Could AI help PLN get more usable plant-waste fuel from its many small suppliers?

PLN’s fuel buyer, and the plant waste it cannot get enough of

Indonesia’s mills throw off huge amounts of plant waste: sawdust from sawmills, husk from rice mills, shell from palm-oil mills, offcuts from plywood plants. All of it burns. A power station built for coal can burn some of it mixed in with the coal, a practice called cofiring. It cuts coal use and gives the waste a buyer.

PLN, the state electricity company, does this at its coal power stations. One company buys all of the fuel for them: PLN EPI, PLN’s fuel-supply arm. In 2025 it delivered 2.352 million tonnes of plant waste to 52 power stations, and it says they could take 8 to 10 million tonnes a year.[1] Its target for 2026 is 3.65 million tonnes. By late October it had delivered 1.842 million.[1][4]

Yet in the same year PLN EPI planned to export about a million tonnes of feedstock, described as surplus that its power stations cannot yet take.[2] It is short of what its plants burn while selling abroad what they will not. The source does not say which material is exported or why the plants cannot take it; the research points to quality and boiler limits. This article follows PLN EPI and asks how AI could help it get more fuel its plants can actually burn.

The fuel comes from almost 200 suppliers, and about 80% of them are small businesses.[5] A few are larger processors: one supplies nine PLN power stations from its own plant.[7] Sawdust made up more than half of all deliveries between 2023 and 2026.[4]

Two things make the job hard. The first is quality. A boiler is designed for fuel of known heat and dryness. Wet fuel gives less heat, and some plant waste carries alkali, mineral salts that can damage a boiler. Indonesia’s Ombudsman, the state body that investigates public services, found the supply sporadic and seasonal, with high moisture and alkali that put boilers at risk.[10]

The second is price. PLN EPI cannot simply pay more for better fuel. An energy ministry rule sets what it pays: 1.2 times a government reference price for coal, scaled down for fuel that carries less heat than coal, plus transport.[3] That averages about Rp700,000 (about US$40) a tonne. Exporters pay Rp1.0 to 1.3 million a tonne for palm shell.[2]

One truckload, from sawmill to power station

Here is how a load of sawdust reaches a PLN power station today, as What Could Work’s stored research on this market describes it.

  1. A small supplier registers with PLN EPI, which also runs a digital marketplace for buying fuel.[4]
  2. The supplier buys sawdust from several sawmills, often wet, and more or less wet with the seasons.[10]
  3. Some loads are dried or screened first, by a larger processor or at one of PLN EPI’s two processing sites, at Tasikmalaya and Ciamis in West Java, each handling up to 10 tonnes an hour.[4] The research infers that many small suppliers’ loads go as they are; no figures are published.
  4. A truck hauls the load. The price rule adds transport on top of the fuel price.[3]
  5. The power station samples and tests the load for heat, moisture and alkali, then accepts or rejects it. How often loads are rejected is not published.
  6. PLN EPI pays by the rule above. A supplier can borrow against its PLN contract from a state bank while it waits.[3]
  7. The plant burns the fuel with coal. When biomass is short or poor, it burns more coal.

PLN EPI is already changing parts of this. It plans 15 of its own processing hubs by 2028, and it is building digital records to trace fuel from source to plant.[4][6]

The problem in one sentence: PLN EPI needs fuel its plants can burn, but it buys most of it from small firms selling wet, uneven waste they cannot afford to dry or test, and a fixed price rule stops it paying more for better fuel.

Our earlier investigation on What Could Work looked at a different buyer: factories that might burn plant waste instead of coal.[13] This article stays with PLN’s fuel buyer, because it is the party that can act on the problem: it sets the terms, holds the records of every load and reaches every supplier.

Improve: plan each month’s deliveries from a forecast of how wet each supplier’s fuel will be

The step this targets is PLN EPI’s delivery planning: deciding which supplier sends how much fuel to which power station, and which loads go through a processing site first. It matters because the plants are not all alike. A load that one boiler rejects may be fine for another, and a load that is too wet for any of them is better dried before it travels. Today the quality of a load becomes known when it is tested at the plant.

What the AI would do. AI uses PLN EPI’s own records of past loads. For each supplier and kind of waste, it forecasts how wet the fuel is likely to be in the coming weeks, and how much there will be. It takes the season and rainfall into account and gives an honest margin of error. It also learns from each plant’s record which fuel it accepted and burned. A planning tool then proposes the month’s plan: drier fuel to the strictest boilers, likely-wet supply through a processing site, and drying capacity kept free for the rainy months. PLN EPI’s planners approve or change it.[A]

Related Business Opportunity:  Could a company pool Java's small farm plots into land blocks for PLN's solar plants?

