Almost everyone builds the merchant model, usually without deciding to. It is worth deciding.
Model A — Merchant processor
You buy RCN, shell it, and sell kernel on your own account, taking the whole spread and the whole risk.
Working capital is 100% of a season’s raw material, bought inside three to four months, plus at least three months of overheads before the first kernel receipt. Build-out runs 12–18 months for a 5,000 t/year semi-mechanised plant, with real processing starting in the second year. You fund a year of cost with no revenue, then a year of raw material, then wait for a letter of credit.
The failure mode runs in both directions. In 2018 RCN fell 50–60% from peak while kernel fell only 20–25%, producing defaults and forced renegotiation across the trade. In 2025 the reverse: RCN import prices rose sharply while kernel did not follow, and processors reported heavy losses with some filing for insolvency.
Being long RCN and short kernel is an unhedgeable position for most origin processors. That is the model’s defining characteristic, not an incidental risk.
Model B — Toll / custom processor
You process someone else’s nuts for a fee. The trader keeps title; you never own RCN.
Commodity price risk falls to near zero, raw-material working capital falls to near zero, and revenue becomes a conversion fee times volume. You give up the spread and gain survivability — plus a business financeable on plant and a contract rather than on inventory.
No published toll processing rate exists anywhere. Not in the development literature, not in the trade press. Anyone quoting you a market rate is quoting their own book.
Derive an indicative fee from your own conversion-cost stack plus a 15–25% return. Because conversion cost is a range set by automation and wage level, so is the fee:
| Basis | Conversion cost | Toll fee, +15–25% | Per kg kernel at 23% KOR |
|---|---|---|---|
| Semi-automatic, Asian wage levels | ~US$155/t RCN | US$180–195/t | ~US$0.78–0.85 |
| Manual, African wage levels (2026-restated) | US$559–579/t RCN | US$645–725/t | ~US$2.80–3.15 |
Two adjustments before you quote from either. Export and logistics normally stay with the owner of the goods under a toll arrangement, so strike that line from your basis. And financing built on raw material a toll processor never owns should come out too — leave it in and you have quoted conservatively in your own favour, which is the safer error.
The other four, in brief
| Model | Capital | Price risk | Margin | Time to first revenue |
|---|---|---|---|---|
| C Grower-processor | Highest of all — orchard, capex, plus RCN for the balance | High, plus agronomic and weather risk | Full spread plus farmgate margin on own nuts | Orchard 3–5 years to bear |
| D Brand / retail packer | Capex, kernel stock, listing and marketing spend | Medium — you buy kernels, not RCN | The retail gap: export price is 20–30% of retail | 18–24 months to a listing |
| E Acquire idle plant | Lower capex, unknown refurbishment | As A | As A | Fastest route to a buying season |
| F Satellite network | Lowest capital per tonne | As A | As A, less transport and coordination cost | Fast |
Model E deserves more attention than it usually gets. In most producing countries there are idle or underused plants, and buying one is frequently the fastest and cheapest route to a first buying season — provided the diagnosis of why it went idle is honest. If it stopped for want of working capital rather than for want of machinery, you are buying the same problem with a building attached.
Why farm-to-retail fails at year two
The integrated plan does not fail in year one. Year one flatters it: the orchard is planted, the plant is commissioned, the first container ships, and the brand deck looks credible.
Year two is when four cash cycles collide.
- The orchard needs three to five years, and spending in every one of them
- The factory needs a full season’s RCN paid for inside ninety days
- The brand needs listing fees, artwork, marketing — against 60-to-90-day retail payment terms
- The kernel book needs forward sales matched to RCN tranches to survive a price move
All four draw on one account, and three peak in the same quarter as the harvest.
Then the management arithmetic. Five functions are already described as inevitable for a processing factory to succeed: production, technology and equipment, human resources, food safety, and finance and administration. Farming adds agronomy and a second workforce. Retail adds brand management, key accounts and packaging development. You are now hiring eight senior functions in a location chosen for its proximity to nuts.
One leg then starves the others, and it is almost always the RCN buying budget — because that is the only cost you can defer by simply buying fewer nuts. Which halves your utilisation, strands your fixed cost, and puts you among the plants running at a fraction of nameplate.
Sequence, do not stack. Do one thing at a time, and add the next only when the first funds itself.