You will be offered dozens of metrics by consultants, software vendors and your own managers. Five of them determine whether you make money. The rest exist to diagnose why one of the five moved.
The five that predict profit
| KPI | Definition | Measured where | Reference point |
|---|---|---|---|
| Packed wholes | Wholes as a share of packed saleable kernel | Grading and packing, by lot | 75% of packed output, against 25% brokens |
| Production KOR | kg of final packed saleable kernel per 100 kg of RCN actually processed | Monthly mass balance — RCN issued from store against kernel packed | 22–24 kg per 100 kg |
| Capacity utilisation | Tonnes of RCN processed ÷ nameplate tonnes available in the period | Daily production log against installed capacity | Africa under 50%, Asia over 80% (UNCTAD) |
| RCN cost per kg of saleable kernel | Landed RCN cost per tonne ÷ kg of saleable kernel per tonne | Purchasing and mass balance combined | Compare weekly against your realised blended kernel price |
| Conversion cost per tonne of RCN | All cost except raw material, per tonne of RCN processed | Monthly management accounts | Roughly US$155/t semi-automatic at Asian wage levels; US$490–580/t manual at African wage levels |
The 75/25 packed split is the figure this guide uses throughout. It sits inside every published range: Azam-Ali & Judge put the whole-to-split ratio at 55–85% of actual output depending on method and management, with 65% as the “satisfactory” standard; more recent equipment data puts a mechanised line at 75–90%; and a peer-reviewed Cambodian project appraisal adopts 75/25 as its working assumption. Anything above 85% packed wholes should be treated as a claim to verify, not a plan to budget on.
Conversion cost is a range, not a number, and the axis is automation and wage level. A peer-reviewed Cambodian appraisal of a 960 t/year semi-automatic plant gives a cash conversion cost of about US$102/t of RCN — roughly US$155/t once straight-line depreciation and finance are added. TechnoServe’s Mozambique dataset, on a manual process, gives US$455–491/t on the same basis. Both are internally consistent; they describe different plants. Labour alone accounts for most of the gap, at about US$50/t against US$280/t. State which basis you are using, state what is inside it, and never mix the two in one model. Neither figure includes packaging or export logistics — carry those as separate lines.
The two KOR figures, and why they must never be compared
Quality-check KOR and production KOR are different quantities. Quality-check KOR is a laboratory cutting test used to price a lot of raw nuts, expressed in pounds of kernel per 80 kg bag. Production KOR is kilograms of final packed saleable kernel per 100 kg of RCN.
The conversion is lbs per 80 kg bag = % × 1.7637. So a 48 lb cutting test is about 27% laboratory outturn, while the realistic factory figure is 22–24%. The laboratory test counts kernel that the line then loses to breakage, rejects and handling.
A plan built on the laboratory number overstates revenue by roughly 10–15%. It is one of the most common and most expensive modelling errors in the sector.
The same warning applies to whole-kernel measurement. Do not benchmark packed wholes against the whole-cutting rate at shelling — they are measured at different points, and packed wholes is necessarily the lower of the two, because peeling, grading and handling each take a further cut after the sheller. Both belong on the board, separately:
| Metric | Measured at | Reference |
|---|---|---|
| Whole cutting rate | Shelling, on a fixed sample mass per machine | 85–90% machine; good manual cutting 90–95% |
| Packed wholes | Final packed output, by lot | 75%, against 25% brokens |
What one point is actually worth
This is the arithmetic worth internalising, because it settles where to spend improvement money.
A point of whole-kernel share. At 230 kg of saleable kernel per tonne of RCN (23% production KOR), one percentage point shifting from wholes to splits moves 2.3 kg per tonne of RCN. Whole grades carry roughly a 15–16% premium over white splits at 2026 reference prices — about US$1.10/kg on a whole-kernel price near US$7.05/kg.
2.3 kg × US$1.10/kg ≈ US$2.53 per tonne of RCN, per point of wholes lost
A point of production KOR. One percentage point is 10 kg of kernel per tonne of RCN. At a blended realised export price of about US$6,810/t — blended, because it includes splits, pieces and butts rather than pricing everything at the W320 quote:
10 kg × US$6.81/kg = US$68 per tonne of RCN, per point of KOR
US$68 ÷ US$2.53 ≈ 27×
A point of yield is worth about twenty-seven times a point of grade, because yield moves the quantity you sell while grade only moves the price on what you already have. Chase yield before you chase grade.
For scale: the industry’s gross spread in 2025 was roughly US$245 per tonne of RCN. Ten points of whole-kernel share is about US$25/t — a tenth of everything there is. Ten points of KOR would be more than the entire spread, which is why nobody achieves it and why the first two points are worth so much effort.
A caution on the price gap. An older Indian data point put the whole-to-split discount at about 23%. Driving operating decisions off that instead of the current 15–16% inflates the apparent cost of breakage by roughly half, and will make you buy machinery you do not need. Use current prices for operating decisions and keep the historic range for sensitivity only.
Move either slider and watch the two deltas diverge. The ratio holds across the whole realistic band — it is not an artefact of the baseline.
kg of packed saleable kernel per 100 kg of RCN. 22–24% is the realistic factory band.
Wholes as a share of packed kernel. 75/25 is the planning figure; above 85% is a claim to verify.
Yield moves the quantity you sell. Grade only moves the price on what you already have.
| KOR \ Packed whole share | 55% | 65% | 75% | 85% |
|---|---|---|---|---|
| 21% | US$1377 | US$1400 | US$1423 | US$1446 |
| 22% | US$1442 | US$1466 | US$1491 | US$1515 |
| 23% | US$1508 | US$1533 | US$1558 | US$1584 |
| 24% | US$1573 | US$1600 | US$1626 | US$1652 |
| 25% | US$1639 | US$1666 | US$1694 | US$1721 |
Two warnings on capacity utilisation
Nameplate is quoted on a Size A input basis. Machine capacity varies with raw-nut size, so your effective nameplate on a small-nut origin is lower than the plate says, and your utilisation looks worse than it is.
Nameplate in tonnes per day hides the season. A 30 t/day plant running a 120-day season is a 3,600 t/year plant. Measure utilisation against the days you can actually feed the line, and report both figures — the daily rate tells you whether the plant is balanced, the annual tells you whether it is financed.
The secondary KPIs, and what each one diagnoses
| KPI | Reference | What a bad reading tells you |
|---|---|---|
| Broken rate at shelling | Under 3%; machine makers claim 3–5% | Blade sharpness, blade adjustment against calibre, feed rate |
| Uncut / unshelled rate | Under 5% | Calibration discipline, steam penetration, machine setting |
| Peeling first-pass efficiency | 80–90% specified; factories report as low as 70% | Humidification control, kernel moisture, testa adhesion at origin |
| Peeling breakage | 4–8% | Over-dried kernel, aggressive settings, excess manual finishing |
| Rework volume | No published benchmark | Where the process is producing product it cannot sell first time |
| Downtime by cause | No published benchmark | Usually material starvation and power, not machinery |
The last row is the one most plants get wrong. Downtime is recorded against machines because machines are visible, but in practice the line stops because nuts did not arrive at it or the power failed. Record the cause, not the location, or you will spend maintenance budget solving a procurement problem.