Why AI, and not just a spreadsheet? Forecasting how wet fuel from 200 suppliers will be, from uneven records and the weather, is the kind of work AI forecasting does well when the patterns recur each year. It does worse when records are thin or conditions change. It may also prove unnecessary: last year’s monthly averages for each supplier might forecast just as well, and the trial has to check that.[A]

What needs no AI. Matching forecast fuel to plants is ordinary planning mathematics. The tests at the plant, the price rule, the contracts and each boiler’s limits stay exactly as they are.

Step Today With the Improve change
Knowing a load’s quality At the plant gate Tested on arrival Forecast weeks ahead Then tested on arrival as now
Sending fuel to plants Not published Quality found at the plant gate By forecast fit Drier fuel to the strictest boilers
Using processing sites Two sites Up to 10 tonnes an hour each Planned ahead Capacity kept for likely-wet supply
Who decides PLN EPI planners PLN EPI planners Approve or change each plan
Price, contracts, tests, boiler limits Unchanged
Our description of the proposed change, built from the process the research records. How PLN EPI plans deliveries today is not published.

Who would use it: PLN EPI itself, or a company selling it the forecasting. What it could change: more of the fuel that arrives fits the plant that receives it, so fewer loads are rejected and the plants can take more. What it cannot change: it makes no fuel drier. If most small suppliers simply cannot dry their waste, a better plan only moves the wet fuel around, and the limit becomes how much the processing sites can dry. It does nothing about the price rule or exporters.

Reinvent: buy raw waste at a hub, paying only for its dry weight

Start again from what PLN EPI actually needs. It does not need 200 small firms to deliver finished fuel to 52 power stations. It needs fuel within each boiler’s limits, every month, from waste that arrives wet. The Reinvent idea moves the purchase to where the waste is.

A network of hubs near the sawmills and rice mills buys raw waste from small suppliers. On arrival, each load’s moisture and type are measured, and the supplier is paid for the dry material in it. The water is not paid for. The hub then dries, screens and blends the waste to the specification of the power station it will go to, and dispatches it when that plant needs it. The hubs could be PLN EPI’s own or run for it under contract by processors.

How the work would change

Step Today With the Reinvent change
What PLN EPI buys Finished fuel Delivered to the plant, by the tonne Dry material in raw waste Bought at a nearby hub
Where quality is measured At the plant gate After the haul At hub intake Before any drying or haul
Who dries and tests Each supplier If it can afford to The hub For all its suppliers
What starts a delivery A supplier’s offer A plant’s forecast need Fuel graded to its boiler
What PLN EPI asks of small suppliers Fuel that passes the plant test They carry a rejected load Raw waste brought to a hub Paid on arrival for dry weight
Price rule, boiler limits, plant tests Unchanged
Our description of the proposed redesign. It reaches beyond the process the research mapped, into how small suppliers are paid.

Where AI does the work

AI is not what makes this possible; moving the purchase is. AI makes two parts of it workable at scale.

  • Measuring every load on arrival. At the weighbridge, AI turns the readings of calibrated moisture sensors into an estimate of each load’s wetness, with a margin of error, and a camera identifies the kind of waste. Because suppliers are paid on this estimate, a share of loads is still tested in a laboratory to check it. Any load the system is unsure of, and any load a supplier disputes, goes to the laboratory.[B]
  • Planning the hubs. AI forecasts how much waste each area will bring in through the seasons and what each plant will need. A scheduling tool then proposes what to dry, hold or send, for the hub’s planner to approve.[C]

The price rule, the payment arithmetic, the laboratory checks, the plants’ tests and each boiler’s limits need no AI. People still set the specifications, approve plans and settle disputes.

Why it is more than a set of improvements. A better plan, as in the Improve change, still leaves quality with small firms that cannot dry or test it. Paying for dry material at a hub moves that job to the party with the dryer and the laboratory.

It also stops water travelling. The price rule adds transport to what PLN pays,[3] so a tonne of water hauled is freight PLN pays for. At legal loads, freight for one tonne over 100 kilometres is about Rp293,000.[11][D] How PLN’s transport allowance is calculated is not published. No change to a single step can do these things; they need the purchase itself to move.

The limits. It needs capital. PLN EPI already plans 15 hubs by 2028.[4] For scale, one listed processor planned to spend about Rp50 billion on a plant making 60,000 tonnes a year.[9] What raw waste costs, and what drying costs a tonne, are not published. Paying on a measurement invites disputes unless sellers trust it. And it does not fix the price rule: exporters can still outbid PLN for some waste.[2]

It also changes PLN EPI’s relationships with its suppliers. The hub, not the small supplier, carries the drying cost and the risk of having paid too much for wet waste. Larger processors that supply PLN today could run hubs under contract, or find PLN EPI competing with them. Small suppliers would earn a collector’s margin rather than a supplier’s, and some may stop selling to PLN.

