The record // Minsawi Smart Mill
Reported by The Star, 19 November 2024 — Perak facility is world’s first palm oil mill to adopt AI
A 41-year-old mill
became the world’s first.
Minsawi Industries, Kuala Kangsar, Perak. Retrofitted from August 2024 and launched on 19 November 2024. The Star reported it that day as the first palm oil mill in the world to adopt AI, quoting the Plantation and Commodities Minister at the launch. It was instrumented while it was still milling.
The delta
What changed, and who
reported it.
These figures are ours. AIREI presented this performance data at the inaugural International Palm Oil Millers Conference 2026. Joseph Tek Choon Yee — former president of the Malaysian Estate Owners’ Association and past chief executive of the Malaysian Palm Oil Association — was in the room, and wrote them up in The Edge Malaysia on 25 May 2026 with his own qualifiers attached. The column says so plainly: “the achievement section showing the post-implementation performance trend”, “the last few performance slides shared”. What that establishes is that an industry authority saw our data and put it into print with caveats, two paragraphs of which push back hard and are quoted in full further down. It does not establish that anybody independent measured anything.
It is an opinion column, not a news report, and it hedges — The Edge Malaysia has OER “reportedly” improved from “about” 17.85%, and breakdown hours from “about” 12.21. We keep its qualifiers. The column also does not name the mill. That identification — Minsawi, the same retrofit, the same technology provider — comes from a separate report in The Star, 19 Nov 2024. Joining the two is our inference.
Ours, relayed in The Edge Malaysia, 25 May 2026
OER — delta
+1.17 pp§OER
Oil extraction rate, against that mill’s own pre-retrofit baseline. A delta on one site — not a target, and not a general performance claim.
Ours, relayed in The Edge Malaysia, 25 May 2026
Mill headcount
66→31§HDC
A 53% fall in headcount at this mill — a count, not a wage cost and not a rate. One lower figure for the same mill is in print elsewhere and named further down this page.
Ours, relayed in The Edge Malaysia, 25 May 2026
Breakdown hours
12.21→3.08§BRK
Monthly average. Ask us how we define a breakdown hour and when the clock starts.
Ours, relayed in The Edge Malaysia, 25 May 2026
Payback
< 2 yrs§PAY
The column calls it “the technology provider’s claim”, not its own finding. The only one of the four that is a projection rather than a measurement — none of the four is independent of us, so that is the difference, and it is the reason this one should carry least weight.
The same figure, charted
Drawn from zero, so the gain is the size it actually is.
Both endpoints sit on one axis that begins at zero and is not truncated, and the axis is drawn rather than asserted — measure the bars against its ticks. The gold segment is the reported movement at the same scale as the baseline it sits on. It is small, and small is the honest picture. A chart that cropped the axis to start near 17 would show the same two numbers as a transformation.
Ours, relayed in The Edge Malaysia, 25 May 2026 — an opinion column, and it hedges both endpoints
Before · the mill’s own baseline17.85%
After · reported19.02%
OER · per centGold — the +1.17 pp movementAxis 0–20, linear · zero baseline, no truncation
points of OER, at Minsawi only. A delta on that site’s own baseline — never an absolute target, never a general performance claim. The arithmetic is yours to check: 19.02 minus 17.85. The Edge Malaysia calls both endpoints “about” and the movement “reportedly”, so read the two decimals as arithmetic on hedged inputs, not as precision.§OER
- Unit basis
- Percentage points of oil extraction rate. The two figures above are endpoints; the delta is the difference between them, never a ratio — a movement in points, not a per-cent improvement.
- Comparator
- The same mill, against its own pre-retrofit baseline. No control, no parallel line, no held-out period — so whatever else changed at that plant in the same window is inside this number.
- Source and date
- The Edge Malaysia, 25 May 2026 — “When palm oil learns to think”, Joseph Tek Choon Yee. An opinion column, not reportage, written from our own conference slides, and not independently verified. Its words are “reportedly improved from about 17.85% to 19.02%”. The column names no mill; the Minsawi identification is ours, joining it to a separate report in The Star, 19 November 2024.
- Individual results will vary
- One mill, one retrofit, one period, and no range or error bar behind either endpoint. Extraction is bounded by ripeness at harvest and by the hours between cut and press — a mill buying most of its crop outside holds neither constant, so this chart is a record of what happened at Kuala Kangsar and not a forecast for your plant.
Measurement notes
Each figure above carries a note, keyed by the mark beside it. The four sit side by side below, answering the same six questions in the same order — provenance, basis, comparator, scope, system version, variance — so one row can be read straight across. Every note prints not established rather than dropping the row it cannot answer. None of the four carries a range or an error bar. For three of them one site and one period leave no spread to quote. The fourth does not have that excuse: breakdown hours are a monthly average, so a distribution exists and we have not published it.
