A cashew factory can run 300 days a year. A cashew harvest lasts three to four months. Those two facts together are the entire working-capital problem, and no amount of engineering makes them go away.
This is the line that most often kills otherwise sound cashew projects — not because the money is unavailable, but because the standard project-report templates size it wrongly, and the error is only discovered after the machines are paid for.
Why the season sets the requirement, not the balance sheet
The requirement is set by the gap between when you must buy and when you can sell. ComCashew states it plainly: working capital is needed to buy raw cashew nuts within the three-to-four-month harvest window in order to stock raw material for processing all year, and again to pay overheads for at least three months before any proceeds from kernel sales arrive.
The pattern repeats across continents. CBI has recorded West African units running only three to four months a year. Cambodian processors report the same constraint — a harvest spanning roughly three months forces them to secure a full year’s supply at once, and many local factories consequently operate only three to six months annually.
Your competitor does not carry this in the same way. A buyer sourcing across Northern and Southern hemisphere harvests finances perhaps three months of inventory. A single-origin processor finances twelve — and historically at a higher cost of capital than the Asian processors they are selling against. Longer inventory at a higher rate, against the same kernel price, compounds hard.
The arithmetic that breaks the standard templates
Substituting one realistic holding period into a published model multiplies its working-capital line roughly eightfold. This is the single most important calculation in cashew project finance, and it does not appear in any published model project report.
NABARD’s 2014 model, for a 500 t/year unit, computes working capital on holding periods of raw material 30 days, work in progress 7 days, finished goods 30 days, debtors 7 days and expenses 30 days. Total current assets in Year 2 come to ₹144.37 lakh, of which the raw-material line is ₹54.08 lakh at 30 days.
Thirty days is entirely defensible in India, because India imports RCN year-round from several origins and a processor can buy monthly.
Now assume a single-origin processor in West Africa, East Africa or Cambodia, who must buy a year’s raw material in one local window and carry it. Substitute 240 days of raw-material stock:
₹54.08 lakh × (240 ÷ 30) = ₹432.6 lakh
— against an original total project cost, in the same model, of ₹158.54 lakh.
The raw-material line alone, correctly modelled, is 2.7 times the entire project cost of the published model. That is not a conservative adjustment. It is a different business, and it means a template built on the Indian format understates a single-origin processor’s true funding requirement by roughly the cost of the factory.
The published models also disagree with each other, which is a fair sign they are not converging on a real number:
| Model | Raw-material holding | Total current assets | Fixed assets | Current : fixed |
|---|---|---|---|---|
| NABARD, 2014 | 30 days | ₹144.37 lakh (Yr 2) | ₹122.45 lakh | 1.18× |
| MSME profile, FY2016-17 | 75-day cycle | ₹75.05 lakh | ₹72.00 lakh | 1.04× |
| NIFTEM-T / PMFME | 7 days | ₹107.70 lakh (Yr 2) | ₹16.73 lakh | 6.4× |
The direction is right in every case — current assets dominate fixed assets in this business — but the magnitude is not. Treat these templates as a format to fill in, never as a source for the holding period itself.
Run the year yourself
Move the two sliders and watch the trough. The peak is not the money you spend — it is the deepest point of the cumulative position, which arrives months after the last purchase and is what your facility has to cover.
A single-origin plant running 300 days needs 8–9; a buyer sourcing two hemispheres finances 3–4.
Months, opening in February in this geography.
| Month | RCN bought (t) | Cash out (US$) | Cash in (US$) | Cumulative (US$) |
|---|---|---|---|---|
| Jan | — | 200,000 | — | -200,000 |
| Feb | 667 | 739,750 | — | -939,750 |
| Mar | 667 | 739,750 | — | -1,679,500 |
| Apr | 667 | 739,750 | 405,375 | -2,013,875 |
| May | — | 139,750 | 405,375 | -1,748,250 |
| Jun | — | 139,750 | 405,375 | -1,482,625 |
| Jul | — | 139,750 | 405,375 | -1,217,000 |
| Aug | — | 139,750 | 405,375 | -951,375 |
| Sep | — | 139,750 | 405,375 | -685,750 |
| Oct | — | — | 405,375 | -280,375 |
| Nov | — | — | 405,375 | 125,000 |
| Dec | — | — | — | 125,000 |
Two things are worth noticing. Shortening the cover reduces the peak but buys idle months at the end of the year, and an idle plant still carries its fixed costs. Compressing the harvest window does not change how much you spend — it changes how fast, and a two-month window puts the whole purchase into the trough before any kernel revenue exists.
How to size your own requirement
Build it from physical facts, in this order, rather than from a template’s percentages.
- Annual RCN requirement. Target kernel output ÷ production KOR. At 22–24 kg of saleable kernel per 100 kg of RCN, a plant selling 700 tonnes of kernel a year needs roughly 3,000 tonnes of RCN.
- Months of cover you must finance. Not the months you would like to hold — the months between the close of your buying window and the exhaustion of the stock. For a single-origin plant running 300 days this is typically 8–9 months of average cover; for a plant sourcing two harvests it can be 3–4.
- Peak inventory value. Tonnes × landed RCN cost per tonne, at the price you actually paid, in the currency you paid it. Peak, not average — the facility has to cover the peak.
- Work in progress and finished goods. The cooling hold and drying cycle alone mean roughly a day of steamed nuts and half a day of wet kernel sit in the plant at all times. Finished goods sit until a container is booked and stuffed.
- Debtors. Entirely a function of your payment terms. A letter of credit at sight is a different business from 30 days after bill of lading.
- Three months of overheads before the first proceeds arrive. That is a floor, not a target.
- A cushion for adverse price movement on stock you have already bought.
Then finance it. A 25% margin from the promoter against 75% from the bank is the norm in the Indian model reports; your margin share is what carries into project cost.
What the ratio looks like at scale
Well-capitalised operators in this sector raise working-capital facilities that dwarf their plant costs. In 2025 Robust International — a Singapore-headquartered trader running processing plants in Burkina Faso, Côte d’Ivoire and Mozambique — upsized a syndicated working-capital facility from US$105 million to US$175 million, arranged by FMO with IFU at US$25 m, OeEB at US$25 m, FMO itself at US$15 m and British International Investment at US$5 m. The facility exists specifically to buy cashew and sesame directly from local cooperatives and farmers rather than exporting raw product to Asia for processing.
That is the shape of the business, at every scale. For your own planning, a first-time single-origin processor should expect a working-capital requirement somewhere between 1.5× and 4× the fixed-asset cost, depending on months of cover and the RCN price.
Anyone presenting you a model in which working capital is a fraction of capex has either found a year-round raw-material market, or has not understood the business.
What to do when the appraisal officer questions the line
Expect the challenge. A working-capital request larger than the capex reads as an error to anyone appraising against the standard template, and you will be asked to justify it.
Answer with the physical facts rather than the ratio: the length of your buying window, the tonnage you must hold to run 300 days, the landed cost per tonne you actually paid, and the date of your first expected proceeds. The number stops looking anomalous the moment it is presented as a consequence of the harvest calendar instead of a percentage of the project cost.