Related Business Opportunity:  What if a school’s paint order came with a lead check?

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, and why.

Favourable for the business Mixed Unfavourable

Potential Improve: delivery planning from a wetness forecast Reinvent: hubs that buy dry weight
Technical feasibility ModerateForecasting recurring seasonal patterns and planning against them are established kinds of work; whether PLN EPI's records of past loads are complete enough is unknown, and last year's averages may do as well. [A] ModerateEstimating moisture from calibrated sensors and recognising the kind of waste are established in fixed set-ups, but paying on the estimate needs calibration for each kind of waste and laboratory checks, and the whole hub chain is untested. [B][C]
Economics Low cost, uncertain gainIt runs on records PLN EPI already holds; what it gains depends on how often loads are rejected or misdirected, which is unpublished. Capital-heavy, margin unknownHubs need capital (one processor planned Rp50 billion for 60,000 tonnes a year) and drying costs are unpublished; it stops PLN paying to haul water, and PLN EPI already plans 15 hubs. [9][D]
Customer value ModerateMore of the fuel that arrives fits the plant that receives it, but no fuel becomes drier. High, if it worksPlants receive fuel graded to their boilers all year, and small suppliers are paid for dry material without needing dryers or laboratories.
Competitive positioning Easy for PLN EPIPLN EPI holds the records and makes the plans, so it can adopt this alone; that leaves little room for a company built on selling it beyond analytics vendors. Strong for PLN EPI; mixed for othersAs the single buyer already building hubs, PLN EPI can lead; processors could run hubs under contract or find PLN EPI competing with them. [4]
Risks LowEvery test, contract and boiler limit stays and planners decide; the main risk is trusting a forecast that is no better than averages, and it is easy to stop. HighPayment rests on a measurement that sellers may dispute; capital is committed before the economics are known; small traders lose part of their margin; exporters can still outbid the price rule. [2]
How it would be tested Replay the forecast on PLN EPI's past records against last year's averages, then run it for one wet season beside the normal plan without using it. One wet season at an existing PLN EPI site: some small suppliers paid on measured dry weight at intake, similar suppliers delivering as today, with random laboratory checks of the measurement.
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 involved is practical, would it create enough value to justify the change?

Improve: potentially yes, as a cheap first step

It runs on records PLN EPI already holds and leaves every safeguard in place, so it is cheap to try and easy to stop. It is best placed with PLN EPI itself. It is worth doing if its forecasts beat last year’s averages and if misdirected or rejected loads are a real share of deliveries. It cannot make the fuel any drier.

Reinvent: potentially yes, and it goes after the cause

It takes quality off the small firms that cannot manage it and puts it with a party that can, and it stops PLN paying to haul water. PLN EPI is best placed to lead it, with processors or new firms running hubs under contract. It is worth the change only if:

  • the measurement at intake is accurate enough to pay on, and suppliers accept it;
  • a tonne of fuel made at a hub, counting the dry material bought, the drying and the haulage, costs less than the price rule pays for a tonne delivered;
  • the hubs are close enough to the mills that small suppliers can reach them cheaply;
  • the capital for the hubs is spent anyway, as PLN EPI’s plan for 15 suggests.

Worth pursuing: both, in that order. The forecast is cheap, works on today’s records, and is exactly the planning a hub network will need. The hub idea goes after the reason small suppliers fail, and it can be tested at the two sites PLN EPI already runs before any new hub is built. Neither will stop exporters outbidding a capped price.

What would have to be tested

For Improve, first replay the forecast on PLN EPI’s past records: would it have predicted which loads were wet, better than last year’s averages? Then run it for one wet season beside the normal plan, without using it. Each month, compare the plan the planners used with the plan the forecast proposed, and check which would have sent fewer loads that were then rejected.

For Reinvent, run one wet season at one of PLN EPI’s existing sites. A group of small suppliers is paid on measured dry weight at intake, while similar suppliers deliver as today. Check the intake measurement against laboratory tests on loads chosen at random. Compare the fuel each group ends up delivering to plants, the cost per tonne delivered, rejections and whether suppliers stay. Stop if measurement disputes swamp payments, or if dry material, drying and haulage cost more than the price rule pays.