The column’s own words
Its words: “reportedly improved from about 17.85% to 19.02%”, and we keep both hedges.
Why those two decimals are not precision
But the column — The Edge Malaysia, 25 May 2026 — calls both endpoints “about”, so read those two decimals as arithmetic on hedged inputs, not as precision. It is not “1.17% better” and it is not “6.6% better”.
How the mill was identified, and what is undated
We infer it by joining this column to a separate report in The Star. Neither period is dated in any published source, and our own record of them is the unsigned document below. Throughput and fruit quality across the two periods are undocumented.
What is not defined
What counts as a breakdown, and when the clock starts, is not defined — which is why it sits in the pending basis below.
What that projection implies
Note what it implies: recovering that capital inside two years is an annual saving of the same order as the ringgit figure we decline to print further down this page. Neither has a documented basis. This one is labelled a projection and attributed to us; that is the whole of the difference.
What these four are missing
They fail our own grading test.
Our grading page withholds an accuracy percentage until five things sit beside it: the reference it was measured against, the sample, who ran the comparison, which build of the model produced it, and an approver’s signature. Hold these four to the same five. The reference method is none — a before-and-after on one mill’s own records. The sample is one mill and one retrofit, across two periods nobody has dated. The tester is us. The model build is unrecorded. The approver line is the unsigned one further down this page.
So what separates these four from the grading figure we withhold is exposure, not rigour — a third party put these into print with his own hedges attached, and the grading figure has never left our brochure. They are here because they are attributed and you can read the column yourself. They are not here because they are proven. And withholding them would not unpublish them — the column is in print and a search finds it in a minute. It would only strip our qualifiers off numbers you were going to read anyway.
What we leave out
Two numbers in The Star we do not use.
The Star’s launch report — the same article this page cites for the mill’s age and its investment, and links in its own footer — carries two figures that appear nowhere on this site. Here they are, and here is why not.
A workforce reduction of “between 30% and 35%”. The Star is explicit about where it came from: “Johari said he was informed” — the Minister, at the launch ceremony, relaying a briefing, about a mill that had by then been running the system for roughly a month. The 66 to 31 above is what our own records showed eighteen months later. The two have never been reconciled against each other, by us or by anybody else.
A ringgit annual-savings figure, and it flatters us. We do not print it, because nobody has established what it counts. The published one is in the article, one click away, and you are welcome to weigh it yourself.
The records themselves would be better than a column about them. They are Minsawi’s, not ours, and we have no standing to publish another company’s operating data. What we can put in front of you is the measurement basis behind these figures, and an assessment at your own plant — where the records would be yours, and the numbers would not need a source line because you would have watched them being taken.
The Edge Malaysia, 25 May 2026 · the same column, quoted in full where it pushes back
“In some cases, certain responses can be executed automatically through the control system. But ultimate accountability must remain with the management team, who must be able to validate, adjust or override recommendations based on operational judgement and real-world conditions.”
“A mill that sources nearly 90% of its crop from outside suppliers lives with a different kind of uncertainty… smart mill technology should hopefully also be tested and implemented in an integrated mill linked to its own estates, where crop supply, harvesting intervals, bunch quality and traceability are more constant and better assured.”
“The Kuala Kangsar mill should be seen as an important working platform, not the final textbook answer.”
The override requirement is a real gap — ask us about it directly. The estate-linked caveat is the honest reason to book an assessment rather than read a number: this result is not yet proven in your configuration.
Read the full analysis: When palm oil learns to think — the AI estate-mill frontier, Part 1, The Edge Malaysia, 25 May 2026.
How to read it
Read the delta. The absolute
will mislead you.
Minsawi’s post-retrofit OER sits below the Malaysian national average published by MPOB. If you read that number on its own, the honest conclusion is “this takes a poor mill to slightly below average” — and you would be drawing a fair conclusion from the wrong number.
Your absolute OER depends on things a mill does not control: the ripeness standard the estates harvest to, the quality of what arrives at the ramp, and the time between cut and press. A mill can be run impeccably and still sit under the national average because of what came off the lorry.
What a mill can act on is what happens between the ramp and the tank. That is where the delta was measured, and the delta is the only figure on this page we will argue for.
Ask and we will send the MPOB citation for the national average, with its scope and its edition.
Capability, not attribution
What we instrument
at that mill.
Three of the stations AIREI instruments at that mill. This is a list of what we instrument — not a claim about what produced the delta above.
01
Sterilizer recipe selection
Which cycle to run against the load in front of you, rather than the cycle the shift always runs.
02
Press mesocarp oil-loss prediction
Predicted loss per press against that press’s own baseline, in the cycle rather than in the following week’s laboratory report.
03
Clarification VCT skimming
Skimming decisions in the clarification station instrumented instead of judged by eye.
