Could AI help Koltiva get fertiliser onto Riau’s palm farms when the rain comes?

Why Riau’s palm growers are putting off their fertiliser

Riau, a province on the east side of Sumatra, has more oil palm than any other part of Indonesia: about 3.4 million hectares in 2022 (a hectare is a square 100 metres on each side). [5] Much of it belongs to smallholders, families who farm a few hectares of their own. Across Indonesia, about 2.9 million such growers farm 6.9 million hectares, an average of about 2.4 hectares each. [5] They sell the fruit bunches their palms produce every two to three weeks, and the fruit is pressed into palm oil at a mill.

Mature palms need fertiliser to keep producing, usually two or three times a year. Government guidance says each round should go on when there is enough rain for the roots to take it up, for example in February–March and September–October. [6] In the dry season, the agriculture ministry tells growers to time fertiliser to the rain, using the national weather agency’s forecasts. [7]

Palm growers pay full shop prices for that fertiliser. The government’s subsidised fertiliser goes to food crops and a few others, and oil palm is expressly left out. [3] [2] In June 2026, a national price round-up listed urea, a nitrogen fertiliser, at 500,000 rupiah (Rp500,000) for a 50 kg sack, and potash, a potassium fertiliser, at Rp450,000, both up from the year before. [2] A farmer group in Bengkalis, in Riau, said its sack of mixed fertiliser had risen from Rp700,000 to Rp930,000. [4]

The bill is small next to what the fruit earns, but it comes first. Each tonne of fruit takes nutrients out of the soil. The urea and potash that put them back cost about Rp129,000 at June shop prices. That tonne sells for about Rp3.8 million, at the price Riau’s plantation office sets for growers whose groups have a written agreement to sell to a mill. [6] [8] [D] The trouble is timing: a grower has to pay for the sacks months before the extra fruit arrives.

On 12 November 2026, the head of Gapki, the association of Indonesia’s palm oil companies, said that farmers in several regions were putting off fertilising. He said they were not doing so by choice; economic conditions forced them. In the same speech he said El Niño, a recurring Pacific weather pattern that brings drought to Indonesia, was reducing the water available to palms, and that palm oil output could fall by up to 3 million tonnes, mostly in 2027. [1] So growers have two reasons to wait, and they look the same from outside: some have no cash, and some have soil that is still too dry.

Koltiva, and how a grower gets fertiliser through it today

This article follows one company that is in a position to act. Koltiva is an Indonesian technology company that records which farms a crop came from, for the companies that buy palm oil, coffee and cocoa. Its field agents have registered more than 170,000 palm smallholders, in 13 provinces including Riau. [11]

Koltiva also sells fertiliser to growers. Its agronomists (crop advisers) check the products, nearby farm-supply shops it has enrolled stock them, and growers can pay later, at harvest, through a digital wallet, a payment account, in Koltiva’s farmer app. [10] [11] Koltiva says it works with lenders it calls financial partners. As far as our research could establish, those lenders put up the money for pay-later orders and decide each loan themselves.

Here is one round of fertiliser for one registered grower, in the steps our research describes. The prices are national shop prices, not a real grower’s receipts. [2] [10] [11]

  1. Koltiva’s agronomist works out what this grower’s palms need, from soil and leaf tests and the standard dose tables. The agronomist also sets the fertiliser window: the weeks when the soil should be moist enough for the fertiliser to work. Koltiva sends that advice to the grower.
  2. The grower orders in the app. For one hectare of palms aged 9 to 13 years, the guidance dose is about 7 sacks of urea and 6 of potash a year. At June prices that costs about Rp6.2 million, before the other fertilisers the tables list. [6] [D] The grower pays cash or asks to pay later.
  3. For a pay-later order, a lender working with Koltiva decides whether to lend, and the grower picks a repayment schedule tied to the harvest.
  4. An enrolled shop hands over the sacks. Koltiva’s app records the sale.
  5. The grower’s household carries the sacks to the palms and spreads the fertiliser, when they judge the soil moist enough.
  6. The grower keeps selling fruit, often to a toke, a local collector who buys at the plot, and repays the loan from that income in the harvest season.

