Mill Quota Came Back at 14 Loads and the Crew Was Set for 32. Demand Models Now Call That Shift Weeks Out.

Fourteen loads. The crew was set up to cut thirty-two.

The quota email comes Wednesday afternoon, usually between two and four. It is three lines long. It says how many loads the mill will take from your vendor number next week, by product, and it does not explain itself. In a good week it says thirty-two and you go back to greasing the skidder. In February it can say fourteen, and by Thursday you are standing in a set with two hundred tons on the deck, four drivers who expect five turns a day, and a landowner promised the tract would be off by month's end.

Nobody is confused about what happens next. You call the wood buyer and beg. You call the second mill sixty-one miles the other direction and ask what they will take, knowing the haul eats four dollars a ton. You send a truck home. You let the loader operator go at noon. Every one of those calls is made off a guess, because the only thing you knew on Monday was last week's number.

Demand forecasting in a logging business means using your own delivered-load history to predict how many tons each mill will actually take, by product, two to six weeks before the quota email tells you. That is the whole idea. Not a market call on lumber futures — a call on your vendor number at four specific mills.

What the guess actually costs between Thursday and Friday

Price the idle day honestly and it stops being an annoyance. A mechanized southern crew — feller-buncher, grapple skidder, knuckleboom loader with a pull-through delimber — carries equipment notes, insurance, workers' compensation at logging rates, and six people who will go work for the contractor down the road if you send them home too often. Cut or not, that bill arrives.

Then the quieter loss that never shows up as a line item: merchandising into the wrong product. When the pulpwood quota is open and the chip-n-saw quota is not, the loader operator bucks for pulpwood, because a load you cannot deliver is worth nothing. Two weeks later the chip-n-saw quota opens back up and the stems that would have made it are already chipped. There is no invoice for that — just a delivered price per ton lower than the tract was capable of.

The third loss is tract selection. You bought the wet-weather tract in September for its rock road, and it turned out to be the driest winter in years while the sandy summer tract sat idle. Or the reverse, which is worse.

How the week actually gets planned today: a whiteboard, a rain gauge and forty years of feel

The current method is not stupid. It is a good pattern-matching machine that happens to live in one person's head. The owner knows the mill takes wood hard in January and February because everyone else is stuck in the mud, that the paper mill's July maintenance shutdown always lands the same two weeks, and that the sawmill goes down between Christmas and the second of January, so December's last three weeks are about building a deck you can sell in January.

The head holding all this is also running two crews, buying timber and doing payroll. And when the pattern breaks — a hurricane two states over dumps salvage pulpwood into the region and kills the price for five months — there is nothing to fall back on but the phone.

flowchart TD
  A["Three years of your own scale tickets, by mill and product"] --> B{"What will each mill take in weeks 2 to 6?"}
  B -->|Pulpwood quota tightening| C["Shift the set to the sawtimber-heavy tract"]
  B -->|Chip-n-saw quota opening| D["Buck long, hold CNS stems on the deck"]
  B -->|Both mills tight, ground wet| E["Move crew to the rock-road tract, cut hours"]
  C --> F["Weekly plan: tract, product mix, truck turns per mill"]
  D --> F
  E --> F
  F --> G["Owner overrides with what the wood buyer said Tuesday"]

What actually changed: the models stopped needing tidy history

Demand forecasting is now one of the most widely adopted uses of AI in manufacturing — roughly 48% of manufacturers are doing it — and pricing optimisation runs around 72% in retail and e-commerce. Those are not adjacent industries by accident. They are the two places where the history is messy in exactly the way a logging business's history is messy.

Here is what was different before. The older statistical methods wanted clean, regular history: the same measurement every week, no gaps, a stable seasonal shape. Yours is not that. You have weeks with zero tons because it rained eleven inches, a six-month hole where you ran one crew instead of two, a mill change in 2024, and product codes a clerk typed two different ways. Feed that into a classical forecast and it either refuses or smooths it into nothing.

