabilities per verdict. Select a count to show only that verdict.
- Buy
- Build
- Pilot
- Defer
- Review
Live count of capabilities by verdict. Pilot and Defer gates come before the weighted preference.
Index points out of 100, derived from illustrative inputs. The lower pan favours that route. Select a card to make it the lens for the matrix, the method stage and the verdict; the lens card opens wide.
03 / Trace the contribution
Every point has an input.
| Criterion | Weight share | Build fit / 5 | Build points | Buy points | Build less buy |
|---|---|---|---|---|---|
| Data history fit | 20.0% | 1 | 4.0 | 16.0 | -12.0 |
| Vendor maturity | 20.0% | 1 | 4.0 | 16.0 | -12.0 |
| Local skills | 20.0% | 1 | 4.0 | 16.0 | -12.0 |
| Strategic control | 20.0% | 2 | 8.0 | 12.0 | -4.0 |
| Build cost advantage | 20.0% | 1 | 4.0 | 16.0 | -12.0 |
| Reconciled total | 100.0% | Not additive | 24.0 | 76.0 | -52.0 |
Build points plus buy points equal 100 before rounding. The margin is a preference index, not a probability. Individual displayed cells may differ from the displayed total because of rounding.
04 / Keep a decision trail
Save A. Challenge it with B.
Save the current portfolio as scenario A, then change the ratings or weights to create B. Until you save, A is the programme default. A survives in a shared link; it is otherwise held only in this page’s memory.
Load B:
AB
A build : buy
Change in build index, B less A
B build : buy
All five capabilities, A dashed over B solid. Select one to put it on the wipe above. Both beams sit level when B equals A.
Defect detection
A 24.0B 24.0+0.0
Buy or license
Demand forecasting
A 20.0B 20.0+0.0
Defer: capture history
Inventory optimisation
A 24.0B 24.0+0.0
Defer: capture history
Generative design
A 28.0B 28.0+0.0
Pilot with partners
Data capture
A 88.0B 88.0+0.0
Build lightly, own it
The A against B table
No A saved. The wipe and the strip above compare your inputs with the programme defaults.
05 / Method and judgement boundary
Make the weighting reproducible.
Four steps take the sliders to the verdict. The stage on the right redraws on every step, using your current inputs.
Share each weight.
Divide each entered weight by the sum of all weights. Five equal weights give every criterion 20 percent. The shares always add to 100 percent, so only relative importance matters.
wᵢ = entered weightᵢ / Σ entered weightsScore the fit to build.
Multiply each share by its build-fit rating divided by 5, then by 100. Vendor maturity is the only reversed rating: build fit equals 5 minus vendor maturity. Buy takes the complement under each criterion.
build pointsᵢ = 100 × wᵢ × fitᵢ / 5Weigh the two pans.
Add the contributions for the build index B. Buy index Y is 100 minus B, so the pans share 100 points. Equal-weight defect detection has fits of 1, 1, 1, 2 and 1: 6 of 25, so B is 24 and Y is 76. That is arithmetic, not a 76 percent chance of success.
B = Σ build pointsᵢ; Y = 100 − B; margin = B − YRead the verdict, not only the margin.
A margin from minus 10 to plus 10 returns Review: an illustrative governance band, not a confidence interval. Forecasting and optimisation show Defer when data-history fit is below 2 of 5. Generative design always stays a partner pilot.
review if |margin| ≤ 10; defer if history < 2
Assumptions and live delta register
Default assumptions and basis, beside your live values.
All rows11 rows shown. Lit row: Defect detection ratings.Point at any column header or row to inspect its rule and live delta.
| Assumption | Default / rule | Live now & sparkline | Status and basis | Jump |
|---|---|---|---|---|
| Capability directions | Buy models; own capture; partner design | A direction, not a score5 routes | GFT programme digital strategy, Build Against Buy, page 14. Qualitative recommendation. | |
| Rating order | History, vendor, skills, control, cost | Fixed order0 to 5 scale | Illustrative ordinal scoring scale, 0 to 5. | Open |
| Defect detection ratings | 1 / 4 / 1 / 2 / 1 | 1 / 4 / 1 / 2 / 1Δ +0.0 | Illustrative translation of the programme direction; not sourced numerical measurements. | Open |
| Demand forecasting ratings | 0 / 4 / 1 / 2 / 1 | 0 / 4 / 1 / 2 / 1Δ +0.0 | Illustrative translation of the programme direction; not sourced numerical measurements. | Open |
| Inventory optimisation ratings | 1 / 4 / 1 / 2 / 1 | 1 / 4 / 1 / 2 / 1Δ +0.0 | Illustrative translation of the programme direction; not sourced numerical measurements. | Open |
| Generative design ratings | 0 / 2 / 0 / 3 / 1 | 0 / 2 / 0 / 3 / 1Δ +0.0 | Illustrative translation of the programme direction; not sourced numerical measurements. | Open |
| Data capture ratings | 5 / 1 / 4 / 5 / 4 | 5 / 1 / 4 / 5 / 4Δ +0.0 | Illustrative translation of the programme direction; not sourced numerical measurements. | Open |
| Criterion weights | 1 / 1 / 1 / 1 / 1 | 1 / 1 / 1 / 1 / 1Equal 20% | Illustrative equal importance. No policy preference assumed. | Open |
| Review band | Absolute margin ≤ 10 | Selected margin -52.0±10.0 pts | Illustrative sensitivity screen, index points. | Open |
| History screen | Fit rating ≥ 2 | Forecast 0.0, optimisation 1.02 gated | Illustrative readiness judgement. No specified months or sample count. | Open |
| Beam angle | Margin × -0.16 degrees | Selected beam 8.3 degrees-0.16° / pt | Illustrative visual mapping only. No additional model input. | Open |
Method source and architecture
The model stays visible.