Two things go wrong along the way. A grower who is refused a loan, or never asks, pays cash, borrows from the toke, or waits. And after the shop hands over the sacks, Koltiva’s records go quiet, as far as our research could find. Nothing published shows whether the fertiliser went on, when, or why not. Koltiva’s only published figures on fertiliser lending come from 136 coffee farmers in South Sumatra, not palm growers. [10]

So the problem, put plainly, is this: Koltiva wants its palm growers to get genuine fertiliser on while the soil can use it, without the service losing money, but growers miss the moment either for lack of cash or because the soil is dry, and Koltiva cannot tell which.

Improve: ask growers one question, and understand the answer

The Improve idea changes one step: the reminder Koltiva sends when the fertiliser window opens. Today that step works like this:

  1. Koltiva’s agronomist decides when the window opens, and Koltiva sends each grower a reminder with the dose.
  2. The grower decides whether to apply. A grower without cash may ask for a loan, borrow from the toke or wait.
  3. As far as our research could find, Koltiva hears nothing back. It does not learn who applied, who waited, or why.

With the change, the same reminder asks for a reply:

  1. Koltiva sends the same reminder, ending with a short question: have you applied, partly applied, or not yet? If not yet, why? The grower can give one of four answers: no money, soil still dry, no fertiliser in the shop, or something else.
  2. The grower answers however suits them: by tapping a button, typing a message, or sending a voice note in their own words. Software turns a voice note into text, sorts each answer into one of those reasons, and asks one follow-up question when the answer is unclear. It keeps the grower’s original words.
  3. Koltiva’s field staff check unclear answers and call growers who cannot use the app. A grower who says “no money” is offered the existing pay-later route, and the lender still decides. A grower who says “no fertiliser” is passed to the shop. A grower who says “soil still dry” goes to the agronomist, not pressed to apply.
  4. Koltiva’s reports count the answers, so for the first time it sees how much waiting is about cash and how much about rain.

The step matters because the two reasons for waiting need opposite responses. If Koltiva lends to a grower whose soil is dry, the grower may waste the fertiliser and still owe the debt. If Koltiva tells a grower who has no cash to wait for rain, the grower misses the window.

What the AI does. It does two narrow jobs. It turns spoken replies into text, and it reads replies written in the grower’s own words and decides which of the four answers the grower gave. A chat model, the kind of AI behind chat assistants, keeps the exchange to that one question and passes anything else to Koltiva’s field staff. [A]

Related Business Opportunity:  Could Bali's Bamboo Buyers Help Pay for the Next Generation of Growers?

Why AI, and not just a form? A form with four buttons needs no AI, and Koltiva should try it first. AI earns its place only if many growers would rather type or speak, in Indonesian or a regional language, than tap. Ordinary software sends the message, stores the answer and passes it to the right team. Speech software transcribes Indonesian and other widely spoken languages reliably, but it makes many more mistakes in most Indonesian regional languages unless someone adapts it locally. Koltiva would have to test it on real replies. [A]

Step Today With Improve
Window reminder One-way message Dose and timing only Message plus one question Applied, partly, or not yet, and why
Learning why growers wait Not recorded No reply is asked for Grower’s own reply Tap, text or voice; software sorts the reason
Unclear answers Not applicable Field staff check They also call growers who cannot reply
Response to the reason Up to the grower They must ask for help Offered by Koltiva Loan route, shop, or agronomist
Loans, dose, applying Unchanged The lender decides loans, the agronomist the dose, the grower when to apply

Koltiva would run this itself, on its own app. It can change two things: how fast a grower who has no cash hears about a loan, and how much Koltiva knows. It cannot create cash, fill an empty shop or make the soil wet. If lenders or shops cannot respond before the window closes, a quicker answer helps nobody. The answers are also growers’ own accounts, not proof that fertiliser went on, and growers may shade them if they fear a “not yet” will count against them.