The missing evidence, and what each piece decides:

  • How often plants reject loads, and why. This decides how much either idea can gain.
  • How PLN EPI plans deliveries today, and how complete its records of past loads are. This decides whether the Improve forecast can be built.
  • What small suppliers pay for raw waste, by type and season. This decides the hub’s purchase price.
  • What drying costs a tonne at a hub. This decides the hub’s margin under the price rule.
  • How PLN’s transport allowance is calculated. This decides how much PLN saves when water stops travelling.
  • How accurately moisture can be measured on a load at a weighbridge. This decides whether suppliers can be paid on it.
Related Business Opportunity:  Could AI help Meratus fill its new container ships to Mamuju?

Research Notes

What this investigation is based on

What Could Work’s stored research on biomass fuel for PLN’s power stations (the sources below) and our own analysis of how the work is done and where AI could change it. No interviews, invoices, plant records or pilots were used. This article replaces a first version published on 18 November 2026, which compared ideas for two different buyers.

Whose problem this article takes, and why. Several parties share this problem: the small suppliers, the larger processors, PLN EPI and factories that might burn plant waste. We follow the one that can act on a change. PLN EPI decides how fuel is bought and tested, can fund a change, holds the records of every load and reaches all of its suppliers. The larger processors can act too, but only for themselves. The small suppliers carry much of the problem but lack the money, records and staff to change how the work is done, so they appear here as the firms PLN EPI’s changes would serve. Factories are a different market, covered in our earlier investigation.

Sources

  1. ANTARA: PLN EPI bidik pemanfaatan biomassa 10 juta ton per tahun (7 November 2026). 2.352 million tonnes at 52 plants in 2025; 1.842 million tonnes by 26 October 2026; average heat value up from 2,520 to 3,440 kcal/kg; 8–10 million tonnes as potential demand. Company statements relayed by a news agency.
  2. Petromindo: PLN EPI targets 1 million tonnes of biomass exports in 2026 (11 March 2026). Exports planned from surplus feedstock domestic power plants cannot yet absorb; domestic cofiring biomass about Rp700,000 a tonne; exported palm shell Rp1.0–1.3 million a tonne. Average prices as reported.
  3. Metro TV: PLN EPI gandeng BNI kucuri pembiayaan bagi pemasok biomassa (21 September 2025). The price rule (1.2 × coal reference price, adjusted for heat, plus transport, under a 2023 energy ministry regulation) and bank finance for suppliers that requires an active PLN contract.
  4. Kontan: PLN EPI targetkan pasokan biomassa co-firing 4,07 juta ton pada 2027 (27 July 2026). Targets of 3.65 million tonnes for 2026 and 4.07 million for 2027; sawdust 54.6% and wood chip 22.4% of 2023–26 supply; processing facilities at Tasikmalaya and Ciamis of up to 10 tonnes an hour; 15 hubs planned by 2028; a biomass marketplace.
  5. Listrik Indonesia: Cofiring meningkat, PLN butuh banyak pemasok biomassa (6 September 2026). Nearly 200 supply partners, about 80% of them small businesses.
  6. Listrik Indonesia: PLN genjot ekosistem biomassa (2 November 2026). Sub-hubs, hubs and main hubs; digital records for transparency and traceability; distribution to power plants and industrial users. Stated plans, no outcome data.
  7. PT Prima Khatulistiwa Sinergi: best supplier award from PLN EPI (company release, 13 November 2025). Supplies nine PLN power stations in Sumatra and Kalimantan from its own plant; also sells to private plants and exports palm shell. A company’s own statement.
  8. Listrik Indonesia: PLN EPI gandeng Panah Perak Megasarana tambah pasokan biomassa. A pellet maker added as a PLN EPI supplier.
  9. Kontan: Maharaksa Biru Energi (OASA) kejar kapasitas produksi biomassa 500.000 ton per tahun (November 2023). A listed company’s plan to invest about Rp50 billion in a plant of 60,000 tonnes a year. A plan, not a measured cost; its outcome is unknown.
  10. Ombudsman RI: tata kelola biomassa belum optimal (press release, 15 January 2026). Sporadic, seasonal supply; high moisture and alkali risking boilers. Qualitative; no rejection statistics.
  11. Kontan: Zero ODOL naikkan biaya distribusi (research by ITL Trisakti). Rp1,084 per tonne-kilometre for overloaded trucks against Rp2,934 at legal load; all goods, not biomass.
  12. Kontan: Jelang Zero ODOL 2027, pengusaha truk minta tarif angkutan dan insentif (9 September 2026). The government’s 2027 target for ending overloaded trucks; the date has slipped before.
  13. Our earlier investigation on What Could Work: Could Java’s factories buy biomass boiler fuel by heat delivered, not by the tonne? (7 November 2026). The same stored research, from the side of factories that burn coal; it proposed a drying yard billing factories for the heat in each truckload.