Build index = Σ (normalised weight × build-fit rating ÷ 5) × 100. Vendor maturity is reversed for build fit. Buy index = 100 − build index. All numerical defaults are illustrative judgements, not measured vendor performance.
Criterion contribution profileDefect detection · B 24.0 : Y 76.0
Every criterion multiplies its normalised weight share by build fit out of 5, then scales to 100 index points.
GFT Cameroon textile programme, digital strategy, preparation 31 January 2024, sources refreshed 2026. Qualitative recommendations only; numerical scoring is illustrative.
A planning aid. Not investment, legal or tax advice. Verify inputs and boundaries before an operational decision.
Live substitution for Defect detection. Select a term to highlight its role or move a slider above.
- Data history fit20.0% × fit 1.0 / 5 × 100
- Vendor maturity20.0% × fit 1.0 / 5 × 100
- Local skills20.0% × fit 1.0 / 5 × 100
- Strategic control20.0% × fit 2.0 / 5 × 100
- Build cost advantage20.0% × fit 1.0 / 5 × 100
B = Σ24.0Y = 100 − B76.0margin-52.0
06 / Use the result on the factory floor
Own the evidence before the commitment.
01/ 04 · Floor reading
Next task
Illustrative pose for this operational check, not a score.
Live floor interpretation
Defect detection: Buy or license. The procurement score favours buy. Validate the bought model on the actual fabric, camera, lighting and reject workflow.
01 / 04
Name the next practical task.
A factory manager can use the score to assign the next task: establish the data boundary, inspect a candidate system, or reserve a capability for later. The index replaces none of those checks.
02 / 04
Challenge the ratings.
An engineer should test each rating against integration effort, edge operation, power interruption, local buffering and ongoing monitoring.
03 / 04
Inspect the lifecycle.
An investor should read the same assumptions beside lifecycle costs, contractual data access and a credible exit route.
04 / 04
Own the data, not every layer.
Owning data means controlling identifiers, definitions, access and portability. A small internal build can still depend on external services.
Before the money moves
Read the limits first.
Conditions that the score cannot see. Select a limit on either side and the scale shows what it does to the index.
Impact on the scale
±4.0
index points on the build index for each rating point on build cost advantage, at its live weight share of 20 percent.
Impact and boundary profile across all six limits
Cost advantage is a judgement, not a cash-flow model.
Put subscription costs, hardware, retraining, maintenance, integration, data export and replacement on the same planning horizon before rating it. A cheap licence with expensive integration can change the decision.
Three checks before the money moves.
Before buying vision, test representative fabrics and defect classes with a held-out inspection set and a human escalation route. Before using a forecast, assess whether the history represents current products, demand and lead times. Before running an optimiser, agree the operational constraints and how planners can reject an infeasible recommendation.
What the screens do not say.
Passing the history screen means only that you judge history adequate for a next assessment. It says nothing about sample bias, seasonal coverage or model accuracy. Ratings need a documented basis from the factory and its candidate suppliers before a real decision. A partner design pilot needs an explicit production review before its output reaches the line.
The source supports directions, not these scores.
GFT’s Cameroon textile programme was prepared on 31 January 2024, with source refreshes in 2026. Its digital strategy recommends buy-before-build for a cold-start operation, buying or licensing defect detection, buying forecasting before tuning, buying or integrating optimisation, piloting generative design with partners, and building lightweight data capture that the venture owns. This tool interprets that qualitative position; it does not disclose programme source documents or certify today’s vendors.
Read the connected research questions.
- WP03, Seeing Every Defect: inspection quality and the conditions for a vision trial.
- WP04, Forecasting From a Cold Start: the history problem behind the readiness screen.
- WP14, The Industrial Data Spine: governance and lightweight model operations.
- WP18, Edge to Dashboard: architecture under power and connectivity constraints.
These are related paper titles from the programme library plan, not claims that released PDFs are available. Related instrument: find production losses with the OEE line tool. Use a measured operating baseline when defining a digital pilot’s success.
Limitations.
Additive weights assume independent criteria; history, skills and costs can overlap in practice. Complementary indices force a relative choice and cannot represent “both routes are unsuitable”. The review band and readiness screen are assumptions. This is a planning aid, not investment, legal or tax advice, a vendor ranking, or approval to deploy.
Step 1 · The margin
BUILD
BUY
Margin -52.0 index points. Review band: plus or minus 10.
Step 2 · The trial
0 of 5 trial checks ticked.
Step 3 · The boundary
Build and buy points share 100. An index, not a probability of success.
07 / Take the verdict to a trial
A procurement preference needs an operating test.
Take the selected route into a bounded trial. Confirm data ownership, support, edge operation, integration cost and an exit path before committing capital.
Verdict for Defect detection
Buy
Buy, then prove it on your own line.
Take this into a bounded trial.
08 / Portfolio hand-off
Carry the settled posture into site selection.
All five capabilities reconciled to 100 index points. Select any capability to inspect its final beam before moving to the next instrument.