Reinvent: a service that applies the fertiliser when the rain comes

The Reinvent idea starts again from what Koltiva is trying to achieve: the right fertiliser on each grower’s palms in the useful window, paid for from later fruit. Today the work runs through each household, one order at a time:

  1. Koltiva sends advice; the grower orders when they have cash or credit.
  2. A lender approves the order; a shop hands over the sacks.
  3. The household carries and spreads the fertiliser when it judges the soil ready, and Koltiva does not see it.
  4. The grower repays from fruit income, whether or not the fertiliser ever went on.

In the new arrangement, Koltiva sells a finished job, fertiliser applied, rather than sacks:

  1. Before the season, a grower who wants the service signs up their plot. They agree the agronomist’s prescription for it, a disclosed price for the fertiliser, a fee for the work, the lender’s charge, and what happens if a round is put off.
  2. A lender sets aside money for the whole round. It pays out only for applications that Koltiva has checked and accepted, and it agrees in advance how much drought loss it will bear.
  3. Koltiva reserves fertiliser its agronomists have approved for each signed-up plot, and stores it at nearby enrolled shops.
  4. Software watches local rain gauges, soil-moisture readings and rainfall records, and forecasts when each group of nearby plots is likely to be ready, so Koltiva can book crews and move stock ahead of time.
  5. Koltiva’s field staff confirm on the ground that the soil is ready, and the agronomist gives the go-ahead. Planning software proposes which crew goes where, and a Koltiva coordinator, the staff member who runs the round, approves it.
  6. A trained crew carries the reserved sacks to the plot and applies the prescribed dose. It records the batch, the number of sacks, the plot and the time. The grower confirms the work or disputes it; a supervisor settles disputes. Checkers who work separately from the crews inspect a sample of plots.
  7. Only then does the lender pay for that application. The grower repays from later fruit sales, through the wallet, and stays free to sell to any buyer. If field staff find the soil too dry, they put the round off. Nothing is applied, and the grower owes nothing for it.
Step Today With Reinvent
What starts the work A grower’s order When they have cash or credit The soil being ready For all signed-up plots in an area
Who applies it The household When it judges the soil ready A trained crew Prescribed dose, recorded on the spot
What the loan pays for Sacks handed over Whether used or not Fertiliser applied Checked work only
What Koltiva knows A sale Nothing after the shop A record of the work Batch, plot, time, grower’s confirmation
Dry season Grower’s risk Debt even if fertiliser waits Round put off No fertiliser debt
Prescriptions, fruit sales Unchanged Agronomist prescribes, lender sets terms, grower sells to any buyer

Where AI does the work. Only in step 4. One piece of software spots when the rain has really set in, or when a dry spell has begun, in the readings for each area. A second forecasts when each area’s window will open, with a range rather than a single date. Both are known techniques, but both become less reliable when the weather changes pattern, as in an El Niño year, and when local records are short. [B]

That is the whole AI case: earlier warning, so stock and crews arrive on time. A simple rule using the weather agency’s forecasts could do as well, and if it does, the service may still be worth running without AI. Koltiva may also use AI to match a grower’s records across its systems where names and numbers differ; a Koltiva staff member decides before any debt is attached. [C]

What needs no AI. The prescriptions, the loan decisions, the crew schedule, the stock allocation, the payments and the fertilising itself. The crew schedule and the stock plan use ordinary planning software that has existed for decades. [C]

Why this is more than a set of improvements. Koltiva could improve each of today’s steps separately: give growers a loan limit before the season, send the dose early, and let shops offer paid visits to spread the fertiliser for households short of hands. Even then, each household would still order, borrow and book a visit on its own after the window opens, and the loan would still pay for sacks, not for fertiliser on the ground. Three things need the new arrangement: one date for a whole area set by the soil, a debt only for work actually done, and a record made by the person who did it.