We do not publish a causal split. No portion of the reported OER delta is attributed here to any one of those three, and you should be suspicious of any vendor who does. No causal decomposition exists in the evidence; producing one would be marketing attribution wearing an engineer’s coat.
These four figures come from one mill, over one period, on the system as it stood then. They are a record of what happened at Minsawi — not a specification for the system you would be quoted. Ask us how any of them was measured and we will send you the basis.
For your process engineer
Measurement basis.
Every figure on this page came from one mill over one period. Ask us how any of them was measured — the dates, the throughput, the fruit quality, what else changed on the plant, and how we count a breakdown hour — and we will send you what we hold, as it stands.
Ask for the measurement basis
It comes to you with the periods, the definitions and the sign-off attached. If a figure on this page matters to your decision, this is the document to read before you act on it.
Where this chain ends
Two of our documents
are on this site.
Every figure on this site names a document. For the journalism that is enough — The Star and The Edge are hyperlinked and you can read them without asking us. For our own documents it was not: a citation reading page 3 is theatre if you cannot turn to page 3. So the two pages that carry the specification are published here, ungated, and every citation of them on this site is a link to the page itself.
Published, no form, no email
Documents you can open
- Smart Mill — the eight stations and what each one does
- Smart Palm Oil Mill Brochure 2026, page 3 (PDF) — the module list every capability line on the Smart Mill and Traceability pages is quoted from.
- Unmanned AI FFB Grading — the specification
- AIREI AI Grading Brochure, page 2 (PDF) — the key-feature list and the four grading tiers, as printed.
- Company profile
- Not published. It is a 17-page internal deck carrying staff and client detail that is not ours to put on a public page. Where this site cites it, the citation is a pointer to a document you must ask for — not one you can open.
- The corporate films
- Not published as files. The quotes and timestamps we take from them are exact, and the clips are sent on request so you can check them.
If a claim on this site matters to your decision and its document is in the second half of that list, ask for it — you will get the document, not a summary of it.
The install
You could not turn
off the mill.
Minsawi had a crop to process. There was no commissioning shutdown to hide the work in, so the retrofit went in around a mill that kept running — sensors, actuators, control and network laid onto live plant, station by station, from August 2024.
The Edge Malaysia, quoting AIREI CEO Surendran Kuranadan — undated p16 clipping, no online copy recovered
Per The Edge Malaysia, the only window where the plant could be stopped was seven hours on a Sunday. Everything that genuinely required a dead plant had to be sequenced into it. Everything else had to be done hot.
If you are weighing this against your own crop calendar, that is the constraint worth asking us about first — before the modules, before the dashboards. It is the part of the job that decides whether a retrofit is possible at your site at all.
AI cannot rescue poor engineering.
The author who published the Minsawi figures also wrote that the technology is not the final answer to every milling challenge — that sound engineering, disciplined maintenance, reliable instruments, clean steam lines, healthy boilers and trained operators come first.
We agree with him, and we would rather you read it here than find it later. It is also why the first thing we sell is a site assessment and not a system: if the plant underneath is not right, instrumenting it will tell you that in high resolution and change nothing.
Coverage
Who else wrote it down.
Named outlets, real headlines, and a link to each one so you can read the original. The launch was covered in three languages.
- The Star Perak facility is world’s first palm oil mill to adopt AI 19 Nov 2024 · English
- New Straits Times AI can boost Malaysia’s palm oil productivity, cut labour dependency Nov 2024 · English
- Bernama Malaysian palm oil mills should adopt AI technology — Johari 19 Nov 2024 · English
- The Edge Malaysia When palm oil learns to think: the AI estate-mill frontier (Part 1) 25 May 2026 · English
- Buletin TV3 Kilang sawit AI pertama di dunia dilancar di Kuala Kangsar 19 Nov 2024 · Bahasa Malaysia
- Berita Harian Malaysia negara pertama dunia integrasi AI dalam kilang sawit Nov 2024 · Bahasa Malaysia
- Harian Metro Kilang sawit pertama integrasi AI Nov 2024 · Bahasa Malaysia
- Berita RTM Kilang proses sawit guna AI pertama di dunia Nov 2024 · Bahasa Malaysia
- Business Today Palm oil mills should adopt AI to reduce cost and increase revenue 19 Nov 2024 · English
- Asia News Network AI boost for palm oil production in world’s first smart mill in Malaysia Nov 2024 · English
- OFI Magazine Malaysia’s first palm oil mill using artificial intelligence launched Nov 2024 · English
- Asia Palm Oil Magazine Malaysia unveils the world’s first AI-driven smart palm oil mill Nov 2024 · English
Next step
Your mill is not this mill.
Nothing on this page transfers to another site on its own. What a site assessment establishes is the only thing that matters to you: what your plant is losing today, where it is losing it, and whether instrumenting it would change that.