The 2026 models handle that. They read gappy, lumpy, seasonal history the way a person does — they know a rain week is a rain week and not a demand collapse, they carry the December shutdown across a mill change, and they say plainly when there is not enough history on a new mill to be useful. That last part matters more than the forecast. And with the cost of running these models down roughly tenfold from 2025, recalculating every mill and product nightly costs less than a case of bar oil.

A Tuesday in February, run the new way

Seven in the morning. The forecast ran overnight on three years of your scale tickets — net tons by date, mill, product, tract and truck — plus the quota emails, rainfall by county, and the published delivered-price series you already subscribe to.

It says three things. First: the pine pulpwood mill has taken above its stated quota in nine of the last eleven wet Februarys and will likely take twenty-eight to thirty-four loads next week, not the twenty-two it published. Second: chip-n-saw quota at the second mill has climbed three weeks running and the spread over pulpwood has widened, so borderline stems are worth bucking long. Third — the one that earns its keep — your tonnage to the third mill has drifted down four weeks running while the others held, which usually means something changed there and you should call the wood buyer today, not Wednesday.

The foreman gets one page: which tract, what to buck for, how many turns per truck per mill. The owner still makes the call — just not from memory at ten o'clock at night.

The arithmetic on one crew, with the assumptions on the table

Illustrative numbers for one southern crew. Put your own in; the shape is what matters.

Against that, a setup that reads your tickets and runs nightly is a couple of hundred dollars a month plus the work of getting three years of tickets into one place. If you recovered only the idle days and never moved a ton between products, it still pays. And you can prove it: write down the forecast every Friday, write down what the quota email said the following Wednesday, and after ten weeks you will know whether to keep it.

What the forecast will not tell you, and where the wood buyer still wins

It cannot see inside the mill. If the chipper goes down Sunday night, no amount of history predicts Monday's quota. The relationship with the procurement forester still gets you that call, and it is still the most valuable asset in this business. The forecast makes you a better caller, not a caller who needs nobody.

It is weak on anything with no history. A brand-new mill, a new product class, a first-time supply agreement — the model has nothing and should say so. If it produces a confident number for a mill you have delivered to four times, that is a reason to distrust it, not a reason to act.

It does not know your ground. It can see that it rained; it cannot see that the branch crossing on the north end will not hold a loaded truck until Thursday. That stays the foreman's call. And it has no view on a competitor bidding tracts to keep his crews busy rather than to make money. Nothing forecasts a man who has decided to lose money.

Frequently asked questions

How much of my own history do I actually need?

Two years is workable, three is comfortable. Length matters less than detail: it has to be your tickets, by mill and product, with dates. A weekly company-wide tonnage figure is close to useless. Loads with mill, product, net tons and tract number are what make it work.

My tickets are PDFs from four different mill portals. Is that a problem?

That is the normal starting point, and it is the real work. Tickets have to get out of the portals and into one place before anything can read them. Most contractors find that is the project, and the forecast is the easy part afterward.

Will this tell me what stumpage to bid?

Indirectly, and carefully. It sharpens your view of delivered price by product and of how many tons you can realistically move per week — two of the inputs to a bid. It does not know the cruise, the terrain, the haul or who else is bidding. Treat it as a check on your number, not a replacement.

What if my history includes a hurricane salvage year?

Say so, in plain words, when you set it up. These models handle a labelled disruption far better than an unlabelled one. Leave a salvage year in as if it were normal and the price collapse gets read as seasonality — bad answers every autumn after.

One practical note. Landowners call about when you will be off their place, drivers call about which mill to run to, and wood buyers call the office while everyone is in the woods. CallSphere builds AI voice and chat agents that answer the business line and web chat around the clock, take the caller's details and book the callback, so the tract-status calls that pile up during a quota shift get captured instead of going to voicemail. It does not forecast quota — that is your tickets and your wood buyer.