What people still decide. Growers decide whether to join, let the crew onto their land and confirm the work. The agronomist sets the dose and releases each round. The lender decides how much money to offer and how much drought loss it will bear. The coordinator approves every crew plan.

The limits. The hard parts are not the AI. When rain returns, many plots open at once, so the bottleneck could move from the grower’s cash to the number of crews. In a dry season, reserved stock and booked crews sit idle, and Koltiva, the shops holding the stock or the crews must bear that cost. Our research found no sign that households lack the hands to spread fertiliser themselves, so growers may not want to pay a crew. The idea also needs a lender willing to pay out only for checked work.

Other routes already exist. The toke lends to growers against their fruit. Pupuk Indonesia, the state fertiliser maker, runs a programme that bundles fertiliser, advice and financing; it reported 451,537 hectares across all crops by the end of 2024. [12] Our earlier investigation on What Could Work (this publication) into helping Riau’s growers form groups that sell to mills at the official price found that collectors also lend growers small sums for farm costs, which ties them to one buyer. [14]

Related Business Opportunity:  Could a regional dairy plant supply fresh milk to Java's free school-meal kitchens?

Our earlier investigation on What Could Work into this same problem examined a different business. A new company would buy fertiliser in large quantities from distributors, deliver it to growers on credit, and have the mill subtract the repayment from what it pays for their fruit. It concluded that the idea needed further investigation, because nobody knew whether fertiliser bought that way would be cheap enough, or whether a mill would agree to subtract repayments. [13]

The Reinvent idea does not depend on either. It uses Koltiva’s growers, shops and lenders, earns a fee for work done rather than a profit on each sack sold, and collects through the wallet. In exchange, it adds the cost of crews, of storing reserved sacks at shops, and of the rain gauges Koltiva would install and maintain, and it still depends on growers repaying through a drought.

What each change could offer

Neither idea has been tried, so the table judges potential: how well each could do if it were built and run sensibly.

Favourable for the business Mixed Unfavourable

Potential Improve: Reminder reply Reinvent: Application service
Technical feasibility Moderate to highButtons need no AI; sorting typed replies into four fixed reasons is an established use. Voice notes in Indonesian regional languages are the uncertain part and must be tested. [A] ModerateThe AI part, spotting rain and forecasting the window, is known but less reliable in an El Niño year and with short local records. [B] The harder questions are crews, stock custody and a lender that pays on checked work.
Economics Unpriced, likely smallCheap to add to an existing message, but no cost or saving is recorded, and a plain set of buttons may deliver most of the benefit. It earns money only if more growers get loans in time. [D] Unknown, heavy setupCrews, staging, rain gauges, checks and loan losses are all unpriced. For a grower, each Rp1 million of fee and loan charge per hectare needs about 260 kg more fruit a year. [8] [D]
Customer value IndirectGrowers short of cash can be offered a loan sooner, and dry-soil growers are not pushed to apply. It creates no cash and does not lower prices. High, if growers delegateFertiliser goes on when the soil is ready, with no cash up front and no debt for a cancelled round. Growers must accept a crew on their land and pay for the work.
Competitive positioning Easy for KoltivaKoltiva can add it to its own app quickly. For the same reason it gives no lasting edge, and leaves little room for a company selling it. ModerateKoltiva's registered growers, enrolled shops, agronomists and records are hard to copy. Toke advances and Pupuk Indonesia's programme remain alternatives. [12] [14]
Risks Low to moderateA misread reason could push fertiliser into dry soil or pressure a household, so staff confirm each referral. Fear of penalties can bias answers; staff may be overloaded. HighCrews may be too few when many plots open at once; idle stock and crews in dry seasons; drought losses across a whole area; disputes over work done on growers' land.
How it would be tested A short trial with one grower group, comparing today's follow-up, buttons with recorded prompts and staff calls, and the AI reply on usable answers, cash-versus-dry confusion, staff time and loans made inside the window. A forecast test on past rainfall against a simple rule, then one season on about 100 hectares of signed-up plots beside the improved current service, measuring plots fertilised in time, cash paid up front, crew days, loan losses and Koltiva's margin per hectare.
Our assessment of potential: how well each could do if it were built and run sensibly. Ratings are editorial judgments built from the research and from what our AI Capability Atlas says about each kind of AI; none is a measured result. The colour shows whether a rating helps or hurts the business case, so high risks show red. The last row says how each idea would be tested; it is not rated.