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. Forecasting each supplier’s fuel, and planning deliveries. Two kinds of AI work, and one that is not AI. The first forecasts a quantity over time, here how wet and how plentiful each supplier’s fuel will be, from past records, the season and rainfall. Our AI Capability Atlas rates this established for patterns that recur within known bounds and conditional when data is sparse or conditions change, and it says a forecast must be checked against simple baselines over past periods. The second attaches an honest margin of error to each forecast; the Atlas rates this established for well-defined statistical targets and conditional when the population shifts. Turning forecasts into a delivery plan is planning mathematics, which the Atlas rates established for well-specified models and conditional in uncertain, changing operations. How this shaped the article: it is why the forecast must first beat last year’s averages on PLN EPI’s own past records, why planners approve every plan, why the plant tests stay, and why technical feasibility is rated Moderate: whether PLN EPI’s records are complete enough is unknown.
  2. Measuring each load on arrival. Three kinds of AI work together. The first cleans a sensor’s signal and estimates the quantity behind it, here moisture, with its uncertainty; the Atlas rates this established for specified physical measurement set-ups and conditional when the system is poorly observed. The second combines readings from several sources, here moisture sensors and a camera, into one estimate; the Atlas rates engineered sensor fusion established and warns that misaligned readings can mislead. The third recognises the kind of waste from an image; the Atlas rates this established for a fixed set of categories and conditional for open-ended scenes and rare defects. How this shaped the article: because suppliers would be paid on this estimate, it is why loads are still tested in a laboratory at random, why any uncertain load goes to the laboratory, why disputes are named as a risk, and why the trial checks the measurement against laboratory tests.
  3. Planning the hubs. The same forecasting as in note A, here of the waste each area will bring in and each plant’s need, and scheduling, which the Atlas rates established for explicit scheduling problems and conditional in highly changeable settings. How this shaped the article: it is why a hub planner approves every schedule and why forecasts are treated as the weak part until they beat simple baselines.
  4. The cost of hauling water. Rp2,934 per tonne-kilometre at legal load × 100 kilometres ≈ Rp293,000 per tonne moved.[11] The study covers all goods, not biomass, and the trip length is illustrative. How PLN’s transport allowance is calculated is not published, so how much of this PLN pays for water in practice is unknown. For scale on hub capital: about Rp50 billion for 60,000 tonnes a year is roughly Rp830,000 of investment per tonne of yearly capacity,[9] one company’s plan for a full processing plant, likely more than a drying hub needs.
  5. Other ideas we considered. For Improve, we also considered quick moisture readings on every load at the plant gate and lighter laboratory sampling for suppliers with good records. Both may help, but both depend on records whose quality is unknown, and neither changes what fuel reaches the plants as much as planning does. For Reinvent, we considered a shared pool reserving fuel across many suppliers and buyers; PLN EPI already plays that pooling role for PLN’s plants. Our first version of this article compared a per-load moisture check at suppliers with a heat service for factories; we replaced it because the two ideas served different buyers.

“One truckload” and both “Today / With the change” tables are our summaries of the process the research records. That a rejected load returns at the supplier’s cost, and that many small suppliers ship undried, are inferences in the research, not documented terms.

Evidence status of the main claims

Claim Status
PLN EPI delivered 2.352 million tonnes to 52 plants in 2025 and is behind its 2026 target Observed [1][4]
PLN EPI planned to export surplus its plants cannot yet absorb Observed [2]
PLN EPI pays 1.2 × a coal reference price, adjusted for heat, plus transport; about Rp700,000 a tonne Observed [2][3]
Supply is sporadic and seasonal, with moisture and alkali problems Observed [10]
Most small suppliers cannot afford to dry and test their fuel well Inferred
How PLN EPI matches fuel to plants today Unknown
A forecast of each supplier’s wetness would raise the fuel plants accept Hypothesis
Buying dry material at hubs would deliver more usable fuel within the price rule Hypothesis

Open questions

  • How often do PLN’s power stations reject loads, and why? PLN EPI’s acceptance records can answer.
  • How does PLN EPI plan deliveries today, and how complete are its records of past loads? PLN EPI can answer.
  • What do small suppliers pay for raw waste through the year? Suppliers and sawmills can answer.
  • What does drying cost a tonne at the Tasikmalaya and Ciamis sites? PLN EPI can answer.
  • How is PLN’s transport allowance calculated? PLN EPI’s contract terms can answer.
  • Can moisture be measured on a whole load at a weighbridge accurately enough to pay on? A trial against laboratory tests can answer.

Comments

Leave a Reply

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