Which is worth pursuing?

For each idea we ask: if the arrangement can be made to work and the AI part is practical, would it create enough value to justify the change?

Improve: potentially yes, as a cheap way to learn

The AI part is practical for typed replies and uncertain for voice notes in regional languages. The change is cheap and easy to undo. Its main value is that Koltiva would learn, for the first time, why its palm growers wait, and could offer a loan while the window is still open. It cannot make fertiliser cheaper or create cash. Koltiva is best placed to do it, on its own app.

It is worth doing on four conditions: Koltiva does not already collect these answers; lenders and shops can respond inside the window; growers trust that “not yet” carries no penalty; and AI gets more usable answers than a plain set of buttons.

Reinvent: potentially yes, and it is the change that could actually get fertiliser on in time

Suppose growers let a crew work their land, and a lender pays only for checked work. Then the service answers Koltiva’s question directly: fertiliser goes on when the soil is ready, growers pay nothing up front, and a cancelled round leaves no debt. Its AI is a small, uncertain part. Koltiva is best placed, because it already has the registered growers, enrolled shops, agronomists and lenders. A newcomer could take part by running the crews under contract for Koltiva, but would struggle to run the whole service without Koltiva’s growers and lenders. It is worth the change on five conditions:

  • a real share of missed windows comes from cash, ordering or labour, not only dry soil;
  • growers accept a crew on their land and a fee for the work;
  • a lender will pay out only on checked applications and carry drought losses up to a limit;
  • enough crews are available when many plots open at once after rain;
  • the fee covers the crews, staging, checks and idle days, at a price growers accept.

Worth pursuing: both, in order. The one-question reply is cheap, and it answers the question the bigger idea stands on: how much of the waiting is about cash and how much about rain. If cash and labour turn out to matter, the application service is the only idea here that could get fertiliser onto the palms on time, and it deserves a one-season trial. If dry soil explains most of the waiting, Koltiva keeps the reply and skips the service.

What would have to be tested

For Improve, Koltiva would run a short trial with one group of palm growers. It would compare three ways of asking: Koltiva’s current follow-up, if it has one; a set of buttons with recorded voice prompts, plus staff calls; and the AI reply path. It would count how many growers give a usable answer and how often the software confuses “no money” with “soil still dry”. It would also measure the staff time each answer costs, and whether growers who need a loan get one before the window closes. Koltiva’s field staff would visit a sample of plots to see how far the answers match what actually happened.

For Reinvent, Koltiva would first test the forecasts against several seasons of past rainfall, beside a simple rule. Then it would run one season on about 100 hectares of signed-up plots, beside growers offered today’s service with a loan limit before the season, early dose advice and paid application visits from shops.

The trial would measure the share of plots fertilised inside the window, cash paid by growers before harvest, crew days used and idle, disputes, loan losses, and Koltiva’s margin per hectare. The idea fails if it gets fertiliser on no faster than that improved service, if losses exceed the lender’s limit, or if the fee does not cover the cost.

Before either, the missing evidence:

  • How much waiting is about cash and how much about dry soil. No record splits them. [1] This decides whether Reinvent has a problem to solve.
  • How many Riau palm growers Koltiva serves with fertiliser loans, and how they repay. Koltiva publishes only a coffee example. [10] This decides the scale of both ideas.
  • Whether Koltiva already asks growers if they applied. If it does, Improve adds little.
  • What crews, staging and rain gauges would cost per hectare, and what fee growers would pay. None is recorded. This decides the Reinvent economics.
  • How much fruit a timely application protects. At about Rp3,800 a kilo, each Rp1 million a year of fee and loan charge per hectare needs about 260 kg more fruit per hectare a year to pay for itself. [8] [D]
  • How often growers fail to repay in a drought year. No palm figures exist. This decides whether a lender will join.
Related Business Opportunity:  What if poultry farmers could see what cleaning chick house leaves behind?

Research Notes

What this investigation is based on

What Could Work’s stored research on fertiliser for Riau’s independent oil-palm growers (the sources below), and our own analysis of how Koltiva’s fertiliser service works and where AI could change it. No interviews or pilots were carried out. Koltiva’s descriptions of its service are the company’s own.

Whose problem this article takes, and why: the research names several parties who could act, including the growers themselves, a partner mill, Pupuk Indonesia’s programme and Koltiva. The growers carry the loss but, as households, lack the money, staff and pooled records to change the service. A mill buys the fruit but does not run the fertiliser purchase. Koltiva and Pupuk Indonesia’s programme both run the work, hold records and reach many growers; we follow Koltiva because its farmer app, enrolled shops and pay-later service already connect every step of the process.

Sources

  1. Katadata, Petani Sawit Tunda Pemupukan, Produksi 2027 Terancam Turun, 12 November 2026. The Gapki chair on farmers deferring fertiliser for economic reasons, El Niño dryness, and output risk of up to 3 million tonnes, mostly in 2027. No count of farmers or regions.
  2. APGtimes, Daftar Harga Pupuk Subsidi dan Non Subsidi Juni 2026, 12 June 2026. Non-subsidised urea Rp500,000, potash Rp450,000, NPK Rp750,000–850,000 per sack; palm and rubber growers buy non-subsidised. A national list, not a Riau survey.
  3. Tabloid Sinar Tani, Permentan 15/2025 resmi berlaku, 11 June 2025. The crops eligible for subsidised fertiliser; oil palm is not among them.
  4. RRI Bengkalis, Harga Pupuk NPK Naik 20 Persen, June 2026. One farmer group’s report; crop not stated.
  5. Republika, Sebaran sawit nasional terkonsentrasi di enam provinsi utama, 3 December 2025, citing the plantation directorate’s 2022 figures: 6.94 million hectares of smallholder palm, about 2.9 million farmers; Riau about 3.39 million hectares of all palm.
  6. BRMP Babel (agriculture ministry), Penerapan standar pemupukan kelapa sawit. Nutrients removed per tonne of fruit, dose per tree by age, two to three applications a year timed to rainfall. Provincial extension guidance.
  7. Plantation directorate, agriculture ministry, mitigasi kemarau kelapa sawit, July 2026. Adjust fertiliser timing to rainfall; use the weather agency’s climate information and early warnings. No thresholds given.
  8. InfoPublik, Harga TBS Riau, 30 September–6 October 2026. Riau plantation office price for partnered growers: Rp3,826.84 a kilo for palms aged 10–20. Growers selling to a toke usually receive less.
  9. detik, Disbun tetapkan harga tinggi, harga sawit di tingkat petani justru rendah, 27 October 2022. Growers outside partnerships sell to toke at the toke’s price (Rp2,175 against an official Rp2,660 near Pekanbaru, in 2022).
  10. Koltiva, Why closed-loop financial inclusion is the missing link in smallholder input access, May 2026. Checked inputs, enrolled shops, pay-later repaid at harvest, crop advice with soil and leaf tests; 41,200 kg of fertiliser to 136 coffee farmers. The company’s own account; no fees, repayment or cost figures.
  11. Koltiva, Ensuring sustainability and transparency in Indonesia’s palm oil supply chain, August 2024. More than 170,000 palm smallholders registered; farmer app with pay-later input financing; palm work in 13 provinces including Riau. The company’s own account; does not say how many palm growers borrow.
  12. Pupuk Indonesia, Program Makmur, 21 January 2025. 451,537 hectares across all crops by December 2024. The producer’s own account.
  13. Our earlier investigation on What Could Work: Could Riau’s oil-palm growers buy fertiliser now and pay for it from their next harvest? (17 November 2026). A bulk distributor-lender repaid at the mill; opportunity requiring further investigation.
  14. Our earlier investigation on What Could Work: Could AI help Riau’s small oil-palm growers into the groups mills must pay the set price? (18 November 2026). Collectors lend growers small sums for farm costs, which ties them to one buyer.

Our explanations and assumptions

Where a note discusses a kind of AI, it draws on our AI Capability Atlas, What Could Work’s own reference on what today’s AI can do. For each kind of task it records what goes in and what comes out, how proven the approach is, its typical failures and how to check for them. Established means proven methods exist for a narrow, well-defined task. Conditional means it can work, but reliability depends heavily on the setting and has to be tested there. Exploratory means it is still unproven beyond experiments. The Atlas describes AI in general; nothing in this article has been tested in the business it describes.

  1. Understanding growers’ replies. Three kinds of AI are involved. Speech recognition turns a voice note into text. A classifier reads the text and picks which fixed reason the grower gave. A conversational model asks one follow-up question when the answer is unclear and hands anything else to a person. A fourth, text-to-speech, can read the question aloud, and translation software can help prepare the question in other languages, with a qualified person reviewing it. Our AI Capability Atlas rates sorting text into a small, stable set of categories as established, and conditional when meaning depends on context; it only captures what the grower said, not hidden reasons. It rates narrow, structured conversations as established, and warns that a fluent exchange can earn more trust than it deserves. It rates speech recognition as established for well-supported languages and good audio, but conditional for noisy audio and for most Indonesian regional languages without local adaptation; it also warns that transcription can invent words during pauses or silence. Translation and spoken output are established for common languages and conditional for most regional ones.

    How this shaped the article: it is why Improve starts with plain buttons, why AI must beat them to earn its cost, why silence never counts as an answer, and why field staff check unclear replies. Its technical feasibility is rated moderate to high, not high, because of voice notes in regional languages.
  2. Spotting the rain and forecasting the window. One kind of AI detects defined events in a stream of readings, here the real onset of rain or the start of a dry spell. Another forecasts future values from past records, here when each area’s soil will be ready, with a range. The Atlas rates event detection as established for well-defined signals and conditional when conditions change pattern; it warns that a seasonal swing can look like an event. It rates forecasting as established for recurring patterns and conditional when records are short or conditions change, warns that a single forecast date hides its uncertainty, and says a forecast must be checked against a simple rule on past data.

    How this shaped the article: it is why the AI part of Reinvent is called small and uncertain, why field staff and the agronomist confirm every round, and why the forecasts are tested against a simple rule before the trial uses them. If the rule does as well, the service can run without AI.
  3. Planning crews and stock, and matching records. Scheduling crews and allocating stock to shops are long-standing planning mathematics: proven planning software, not AI. The Atlas rates both as established for clearly specified problems and conditional when conditions keep changing, and warns that forecast errors can outweigh what good planning gains. Matching records that refer to the same grower or plot across different systems is a kind of AI the Atlas rates established when strong identifiers exist and conditional when records are sparse or ambiguous; wrong matches can attach a debt to the wrong person.

    How this shaped the article: it is why we say the schedule and stock plan need no AI, why a coordinator approves every crew plan, and why a person resolves any uncertain match before a debt is recorded.
  4. The arithmetic.
    • Per tonne of fruit, the guidance lists 6.3 kg of urea and 7.3 kg of potash removed from the soil. [6] At Rp10,000 a kg (Rp500,000 ÷ 50 kg) and Rp9,000 a kg (Rp450,000 ÷ 50 kg) [2]: 6.3 × 10,000 + 7.3 × 9,000 = Rp63,000 + Rp65,700 ≈ Rp129,000. One tonne at Rp3,826.84 a kg [8] ≈ Rp3.83 million, so the fertiliser is about 3.4% of the fruit’s value. Phosphate and magnesium fertilisers are not counted.
    • Per hectare a year, at the guidance dose for palms aged 9 to 13 (2.5 kg urea and 2.25 kg potash per tree) [6] and an assumed 136 palms a hectare: urea 2.5 × 136 = 340 kg = 6.8 sacks; potash 2.25 × 136 = 306 kg ≈ 6.1 sacks. Cost: 6.8 × Rp500,000 + 6.1 × Rp450,000 = Rp3.4 million + Rp2.75 million ≈ Rp6.15 million. Planting density is our assumption.
    • Average smallholding: 6.94 million ha ÷ 2.9 million farmers ≈ 2.4 ha (national, 2022). [5]
    • Trial size, for illustration: 100 ha of signed-up plots, two applications a year = 200 hectare-applications, about 1,290 sacks of urea and potash, and about Rp614.5 million of fertiliser financed a year (680 × Rp500,000 + 610 × Rp450,000). The 100 ha is the analysis’s design choice, not a source figure. The cost of crews, transport, staging, tests, rain gauges, coordination, checks, insurance, the lender’s charge and loan losses is unknown; no figure is assumed.
    • Break-even for Koltiva: the fee per checked hectare-application must exceed its cost per hectare of crews, transport, staging, tests, observation, coordination, checks and insurance, plus its share of idle reservations, cancelled rounds and losses. The fertiliser itself passes through to growers.
    • Break-even for a grower: Rp1,000,000 ÷ Rp3,827 a kg ≈ 261 kg more fruit per hectare a year for each Rp1 million of fee and loan charge per hectare, and more for growers paid the lower toke price. [9] The yield gain from timely application is not recorded.
    • Improve: none of its costs (integration, language review, message delivery, transcription, staff calls) or savings is priced in the records. It pays if the staff time saved and the margin on extra loans made in time exceed those costs, and if it beats a plain set of buttons.
  5. Other ideas we considered. For Improve, we also examined software that prepares each grower’s loan file for the lender from Koltiva’s records and paper statements, and a photo check of each sack’s batch code against the supplier’s dispatch record at the shop. We chose the reply because it addresses the step that most decides whether fertiliser goes on in time, and because it settles the question both ideas depend on; nobody has shown that loan paperwork delays orders, and a photo of a label cannot prove what is inside the sack. For Reinvent, we considered a standing fertiliser account topped up by fruit sales; on inspection it was a set of improvements to today’s steps rather than a new arrangement.
  6. General explanations. Riau’s location, the description of smallholders and toke, and both “today / with the change” tables are our summaries of the recorded process and sources.

Evidence status of the main claims

Claim Status
Oil palm is excluded from subsidised fertiliser Observed [3] [2]
Shop prices: urea Rp500,000, potash Rp450,000 a sack (June 2026) Observed [2]
Farmers in several regions are deferring fertiliser for economic reasons Reported by an industry association [1]
Koltiva offers checked fertiliser, enrolled shops and pay-later at harvest Reported by the company [10] [11]
Koltiva does not know whether or why palm growers apply Inferred from the records
Cash, not only dry soil, causes a real share of missed windows Unknown
A reply question gets usable answers and quicker loans Hypothesis
An application service fertilises more plots on time and pays Hypothesis

Open questions

  • How many palm growers in Riau use Koltiva’s fertiliser loans, and how do they repay? Koltiva could answer.
  • How much of the deferral Gapki describes is cash and how much dry soil? A short survey of registered growers could answer.
  • Will a lender pay out only for checked applications, and carry drought losses up to a limit? Koltiva’s lending partners could answer.
  • What would a trained crew cost per hectare in Riau, and would growers pay for one? Local contractors and grower groups could answer.
  • How much fruit does an on-time application protect compared with a late one? Agronomists with plot records could answer.
  • How do Pupuk Indonesia’s programme and the toke respond if growers switch? Unknown.

Comments

Leave a Reply

Your email address will not be published. Required fields are marked *